Economics Nobel Prize 2000 To 2013

The Economics Nobel Prize 2000 to 2013
Compiled by Rajnish Kumar

The Nobel Prize in Economics is officially called The Sveriges Riksbank Prize in
Economic Sciences in Memory of Alfred Nobel

Interesting theories have been professed by academicians across the world; the website hosts
the advanced as well as popular information on the prize. The most recent awards from 2000
to 2013 have been compiled. The purpose is to understand whether these theories have
actually been implemented. Do they have practical value? Are they understood by politicians
of the world? Economics they say is a practical social science subject.

CONTENTS

2013
The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2013
Eugene F. Fama, Lars Peter Hansen and Robert J. Shiller
"for their empirical analysis of asset prices"

2012
Alvin E. Roth and Lloyd S. Shapley
"for the theory of stable allocations and the practice of market design"

2011
Thomas J. Sargent and Christopher A. Sims
"for their empirical research on cause and effect in the macroeconomy"

2010
Peter A. Diamond, Dale T. Mortensen and Christopher A. Pissarides
"for their analysis of markets with search frictions"

2009
Elinor Ostrom
"for her analysis of economic governance, especially the commons"

Oliver E. Williamson
"for his analysis of economic governance, especially the boundaries of the firm"

2008
Paul Krugman
"for his analysis of trade patterns and location of economic activity"

2007
Leonid Hurwicz, Eric S. Maskin and Roger B. Myerson
"for having laid the foundations of mechanism design theory"

2006
Edmund S. Phelps
"for his analysis of intertemporal tradeoffs in macroeconomic policy"

2005
Robert J. Aumann and Thomas C. Schelling
"for having enhanced our understanding of conflict and cooperation through game-theory analysis"

2004
Finn E. Kydland and Edward C. Prescott
"for their contributions to dynamic macroeconomics: the time consistency of economic policy and the
driving forces behind business cycles"

2003
Robert F. Engle III
"for methods of analyzing economic time series with time-varying volatility (ARCH)"

Clive W.J. Granger
"for methods of analyzing economic time series with common trends (cointegration)"

2002
Daniel Kahneman
"for having integrated insights from psychological research into economic science, especially
concerning human judgment and decision-making under uncertainty"

Vernon L. Smith
"for having established laboratory experiments as a tool in empirical economic analysis, especially in
the study of alternative market mechanisms"

2001
George A. Akerlof, A. Michael Spence and Joseph E. Stiglitz
"for their analyses of markets with asymmetric information"

2000
James J. Heckman
"for his development of theory and methods for analyzing selective samples"

Daniel L. McFadden
"for his development of theory and methods for analyzing discrete choice"

Web links to the Nobel Prize webpage with more information

Eugene F. Fama, Lars Peter Hansen and Robert J. Shiller
"for their empirical analysis of asset prices"

Alvin E. Roth and Lloyd S. Shapley
"for the theory of stable allocations and the practice of market design"

"for their empirical research on cause and effect in the macroeconomy"

Peter A. Diamond, Dale T. Mortensen and Christopher A. Pissarides
"for their analysis of markets with search frictions"

Elinor Ostrom
"for her analysis of economic governance, especially the commons"
Oliver E. Williamson
"for his analysis of economic governance, especially the boundaries of the firm"

Paul Krugman
"for his analysis of trade patterns and location of economic activity"

Leonid Hurwicz, Eric S. Maskin and Roger B. Myerson
"for having laid the foundations of mechanism design theory"

Edmund S. Phelps
"for his analysis of intertemporal tradeoffs in macroeconomic policy"

Robert J. Aumann and Thomas C. Schelling
"for having enhanced our understanding of conflict and cooperation through game-theory analysis"

Finn E. Kydland and Edward C. Prescott
"for their contributions to dynamic macroeconomics: the time consistency of economic policy and the driving forces behind business cycles"

Robert F. Engle III
"for methods of analyzing economic time series with time-varying volatility (ARCH)"
Clive W.J. Granger
"for methods of analyzing economic time series with common trends (cointegration)"

Daniel Kahneman
"for having integrated insights from psychological research into economic science, especially concerning human judgment and decision-making
under uncertainty"
Vernon L. Smith
"for having established laboratory experiments as a tool in empirical economic analysis, especially in the study of alternative market mechanisms"

George A. Akerlof, A. Michael Spence and Joseph E. Stiglitz
"for their analyses of markets with asymmetric information"

James J. Heckman
"for his development of theory and methods for analyzing selective samples"
Daniel L. McFadden
"for his development of theory and methods for analyzing discrete choice"

14 OCTOBER 2013
Scientifc Background on the Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2013
UNDERSTANDI NG ASSET PRI CES
compiled by the Economic Sciences Prize Committee of the Royal Swedish Academy of Sciences
THE ROYAL SWEDISH ACADEMY OF SCIENCES has as its aim to promote the sciences and strengthen their inuence in society.
BOX 50005 (LILLA FRESCATIVGEN 4 A), SE-104 05 STOCKHOLM, SWEDEN
TEL +46 8 673 95 00, FAX +46 8 15 56 70, INFO@KVA.SE HTTP://KVA.SE
1

UNDERSTANDING ASSET PRICES

1. Introduction
The behavior of asset prices is essential for many important decisions, not only for
professional investors but also for most people in their daily life. The choice between saving
in the form of cash, bank deposits or stocks, or perhaps a single-family house, depends on
what one thinks of the risks and returns associated with these different forms of saving. Asset
prices are also of fundamental importance for the macroeconomy because they provide crucial
information for key economic decisions regarding physical investments and consumption.
While prices of financial assets often seem to reflect fundamental values, history provides
striking examples to the contrary, in events commonly labeled bubbles and crashes.
Mispricing of assets may contribute to financial crises and, as the recent recession illustrates,
such crises can damage the overall economy. Given the fundamental role of asset prices in
many decisions, what can be said about their determinants?
This years prize awards empirical work aimed at understanding how asset prices are
determined. Eugene Fama, Lars Peter Hansen and Robert Shiller have developed methods
toward this end and used these methods in their applied work. Although we do not yet have
complete and generally accepted explanations for how financial markets function, the
research of the Laureates has greatly improved our understanding of asset prices and revealed
a number of important empirical regularities as well as plausible factors behind these
regularities.
The question of whether asset prices are predictable is as central as it is old. If it is possible to
predict with a high degree of certainty that one asset will increase more in value than another
one, there is money to be made. More important, such a situation would reflect a rather basic
malfunctioning of the market mechanism. In practice, however, investments in assets involve
risk, and predictability becomes a statistical concept. A particular asset-trading strategy may
give a high return on average, but is it possible to infer excess returns from a limited set of
historical data? Furthermore, a high average return might come at the cost of high risk, so
predictability need not be a sign of market malfunction at all, but instead just a fair
compensation for risk-taking. Hence, studies of asset prices necessarily involve studying risk
and its determinants.
2

Predictability can be approached in several ways. It may be investigated over different time
horizons; arguably, compensation for risk may play less of a role over a short horizon, and
thus looking at predictions days or weeks ahead simplifies the task. Another way to assess
predictability is to examine whether prices have incorporated all publicly available
information. In particular, researchers have studied instances when new information about
assets becomes became known in the marketplace, i.e., so-called event studies. If new
information is made public but asset prices react only slowly and sluggishly to the news, there
is clearly predictability: even if the news itself was impossible to predict, any subsequent
movements would be. In a seminal event study from 1969, and in many other studies, Fama
and his colleagues studied short-term predictability from different angles. They found that the
amount of short-run predictability in stock markets is very limited. This empirical result has
had a profound impact on the academic literature as well as on market practices.
If prices are next to impossible to predict in the short run, would they not be even harder to
predict over longer time horizons? Many believed so, but the empirical research would prove
this conjecture incorrect. Shillers 1981 paper on stock-price volatility and his later studies on
longer-term predictability provided the key insights: stock prices are excessively volatile in
the short run, and at a horizon of a few years the overall market is quite predictable. On
average, the market tends to move downward following periods when prices (normalized, say,
by firm earnings) are high and upward when prices are low.
In the longer run, compensation for risk should play a more important role for returns, and
predictability might reflect attitudes toward risk and variation in market risk over time.
Consequently, interpretations of findings of predictability need to be based on theories of the
relationship between risk and asset prices. Here, Hansen made fundamental contributions first
by developing an econometric method the Generalized Method of Moments (GMM),
presented in a paper in 1982 designed to make it possible to deal with the particular features
of asset-price data, and then by applying it in a sequence of studies. His findings broadly
supported Shillers preliminary conclusions: asset prices fluctuate too much to be reconciled
with standard theory, as represented by the so-called Consumption Capital Asset Pricing
Model (CCAPM). This result has generated a large wave of new theory in asset pricing. One
strand extends the CCAPM in richer models that maintain the rational-investor assumption.
Another strand, commonly referred to as behavioral finance a new field inspired by Shillers
early writings puts behavioral biases, market frictions, and mispricing at center stage.
3

A related issue is how to understand differences in returns across assets. Here, the classical
Capital Asset Pricing Model (CAPM) for which the 1990 prize was given to William Sharpe
for a long time provided a basic framework. It asserts that assets that correlate more strongly
with the market as a whole carry more risk and thus require a higher return in compensation.
In a large number of studies, researchers have attempted to test this proposition. Here, Fama
provided seminal methodological insights and carried out a number of tests. It has been found
that an extended model with three factors adding a stocks market value and its ratio of book
value to market value greatly improves the explanatory power relative to the single-factor
CAPM model. Other factors have been found to play a role as well in explaining return
differences across assets. As in the case of studying the market as a whole, the cross-sectional
literature has examined both rational-investorbased theory extensions and behavioral ones to
interpret the new findings.
This document is organized in nine sections. Section 2 lays out some basic asset-pricing
theory as a background and a roadmap for the remainder of the text. Sections 3 and 4 discuss
short- and longer-term predictability of asset prices, respectively. The following two sections
discuss theories for interpreting the findings about predictability and tests of these theories,
covering rational-investorbased theory in Section 5 and behavioral finance in Section 6.
Section 7 treats empirical work on cross-sectional asset returns. Section 8 briefly summarizes
the key empirical findings and discusses their impact on market practices. Section 9 concludes
this scientific background.

2. Theoretical background
In order to provide some background to the presentation of the Laureates contributions, this
section will review some basic asset-pricing theory.

2.1 Implications of competitive trading
A set of fundamental insights, which go back to the 19
th
century, derive from a basic
implication of competitive trading: the absence of arbitrage opportunities. An arbitrage
opportunity is a money pump, which makes it possible to make arbitrary amounts of money
without taking on any risk. To take a trivial example, suppose two assets pay safe rates of
return
and
, where
>
. If each asset can be sold short, i.e., held in negative

4

amounts, an arbitrage gain could be made by selling asset b short and investing the proceeds
in asset a: the result would be a safe rate profit of
. Because this money pump could

be operated at any scale, it would clearly not be consistent with equilibrium; in a competitive
market,
and
must be equal. Any safe asset must bear the same return
( for safe);
the rate at which future payoffs of any safe asset are discounted.
This simple reasoning can be generalized quite substantially and, in particular, can deal with
uncertain asset payoffs. The absence of arbitrage opportunities can be shown to imply that the
price of any traded asset can be written as a weighted, or discounted, sum of the payoffs of the
asset in the different states of nature next period, with weights independent of the asset in
question (see, e.g., Ross, 1978 and Harrison and Kreps, 1979). Thus, at any time t, the price
of any given asset i is given by
,
=
+1
()
+1
()
,+1
().
Here, s denotes a state of nature, the s the probabilities with which these states occur, and
the s non-negative discounting weights. The s are the payoffs, which in the case of stocks
are defined as next-period price plus dividends:
,+1
=
,+1
+
,+1
. In general, all these
items depend on the state of nature. Note that the discounting weights m are the same for all
assets.
1
They matter for the price of an individual asset i only because both m and
depend
on s.
For a safe asset f, does not depend on , and the formula becomes
,
=
,+1

+1
()
+1
().
Thus, we can now interpret
+1
()
+1
() as defining the time t risk-free discount rate
,
for safe assets:

+1
()
+1
() 1 1 +
,
.
More generally, though, the dependence of
+1
() on the state of nature s captures how the
discounting may be stronger in some states of nature than in others: money is valued
differently in different states. This allows us to capture how an assets risk profile is valued by
the market. If it pays off particularly well in states with low weights, it will command a lower
price.
1
In addition, if markets are complete (i.e., if there are as many independent assets as there are states of nature),
the that determines the prices for all assets is also unique.

5

The no-arbitrage pricing formula is often written more abstractly as

,
=
(
+1
,+1
), (1)
where E now subsumes the summation and probabilities: it is the expected (probability-
weighted) value. This formula can be viewed as an organizational tool for much of the
empirical research on asset prices. With
,+1
=
,+1
+
,+1
, equation (1) can be iterated
forward to yield the price of a stock as the expected discounted value of future dividends.
2

Are asset prices predictable?
Suppose, first, that we consider two points in time very close to each other. In this case, the
safe interest rate is approximately zero. Moreover, over a short horizon, m might be assumed
not to vary much across states: risk is not an issue. These assumptions are tantamount to
assuming that m equals 1. If the payoff is simply the assets resale value
+1
, then the
absence of arbitrage implies that
+1
. In other words, the asset price may go up or
down tomorrow, but any such movement is unpredictable: the price follows a martingale,
which is a generalized form of a random walk. The unpredictability hypothesis has been the
subject of an enormous empirical literature, to which Fama has been a key contributor. This
research will be discussed in Section 3.

Risk and the longer run
In general, discounting and risk cannot be disregarded, so tests of the basic implications of
competitive trading need to account for the properties of the discount factor m: how large it is
on average, how much it fluctuates, and more generally what its time series properties are.
Thus, a test of no-arbitrage theory also involves a test of a specific theory of how m evolves, a
point first emphasized by Fama (1970).
Suppose we look at a riskless asset f and a risky asset i. Then equation (1) allows us to write
the assets price as
,
=

(
,+1
)
1+
,
+
(
+1
)

(
+1
,
,+1
)
(
+1
)
.
2
This presumes the absence of a bubble, i.e., that the present value of dividends goes to zero as time goes to infinity.
See Tirole (1985).

6

The discount factor
+1
() can be regarded as the value of money in state s. The above
pricing equation thus says that the assets value depends on the covariance with the value of
money. If the covariance is negative, i.e., if the assets payoff is high when the value of
money is low, and vice versa, then the asset is less valuable than the expected discounted
value of the payoff. Moreover, the discrepancy term can be factorized into
(
+1
), the
risk loading (amount of risk), and
(
+1
,+1
)
(
+1
)
, the risk exposure, of the asset.
The pricing formula can alternatively be expressed in terms of expected excess returns over
the risk-free asset:
[(
,+1

,
)
+1
] = 0, where 1 +
,+1
=
,+1
/
. This allows us
to write
,+1

,
= (1 +
,
)
(
+1
,+1
).
An asset whose return is low in periods when the stochastic discount factor is high (i.e., in
periods where investors value payoffs more) must command a higher risk premium or
excess return over the risk-free rate. How large are excess returns on average? How do they
vary over time? How do they vary across different kinds of assets? These fundamental
questions have been explored from various angles by Fama, Hansen and Shiller. Their
findings on price predictability and the determinants and properties of risk premia have
deepened our understanding of how asset prices are formed for the stock market as a whole,
for other specific markets such as the bond market and the foreign exchange market, and for
the cross-section of individual stocks. In Section 4, we will discuss the predictability of asset
prices over time, whereas cross-sectional differences across individual assets will be treated in
Section 7.

2.2 Theories of the stochastic discount factor
The basic theory, described above, is based on the absence of arbitrage. The obvious next step
is to discuss the determinants of the stochastic discount factor m. Broadly speaking, there are
two approaches: one based on rational investor behavior, but possibly involving institutional
complications, investor heterogeneity, etc., and an alternative approach based on
psychological models of investor behavior, often called behavioral finance.

7

Rational-investor theory
Theory based on the assumption of rational investor behavior has a long tradition in asset
pricing, as in other fields of economics. In essence, it links the stochastic discount factor to
investor behavior through assumptions about preferences. By assuming that investors make
portfolio decisions to obtain a desired time and risk profile of consumption, the theory
provides a link between the asset prices investors face in market equilibrium and investor
well-being. This link is expressed through , which captures the aspects of utility that turn
out to matter for valuing the asset. Typically, the key link comes from the time profile of
consumption. A basic model that derives this link is the CCAPM.
3
It extends the static CAPM
theory of individual stock prices by providing a dynamic consumption-based theory of the
determinants of the valuation of the market portfolio. CCAPM is based on crucial
assumptions about investors utility function and attitude toward risk, and much of the
empirical work has aimed to make inferences about the properties of this utility function from
asset prices.
The most basic version of CCAPM involves a representative investor with time-additive
preferences acting in market settings that are complete, i.e., where there is at least one
independent asset per state of nature. This theory thus derives as a function of the
consumption levels of the representative investor in periods t+1 and t. Crucially, this function
is nonlinear, which has necessitated innovative steps forward in econometric theory in order
to test CCAPM and related models. These steps were taken and first applied by Hansen.
In order to better conform with empirical findings, CCAPM has been extended to deal with
more complex investor preferences (such as time non-separability, habit formation, ambiguity
aversion and robustness), investor heterogeneity, incomplete markets and various forms of
market constraints, such as borrowing restrictions and margin constraints. These extensions
allow a more general view of how depends on consumption and other variables. The
progress in this line of research will be discussed in Section 5.

Behavioral finance
Another interpretation of the implied fluctuations of observed in the data is based on the
view that investors are not fully rational. Research along these lines has developed very
rapidly over the last decades, following Shillers original contributions beginning in the late
3
The CCAPM has its origins inwork by Merton (1973), Lucas (1978) and Breeden (1979).

8

1970s. A number of specific departures from rationality have been explored. One type of
departure involves replacing the traditional expected-utility representation with functions
suggested in the literature on economic psychology. A prominent example is prospect theory,
developed by the 2002 Laureate Daniel Kahneman and Amos Tversky. Another approach is
based on market sentiment, i.e., consideration of the circumstances under which market
expectations are irrationally optimistic or pessimistic. This opens up the possibility, however,
for rational investors to take advantage of arbitrage opportunities created by the
misperceptions of irrational investors. Rational arbitrage trading would push prices back
toward the levels predicted by non-behavioral theories. Often, therefore, behavioral finance
models also involve institutionally determined limits to arbitrage.
Combining behavioral elements with limits to arbitrage may lead to behaviorally based
stochastic discount factors, with different determinants than those derived from traditional
theory. For example, if the is estimated from data using equation (1) and assuming rational
expectations (incorrectly), a high value may be due to optimism and may not reflect
movements in consumption. In other words, an equation like (1) is satisfied in the data, but
since the expectations operator assigns unduly high weights to good outcomes it makes the
econometrician overestimate . Behavioral-finance explanations will be further discussed in
Section 6.

CAPM and the cross-section of asset returns
Turning to the cross-section of assets, recall from above that an individual stock price can be
written as the present value of its payoff in the next period discounted by the riskless interest
rate, plus a risk-premium term consisting of the amount of risk,
(
+1
), of the asset
times its risk exposure,
(
+1
,+1
)
(
+1
)
. The latter term is the beta of the particular asset, i.e.,
the slope coefficient from a regression that has the return on the asset as the dependent
variable and as the independent variable. This expresses a key feature of the CAPM. An
asset with a high beta commands a lower price (equivalently, it gives a higher expected
return) because it is more risky, as defined by the covariance with . The CAPM specifically
represents by the return on the market portfolio. This model has been tested systematically
by Fama and many others. More generally, several determinants of can be identified and
richer multi-factor models can be specified of the cross-section of asset returns, as stocks
9

generally covary differently with different factors. This approach has been explored
extensively by Fama and other researchers and will be discussed in Section 7.

3. Are returns predictable in the short term?
A long history lies behind the idea that asset returns should be impossible to predict if asset
prices reflect all relevant information. Its origin goes back to Bachelier (1900), and the idea
was formalized by Mandelbrot (1963) and Samuelson (1965), who showed that asset prices in
well-functioning markets with rational expectations should follow a generalized form of a
random walk known as a submartingale. Early empirical studies by Kendall (1953), Osborne
(1959), Roberts (1959), Alexander (1961, 1964), Cootner (1962, 1964), Fama (1963, 1965),
Fama and Blume (1966), and others provided supportive evidence for this hypothesis.
In an influential paper, Fama (1970) synthesized and interpreted the research that had been
done so far, and outlined an agenda for future work. Fama emphasized a fundamental problem
that had largely been ignored by the earlier literature: in order to test whether prices correctly
incorporate all relevant available information, so that deviations from expected returns are
unpredictable, the researcher needs to know what these expected returns are in the first place.
In terms of the general pricing model outlined in section 2, the researcher has to know how
the stochastic discount factor m is determined and how it varies over time. Postulating a
specific model of asset prices as a maintained hypothesis allows further study of whether
deviations from that model are random or systematic, i.e., whether the forecast errors implied
by the model are predictable. Finding that deviations are systematic, however, does not
necessarily mean that prices do not correctly incorporate all relevant information; the asset-
pricing model (the maintained hypothesis) might just as well be incorrectly specified.
4
Thus,
formulating and testing asset-pricing models becomes an integral part of the analysis.
5

Conversely, an asset-pricing model cannot be tested easily without making the assumption
4
The joint-hypothesis problem has been generalized by Jarrow and Larsson (2012). They prove that the proposition
that prices incorporate available information in an arbitrage-free market can be tested if the correct process for asset
returns can be specified. Specifying an asset-pricing model can be viewed as a special case of this, since such a model
implies an equilibrium process for asset returns.
5
One exception is when two different assets have exactly identical payoffs. In such a case, an arbitrage-free market
implies that these assets should trade at an identical price, regardless of any asset-pricing model. Hence, if we could
find instances where two such assets trade at different prices, this would violate the assumption that no arbitrage is
possible. Such violations have been documented in settings where market frictions limit arbitrage opportunities.
Examples include documentation by Froot and Dabora (1999) of price deviations of the Royal Dutch Shell stock
between the U.S. and Dutch stock market, and studies by Lamont and Thaler (2003) and Mitchell, Pulvino and
Stafford (2002), who looked at partial spinoffs of internet subsidiaries, where the market value of a company was less
than its subsidiary (implying that the nonsubsidiary assets have negative value).

10

that prices rationally incorporate all relevant available information and that forecast errors are
unpredictable. Famas survey provided the framework for a vast empirical literature that has
confronted the joint-hypothesis problem and provided a body of relevant empirical evidence.
Many of the most important early contributions to this literature were made by Fama himself.
In Fama (1991) he assessed the state of the art two decades after the first survey.
In his 1970 paper, Fama also discussed what available information might mean. Following
a suggestion by Harry Roberts, Fama launched the trichotomy of (i) weak-form informational
efficiency, where it is impossible to systematically beat the market using historical asset
prices; (ii) semi-strongform informational efficiency, where it is impossible to systematically
beat the market using publicly available information; and (iii) strong-form informational
efficiency, where it is impossible to systematically beat the market using any information,
public or private. The last concept would seem unrealistic a priori and also hard to test, as it
would require access to the private information of all insiders. So researchers focused on
testing the first two types of informational efficiency.

3.1 Short-term predictability
Earlier studies of the random-walk hypothesis had essentially tested the first of the three
informational efficiency concepts: whether past returns can predict future returns. This work
had addressed whether past returns had any power in predicting returns over the immediate
future, days or weeks. If the stochastic discount factor were constant over time, then the
absence of arbitrage would imply that immediate future returns cannot be predicted from past
returns. In general, the early studies found very little predictability; the hypothesis that stock
prices follow a random walk could not be rejected. Over short horizons (such as day by day),
the joint-hypothesis problem should be negligible, since the effect of different expected
returns should be very small. Accordingly, the early studies could not reject the hypothesis of
weak-form informational efficiency.
In his PhD dissertation from 1963, Fama set out to test the random-walk hypothesis
systematically by using three types of test: tests for serial correlation, runs tests (in other
words, whether series of uninterrupted price increases or price decreases are more frequent
than could be the result of chance), and filter tests. These methods had been used by earlier
researchers, but Famas approach was more systematic and comprehensive, and therefore had
a strong impact on subsequent research. In 1965, Fama reported that daily, weekly and
11

monthly returns were somewhat predictable from past returns for a sample of large U.S.
companies. Returns tended to be positively auto-correlated. The relationship was quite weak,
however, and the fraction of the return variance explained by the variation in expected returns
was less than 1% for individual stocks. Later, Fama and Blume (1966) found that the
deviations from random-walk pricing were so small that any attempt to exploit them would be
unlikely to survive trading costs. Although not exactly accurate, the basic no-arbitrage view in
combination with constant expected returns seemed like a reasonable working model. This
was the consensus view in the 1970s.

3.2 Event studies
If stock prices incorporate all publicly available information (i.e., if the stock market is semi-
strong informationally efficient, in the sense used by Fama, 1970), then relevant news should
have an immediate price impact when announced, but beyond the announcement date returns
should remain unpredictable. This hypothesis was tested in a seminal paper by Fama, Fisher,
J ensen and Roll, published in 1969. The team was also the first to use the CRSP data set of
U.S. stock prices and dividends, which had been recently assembled at the University of
Chicago under the leadership of J ames Lorie and Lawrence Fisher. Fama and his colleagues
introduced what is nowadays called an event study.
6
The particular event Fama and his co-
authors considered was a stock split, but the methodology is applicable to any piece of new
information that can be dated with reasonable precision, for example announcements of
dividend changes, mergers and other corporate events.
The idea of an event study is to look closely at price behavior just before and just after new
information about a particular asset has hit the market (the event). In an arbitrage-free
market, where prices incorporate all relevant public information, there would be no tendency
for systematically positive or negative risk-adjusted returns after a news announcement. In
this case, the price reaction at the time of the news announcement (after controlling for other
events occurring at the same time) would also be an unbiased estimate of the change in the
fundamental value of the asset implied by the new information.
6
A note on precedence is warranted here. The basic idea of an event study may be traced at least back to James
Dolley (1933), who studied the behavior of stock prices immediately after a split and provided a simple count of
stocks that increased and stocks that decreased in price. A contemporaneous event study was presented by Ball and
Brown (1968), and appeared in print a year before the 1969 paper by Fama et al. Ball and Brown acknowledge,
however, that they build on Fama and his colleagues methodology and include a working-paper version of that
paper among their references. Rather than casting doubt on the priority of the 1969 paper, this illustrates how fast
their idea spread in the research community.

12

Empirical event studies are hampered by the noise in stock prices; many things affect stock
markets at the same time making the effects of a particular event difficult to isolate. In
addition, due to the joint-hypothesis problem, there is a need to take a stand on the
determinants of the expected returns of the stock, so that market reactions can be measured as
deviations from this expected return. If the time period under study the event window is
relatively short, the underlying risk exposures that affect the stocks expected return are
unlikely to change much, and expected returns can be estimated using return data from before
the event.
Fama and his colleagues handle the joint-hypothesis problem by using the so-called market
model to capture the variation in expected returns. In this model, expected returns
,
are
given by

,
,

Here
,
is the contemporaneous overall market return, and
and
are estimated
coefficients from a regression of realized returns on stock i,
,
, on the overall market returns
using data before the event.
7
Under the assumption that
captures differences in expected

return across assets, this procedure deals with the joint-hypothesis problem as well as isolates
the price development of stock i from the impact of general shocks to the market.
For a time interval before and after the event, Fama and his colleagues then traced the rate of
return on stock i and calculated the residual
,
=
,

,
. If an event contains relevant

news, the accumulated residuals for the period around the event should be equal to the change
in the stocks fundamental value due to these news, plus idiosyncratic noise with an expected
value of zero. Since lack of predictability implies that the idiosyncratic noise should be
uncorrelated across events, we can estimate the value impact by averaging the accumulated
,
values across events.
The event studied in the original paper was a stock split. The authors found that, indeed,
stocks do not exhibit any abnormal returns after the announcement of a split once dividend
changes are accounted for. This result is consistent with the price having fully adjusted to all
available information. The result of an event study is typically presented in a pedagogical
7
The market model is closely related to the Capital Asset Pricing Model (CAPM), according to which
), where
is the risk-free rate and
is the expected market return. This is a sufficient but

not necessary condition for the market model to be a correct description of asset returns, but CAPM puts the
additional restriction on the coefficients that
= (1
. See Sharpe (1964).

13

diagram. Here we reproduce the diagram from a study by Asquith and Mullins (1986) of the
stock price reaction for 88 U.S. stocks around the time when the firms announced that they
would start paying dividends. Time 0 marks the day the announcement of the dividend
initiation was published in The Wall Street Journal, implying that the market learned the
dividend news the day before, i.e., at time -1. The diagram plots the cumulative abnormal
returns, i.e., the accumulated residual return
from 12 trading days before until 12 trading

days after the publication of the announcement. As seen in the diagram, dividend news is
quickly incorporated in stock prices, with a large stock price reaction of about 5% around the
announcement day, and insignificant abnormal returns before or after the announcement. This
pattern indicates that this type of news has no predictability.

Figure 1: Abnormal stock returns for initial dividend announcements

The event-study methodology may seem simple, but the force of the original study by Fama,
Fisher, J ensen and Roll and its results created a whole new subfield within empirical finance.
An event study arguably offers the cleanest way of testing for whether new information is
incorporated fully in prices, without generating predictable price movements. By and large,
the vast majority of event studies have supported this hypothesis. Some exceptions have been
found, however. The most notable and pervasive one probably is the so-called post-earnings
announcement drift, first documented by Ball and Brown (1968).
14

One of the most common uses of event studies is to measure the value consequences of
various events. If the market correctly incorporates the new information, the value effects of a
particular event, such as a corporate decision, a macroeconomic announcement or a regulatory
change, can be measured by averaging the abnormal returns across a large number of such
events for different assets and time periods. This method has become commonly used to test
predictions from various economic theories, in particular in corporate finance. See MacKinlay
(1997) and Kothari and Warner (2007) for reviews of this extensive literature.

3.3 Subsequent studies of short-term predictability
A flood of empirical studies using longer time series and more refined econometric methods
followed in the footsteps of the early work on predictability by Fama and others. Researchers
found statistically significant short-term predictability in stock returns, but that such
predictability is small in magnitude (e.g., French and Roll, 1986, Lo and MacKinlay, 1988,
Conrad and Kaul, 1988). The autocorrelation turns out to be stronger for smaller and less
frequently traded stocks, indicating that exploiting this predictability is very difficult, given
trading costs. Focusing on the very short horizon, French and Roll (1986) compare the
variance of per-hour returns between times when the market is open and weekends and nights
when the market is closed. It turns out that prices vary significantly more when the market is
open than they do over nights or weekends, measuring the price evolution per hour from
closing to opening. This finding is intriguing, unless the intensity of the news is
correspondingly much higher when the market is open. One interpretation is that uninformed
noise trading causes short-term deviations of price from its fundamental value. Consistent
with this, French and Roll found that higher-order autocorrelations of daily returns on
individual stocks are negative. Although the interpretation of these findings is still subject to
debate, a common explanation is that some of this predictability is due to liquidity effects,
where the execution of large trades leads to short-term price pressure and subsequent reversals
(Lehmann, 1990).
The research program outlined by Fama in his 1970 paper has by now yielded systematic
evidence that returns on exchange-traded stocks are somewhat predictable over short
horizons, but that the degree of predictability is so low that hardly any unexploited trading
profits remain, once transaction costs are taken into account. In this specific sense, stock
markets appear to be close to the no-arbitrage model with unpredictable forecasting errors.
Lack of short-term predictability does not, however, preclude that longer-term stock market
15

returns could display considerable predictability. Even if short-term returns are nearly
unpredictable, returns could quite possibly be predictable over longer time horizons. In the
next section we turn to the evidence on longer-term predictability.

4. Longer-term predictability
Studies of longer-term predictability have to confront the joint-hypothesis problem head on.
To the extent that one is willing to maintain the hypothesis of arbitrage-free pricing, long-term
return predictability would allow inference about the correct asset-pricing model. Conversely,
finding long-term predictability may suggest the existence of arbitrage opportunities given a
particular asset-pricing model.
Longer-term predictability of asset returns became a major research issue in the 1980s. The
seminal contributions are attributable to Shiller. Important early contributions were also made
by Fama; for example, Fama and Schwert (1977) showed that the short-term interest rate
could be used to forecast the return on the stock market.

4.1Variance ratio tests
Are expected market returns constant over time or do they vary in a predictable way? Shiller
addressed this question for bond markets (1979), as well as for stock markets (1981). He
realized that the simple no-arbitrage hypothesis, with a constant expected return, could be
tested by comparing the variance of asset returns in the short term and the long term. Until the
early 1980s, most financial economists believed that cash-flow news was the most important
factor driving stock market fluctuations. In the title of his 1981 paper, Shiller challenged this
view by asking, Do stock prices move too much to be justified by subsequent changes in
dividends?
To understand Shillers insight, recall that the basic pricing equation (1) implies that an asset
price in an arbitrage-free market can be written as an expected present value of future
fundamentals: the discounted value of future cash flows (dividends in the case of stocks),
where discounting is represented by future values of . As pointed out above, dividends as
well as the discount factor are stochastic. Let
,
denote the realization of the fundamental

value of a stock at time , i.e., the discounted sum of future realized dividends from time t+1
and onwards. This value is not known at time t but has to be predicted by investors. Any
16

unexpected movement in stock prices must come from a surprise change to
,
, either due to
a dividend movement or a movement in the stochastic discount factor. The theory thus says
that
,
=
[
,
], so that the forecast error,

,

,
, must be uncorrelated with any

information available today, in particular the current price. Otherwise the expectations would
not make rational use of the available information. Because by definition
,

,
+

,
and the price and the forecast error are uncorrelated, it follows that (
,
) =
(
,
) +
,

,
, i.e., the variance of the realized fundamental value
in a no-
arbitrage market equals the sum of the variance of the price P and the variance of the forecast
error. This implies that
,
>
,
. In other words, the variance of the price must
be smaller than the variance of the realized discounted value of future dividends.
To investigate this relation empirically, Shiller (1981a) assumed a constant discount factor,
which implies that (realized) fundamentals are given by
=1

,+
.
The resulting time series, based on dividends in the New York Stock Exchange, is displayed
in the figure below together with the stock index itself. The contrast in volatility between the
two series is striking. Contrary to the implication of the present-value model with constant
discount rates, the price variance is much larger than the variance of the discounted sum of
future dividends.
8

The early excess-volatility findings were challenged on econometric grounds by Marsh and
Merton (1986) and Kleidon (1986), who noted that the test statistics used by Shiller (1979,
1981) are only valid if the time series are stationary. This issue was addressed by Campbell
and Shiller (1987). They used the theory of cointegrated processes, which had been recently
developed by the 2003 Laureates Clive Granger and Robert Engle
9
, to design new tests of the
present-value model that allow the processes generating prices and dividends to be
nonstationary. The model was again rejected, even under these more general and realistic
conditions. The paper by Campbell and Shiller (1987) also was important in showing how
8
At about the same time, and independently, LeRoy and Porter (1981) also studied the excess volatility of stock
prices using a different methodology, where they constructed a joint test of price volatility and payoff volatility from
a bivariate model for dividends and prices. They found evidence of excess volatility, but it appeared to be of
borderline statistical significance.
9
See Engle and Granger (1987).

17

cointegration methods can be used as a natural extension of Famas (1970) notion of weak
form tests.
10

Figure 2: Real Standard and Poors Composite Stock Price Index (solid line p) and ex post rational price (dotted
line p*), 18711979, both detrended by dividing a long-run exponential growth factor. The variable p* is the
present value of actual subsequent real detrended dividends, subject to an assumption about the present value in
1979 of dividends thereafter.

4.2 Predictability in stock returns
The finding that stock and bond returns are more volatile in the short term than in the long
term implies that returns are mean reverting, i.e., above-average returns tend to be followed
by below-average returns and vice versa. This also implies that future returns can be predicted
from past returns. Evidence that stock returns may be predictable in the medium and long
term had started to emerge already in the 1970s. Basu (1977, 1983) documented that stocks
with high earnings-to-price or dividend-to-price ratios outperform stocks with low ratios.
Fama and Schwerts (1977) investigation of the relationship between stock returns and
inflation showed that periods of high short-term interest rates tend to be followed by lower
subsequent stock-market returns.
10
Campbell and Shiller were also inspired by the work of Hansen and Sargent (1980), who showed how the concept
of Granger causality could be used in testing rational-expectation models.

18

The finding that stock prices are excessively volatile relative to dividends made it natural to
focus on current dividend levels as a predictor of future returns. Shiller (1984) studied U.S.
stock market data going back to the 1870s. By regressing the one-year-ahead rate of return on
the current dividend-price ratio, he found a positive relationship: high dividends relative to
price predict above-normal returns. Apparently, an investor could earn higher returns by
going against the market, buying when prices are low relative to dividends and selling when
prices are high. In a later paper, Campbell and Shiller (1988a) studied the predictive power of
a long moving average of real earnings. They found that this variable has a strong power in
predicting future dividends and that the ratio of this earnings variable to the current stock
price is a powerful predictor of future stock returns. Other early studies of stock-return
predictability include Keim and Stambaugh (1986) and Campbell (1987). These and other
studies identified a variety of variables that forecast future stock returns. Typically these
variables are correlated with key macroeconomic indicators, suggesting that the discount
factor varies with the state of the business cycle.
Consistent with the limited predictability over very short horizons discussed in section 3,
Fama and French (1988a) documented that predictability increases with the horizon. This
finding is illustrated in the table below, taken from Cochrane (2001). Over a one-year horizon,
the dividend/price ratio explains 15% of the variation in excess returns, but over a five-year
horizon, the explanatory power is as high as 60%.
11

Coefficients from regressing excess returns over different horizons
on the ratio of dividend to price
Horizon (years) Coefficient
(standard error)
R
2
1 5.3 (2.0) 0.15
2 10 (3.1) 0.23
3 15 (4.0) 0.37
5 33 (5.8) 0.60
11
These regressions are associated with some econometric problems. The dividend-yield series is very persistent, and
return shocks are negatively correlated with dividend-yield shocks. As a result, the return-forecast regression inherits
the near-unit-root properties of the dividend yield. For such time series, standard test statistics may suffer from small
sample biases. Nelson and Kim (1993) and Stambaugh (1999) have proposed methods for dealing with this problem.
See also Cochrane (2007).

19

In a related contribution, Campbell and Shiller (1988b) explore the determinants of the
dividend-price ratio,
. Basic pricing theory implies that this ratio should reflect

expectations of future dividend growth and discount rates. In the simplest case of no
uncertainty, constant dividend growth at rate g, and a constant discount rate R, the pricing
expression simplifies to
= , the so-called Gordon formula. In general, however,

given its nonlinearity, implementing an asset-pricing equation for empirical studies is not
straighforward. The methodology developed by Campbell and Shiller allows an analyst to
gauge to what extent variations in d/P can be explained by variations in expected dividends
and discount rates, respectively. It builds on a linearization that decomposes the logarithm of
d/P into a weighted sum of future expected log discount rates and log dividend changes. To
generate expectations, Campbell and Shiller estimated a vector-autoregression system based
on alternative measures of discount rates, e.g., interest rates and consumption growth. They
found some evidence that d/P is positively affected by future dividend growth. None of the
discount rate measures used, however, helped to explain the dividend-price ratio, and overall,
most of the variation in this ratio remained unexplained. The Campbell-Shiller decomposition
has become very influential both by providing an empirical challenge for understanding what
drives asset prices and by providing a methodology for addressing this challenge.

4.3 Predictability in other asset markets
The findings of excess volatility and predictability by Shiller and others turned out to be a
pervasive phenomenon, not only in the stock market but also in other asset markets. As a
precursor to his work in 1980, Shiller (1979) already found evidence of excess volatility for
government bonds. Under the assumption of a constant risk premium (the so-called
expectations hypothesis), long-term interest rates should equal weighted averages of expected
future short-term rates, and consequently the volatility of long-term rates should be smaller
than the volatility of short-term rates. Shiller found just the opposite. The volatility of long-
term rates turned out to be many times larger than the volatility of short-term rates. Similar to
stock prices, the excess volatility of long-term bond prices implies that bond returns are
predictable. Subsequently, Shiller, Campbell and Schoenholtz (1983), Fama and Bliss (1987),
and Campbell and Shiller (1991) all found that the slope of the U.S. Treasury yield curve
predicts bond returns at all maturities. Moreover, Campbell (1987) and Fama and French
20

(1989) showed that the term structure of interest rates predict stock returns as well, and that
excess returns on long-term bonds and stocks move together.
Similar results were found in foreign exchange markets. According to the expectations
hypothesis, forward exchange rates should be equal to expected spot rates. The expectations
hypothesis implies that the so-called carry trade, which involves borrowing in a low-interest
currency and investing in a high-interest currency, should not yield positive excess returns, as
the higher interest rate should be offset by currency depreciation. Hansen and Hodrick (1980)
developed econometric tests using multiple forward rates of different maturities, and were
able to reject the expectations hypothesis in foreign exchange markets.
12
Similarly, Fama
(1984) showed that the coefficient of the forward rate in a regression on future spot rates is
actually negative, rather than plus one as the expectations hypothesis would predict. These
studies, as well as many others that followed, indicated that foreign exchange markets exhibit
significant return predictability as well.
The upshot from these results is that the volatility and predictability of stock, bond and
foreign exchange returns can only be consistent with arbitrage-free markets if the expected
return, i.e., the discount factor, is highly variable over time. The question then is whether
theoretical models are able to generate such high variability in the discount factor.

5. Risk premia and volatility in rational-agent models
The findings of excess volatility and predictability and related findings, such as high return
premia on stocks by Shiller and other researchers illustrate the need for a deeper
understanding of what drives the variation in expected returns over time. A major line of
research, initiated in the 1970s, continues to strive to construct dynamic asset-pricing models
that build on optimizing behavior, implying arbitrage-free prices. In a dynamic model, risk
preferences of investors can vary over time, e.g., as a result of consumption or wealth shocks,
thus generating fluctuations in risk premia and predictability of returns.

12
The econometric approach taken by Hansen and Hodrick (1980) can be viewed as a precursor to Hansens (1982)
GMM, discussed in section 5.3 below.

21

5.1 The consumption capital-assetpricing model (CCAPM)
The most basic dynamic pricing model, the CCAPM, starts from the assumption that the
economy can be described by a representative agent who maximizes expected utility given by

(
+
=0
)|
,
where u is a utility function of consumption c and is the subjective discount factor. Here, we
write the conditional expectation
() as ( |
) in order to specify explicitly the information

set
on which the expectation is based. The agent faces a simple budget constraint
,
+
,1
(
,
+
,
) +
,
where
,
is the number of units invested in the risky asset i at time t,
,
is the dividend
generated by that asset, and
is labor income at time t. The key equation of CCAPM is the

first-order condition for utility maximum:
) =
(
+1
)
,+1
,
|
.
Here, as before,
,+1

,+1
+
,+1
is the assets payment at time t + 1. An optimizing
agent is indifferent between consuming a unit at time t, thus receiving the marginal utility of
one unit at that time (the left-hand side), and investing it to earn a rate of return
+1
/
and
obtaining the discounted marginal utility from consuming that at t + 1 (the right-hand side).
This so-called Euler equation can be rewritten as an asset-pricing equation:

,
=
(
+1
)
)

,+1
|
. (2)
Equation (2) is thus a present-value equation of the same type as the one derived from the
absence of arbitrage, equation (1), with the stochastic discount factor
+1
now given by
(
+1
)/
), which is the marginal rate of substitution between consumption today and

tomorrow. This equation shows why the discount rate would be low during recessions: in bad
times, when
is low, the marginal utility
) is high, and thus the ratio of marginal

utilities
(
+1
)/
) is correspondingly low (conversely, the discount factor should be

high during booms).
The CCAPM would thus seem to give a possible qualitative explanation for the findings of
predictability and excess volatility based on rational behavior. What about the quantitative
content of the theory?

22

5.2 Testing the consumption capital-assetpricing model (CCAPM)
Confronting economic theory with data is a methodological challenge, especially when the
theory gives rise to nonlinear dynamic equations. For that reason, researchers often evaluate
models informally, for example, by using calibration, where model parameters are selected
based on non-statistical criteria and the model is solved and simulated. By comparing the
resulting model-generated time series to actual data, calibration can be useful in assessing
whether a model may be capable of quantitatively matching the actual data at all. A more
rigorous approach, of course, would be to use formal statistical methods. But before the
1980s, the methodological challenges were daunting, and not until Hansens development of
the GMM did formal tests of the CCAPM become commonplace. Thus, empirical evaluation
of the CCAPM began with informal methods.

using calibration and informal statistics
Grossman and Shiller (1981) were the first to evaluate the CCAPM quantitatively. They
assumed utility to be given by a power function (implying constant relative risk aversion).
The discount factor
(
+1
)/
) can then be calculated from consumption data for any

given value of relative risk aversion. Using U.S. consumption data, Grossman and Shiller
found, however, that the observed stock-price volatility could only be consistent with
CCAPM if the marginal utility of consumption was extremely sensitive to variations in
consumption, i.e., if the representative consumer was extremely risk averse. This finding left
excess volatility as a challenge for future asset-pricing research. Subsequently, Shiller (1982)
showed that the model implied a lower bound on the marginal rate of intertemporal
substitution. This insight is a precursor to the influential contribution by Hansen and
J agannathan (1991), which is further discussed in Section 5.3.
In passing, Grossman and Shiller also noted that the CCAPM implied a much lower level of
equity returns than observed in data, hence providing an early illustration of what Mehra and
Prescott (1985) subsequently came to term the equity-premium puzzle. In their paper, Mehra
and Prescott highlighted the extreme difficulty that traditional models have in matching an
observed excess return of stocks relative to a risk-free asset of over 5% per year, a magnitude
that had been observed in data for the U.S. and many other countries. To match the data,
coefficients of relative risk aversion of around 50 were needed, and such levels of risk
23

aversion were viewed to be unrealistic from an applied microeconomic perspective, at least
for the average investor.

and using formal statistical methods
As discussed above, the CCAPM implies that returns are predictable as long as agents are risk
averse and variations in consumption can be predicted. However, in order to test this theory
researchers face several difficulties. One difficulty is the inherent nonlinearity of the main
estimating equation. Another is the need to specify a full stochastic process for consumption.
In fact, these difficulties, along with serial correlation of any errors in the dynamic system, are
shared by a large set of models used in economics. In the early 1980s, the only way to handle
these difficulties was by making a range of very specific assumptions assumptions that were
not even perceived to be central to the main issue at hand. Thus, any statistical rejection
would be a rejection of the joint hypothesis of the main asset-pricing equation and all the
specific assumptions to which the researcher was not necessarily wed.
An influential illustration of this point was given by Hansen and Singleton (1983), who dealt
with the difficulties by a combination of approximations and specific assumptions. Assuming
jointly normally distributed error terms, they developed the following log-linear version of
CCAPM:
ln1 +
,+1
= ln
[ln c
t+1
] + [
+
2
]/2.
This equation expresses expected log returns as the sum of three terms: the log rate of time
preference , a term that is multiplicative in the rate of risk aversion and the expected rate of
consumption change, and a term that depends on variances and covariances. Hansen and
Singleton then estimated this linearized model for monthly stock returns, using a maximum-
likelihood estimator. Based on a value-weighted stock index, the model worked relatively
well, giving estimates of relative risk aversion between zero and two and showing little
evidence against the parameter restrictions. When estimated based on returns of individual
stocks and bonds, however, the model was strongly rejected. This failure is an early indication
of a serious challenge to the rational-agentbased asset-pricing model. At the time, however,
it was unclear how much of the rejection was due to the linearization and error process
assumptions and how much was an inherent limitation of the theory. The GMM provided a
way to address these problems.

24

5.3 The Generalized Method of Moments (GMM)
The asset-pricing context
Consider again the main equation of the CCAPM model, equation (2). Defining
,+1

,+1
/
, it can be rewritten as
1 =
(
+1
)
)

,+1
|
. (3)
This is a nonlinear function of the stochastic processes for consumption and returns and any
relevant additional variables in the conditioning set
. The expression
(
+1
)
)

,+1
1 can
be regarded as a one-period-ahead forecast error. Under rational expectations, this error must
be independent of any information It available at time t. Let us use zjt to denote a variable in
the information set It, e.g., a historical asset price. This implies, for any asset i and
conditioning variable or instrument zj, that

(
+1
)
)

,+1
1
= 0. (4)
This equation, which is an implication of equation (3), is the basis for GMM estimation of an
asset-pricing model.

Some econometric theory
Equation (4) can be viewed as an element in the following vector equation
(
, ) = (5)
where
is a vector stochastic process (sequence of random variables) and is a parameter

vector to be estimated. The vector-valued function expresses the key orthogonality
condition one equation for each asset i and instrument
. In the example,
consists of ,
(for all assets ), and
(for at least one instrument j), and consists of and the other
parameters in (4). The (i,j)th element of the vector would thus be
(
+1
)
)

,+1
1
,
for asset and a particular instrument
. This element has expectation zero and can be

interpreted as a form of forecast error.
In a paper that has turned out to be one of the most influential papers in econometrics, Hansen
(1982) suggested the GMM as an attractive approach for estimating nonlinear systems like
equation (5). A main reason why this estimator has become so popular is that it places only
25

very weak restrictions on the stochastic process
, which is allowed to be any weakly

stationary, ergodic process, and on , which is allowed to be nonlinear. This generality is
particularly important in panel-data and time-series applications, such as asset pricing ones,
where the stochastic process is correlated and the key relationships are nonlinear. Moment
conditions such as (5) had been used in parameter estimation since Pearson (1894, 1900), see
also Neyman and Pearson (1928), but their use had been confined to cases where the
components of
are independent over time, e.g., as in the case of repeated independent

experiments. Hansens contribution was to generalize the previous theory of moment
estimation to the case where
is a stationary and ergodic process.

The GMM estimator can be defined using the sample moment function
()
1
, )
=1

and the quadratic form
()
()
(),
where is a positive definite weight matrix. The GMM estimator
minimizes
().
Hansen (1982) showed that this estimator is consistent for the true parameter vector under
certain regularity conditions and that it is asymptotically normal given some mild restrictions
on (
, ). As already indicated, the proof allows rather general stochastic temporal

dependence for the stochastic process
.
Furthermore, Hansen defined the asymptotic covariance matrix

()
()
=
.
Hansen showed that the selection =
1
ensures that the resulting estimator
minimizes
(in the matrix sense) the asymptotic covariance matrix of the estimator. This result provides
an asymptotic efficiency bound for the GMM estimator a bound because the true is not
known.
Hansen also showed how to estimate the asymptotic covariance matrix and, using its inverse
as a weighting matrix, derived the resulting asymptotic normal distribution. Hansens
construction of the estimate of is based on a consistent estimate of for the sample at hand,
but at the same time, the estimated is needed to construct an efficient estimate of This
conundrum means that there is no straightforward way of obtaining the efficient estimate.
Hansen therefore proposed a two-stage procedure: start with an arbitrary weighting matrix
26

and use it to construct a consistent estimator and use that estimator to estimate the asymptotic
covariance matrix; then use that matrix to obtain the efficient estimator of . Alternative
procedures were proposed later to improve on this two-stage approach.
Finally, Hansen demonstrated how to construct a test of over-identifying restrictions, based on
a method proposed by Sargan (1958). Under the null hypothesis this test statistic has an
asymptotic
2
distribution with degrees of freedom, where is the number of moment
conditions and the number of linear combinations of these conditions (to find parameters
of interest).
13

In summary, Hansen provided the necessary statistical tools for dealing with estimating
dynamic economic models using panel data, where serially correlated variables are
commonplace and where specifying a full model is not always desirable or even possible;
GMM can be applied to a subset of the model equations. GMM has made a huge impact in
many fields of economics where dynamic panel data are used, e.g., to study consumption,
labor supply or firm pricing. It is now one of the most commonly used tools in econometrics,
both for structural estimation and forecasting and in microeconomic as well as
macroeconomic applications.
14

The asset-pricing application
Equipped with GMM, researchers analyzing asset prices could now go to work. The first
direct application of Hansens GMM procedure is reported in the paper on asset-pricing by
Hansen and Singleton (1982). But an earlier use of the essential idea behind GMM can be
found in work by Hansen and Hodrick (1980), who looked at currencies and asked whether
forward exchange rates are unbiased predictors of future spot rates. Serial correlation in errors
and nonlinearities make traditional approaches invalid for this issue, and the authors derived
asymptotic properties based on methods that turned out to be a special case of GMM.
The main purpose of Hansen and Singleton (1982) was to test the CCAPM. To operationalize
the model, the authors assumed utility, as did Grossman and Shiller (1981), to display
13
Hansen has followed up his seminal piece with a number of important extensions, including alternative estimators
(Hansen, Heaton and Yaron, 1996), the choice of instruments (Hansen, 1985, and Hansen, Heaton and Ogaki, 1988),
continuous-time models (Hansen and Scheinkman, 1995), and GMM with non-optimal weighting matrices (Hansen
and Jagannathan, 1997).
14
See the review articles by Hansen and West (2002) and Jagannathan, Skoulakis and Wang (2002) for illustrations of
the use of GMM in macroeconomics and finance. In microeconometrics, GMM has also been a commonly used
model for estimation with panel data see Arellano and Bond (1991) and Blundell and Bond (1998).

27

constant relative risk aversion: () =
(1)
/(1 ). With this specification, an element of
, representing a certain asset and an instrument
, takes the form

+1

,+1
1
, with and as the parameters to be estimated. The corresponding moment condition thus
becomes
1
+1

,+1
1
=1
= 0.
With n assets and m instruments, there are nm such moment conditions. With fewer than nm
parameters to estimate, the model may be tested for over-identifying restrictions. This
instrumental-variables formulation illustrates the point made by Fama (1970): testing an asset-
pricing model amounts to a joint test of the model-generated hypothesis and the lack of
predictable forecast errors. If the over-identifying restrictions are rejected, this means either
that the model is incorrect i.e., that the no-arbitrage condition is violated or that the
orthogonality condition of the instruments is violated, or both.
Hansen and Singleton (1982) estimated this model by GMM, using lagged values of
as
instruments. The data are aggregate indexes for the New York Stock Exchange as well as
indexes for different industries, and the model is estimated both on single and multiple return
series. All versions of the model yield economically meaningful estimates with close to
unity (although with a large standard error) and slightly smaller than unity. When applied to
more than one stock index, the over-identifying restrictions are generally rejected, however.
15

Hence, in line with the excess-volatility findings of Grossman and Shiller (1981), this simple
version of CCAPM does not fit the data very well. This result has led to a vast amount of
research aimed at understanding the shortcomings of the basic model.
In the search for a model that better fits the data, a diagnostic tool that states the properties
that the stochastic discount factor must possess would be useful. Hansen and J agannathan
(1991) showed that the so-called Sharpe ratio
16
expressed by the ratio of the expected
excess return of an asset over the risk-free rate to the standard deviation of the excess return
gives a lower bound to the volatility of the discount factor. Specifically,
(
+1
)
(
+1
)

(
+1
+1

,
15
In a later paper, Hansen and Singleton (1984) corrected an error in their original data series. Using the revised data,
the CCAPM was more strongly rejected.
16
Sharpe (1966).

28

where the left-hand side is the ratio of the standard deviation of the discount factor to its
expected value and the right-hand side is the Sharpe ratio. This relation was originally stated
by Shiller (1982) for a single risky asset and generalized by Hansen and J agannathan to cover
many assets and no risk-free asset. In subsequent work, Hansen and J agannathan (1997)
extended their analysis and derived formal tests of the performance of different stochastic
discount factor proxies.
Hansen-J agannathan bounds have become widely used in practical applications. Many assets
and investment strategies, such as momentum (going long in stocks with high past returns and
shorting stocks with low past returns; see Section 7) or carry trade (borrowing in low-interest-
rate currencies and investing in high-interest-rate currencies) have very high Sharpe ratios.
For the postwar U.S. stock market, the Sharpe ratio in annual data is around one half, which
implies that the annualized standard deviation of the discount factor has to be at least 50%,
which is very high considering that the mean of the discount factor should be close to one.
This discrepancy poses a serious problem for consumption-based models such as the
CCAPM, since the low volatility of observed consumption, along with a realistic level of risk
aversion, implies too low a volatility of the stochastic discount factor according to CCAPM.

5.4 Extensions of the CCAPM
Rejection of the CCAPM with standard preferences does not necessarily reject the basic
intuition of the model, i.e., that the expected return on equity is higher in bad times when
current consumption is low. Starting with the study by Fama and French (1989), several
studies have related predictability to business cycle conditions, showing that expected returns
are lower at the peak of the business cycle and higher in the trough. Fama and French (1989)
also showed that expected returns in equity markets and bond markets move together, and that
the term premium (the difference in yields between long- and short-term bonds) has additional
predictive power for stock returns, in addition to dividend yields. Similarly, macro variables
such as the consumption-wealth ratio (Lettau and Ludvigson, 2001) have been shown to
predict equity returns (in addition to dividend yields and term premia).
17
Rather, the problem
is that the covariation between asset returns and consumption is not large enough to generate
high enough expected returns and volatility using standard expected utility preferences.
17
See Cochrane (2011) for an overview.

29

These results have led many researchers to explore alternative model specifications, changing
assumptions about investor utility, market completeness, the stochastic process for
consumption or these assumptions in combination. Several of these approaches have had
some success in explaining equity premia, volatility and predictability within a modified
CCAPM framework, although it is fair to say that currently no widely accepted consensus
model exists.
One approach has been to address one of the main shortcomings of the standard von
Neumann-Morgenstern expected utility model, namely that the same parameter determines
both risk aversion and intertemporal substitution, even though there is no compelling
economic or behavioral reason for this to be the case. Building on Kreps and Porteus (1978),
Epstein and Zin (1989) developed a class of recursive preferences that allow preferences for
risk and intertemporal substitution to be separated and argued that these preferences could
help resolve the consumption-based model.
18
Hansen contributed to this line of research
(Eichenbaum, Hansen and Singleton, 1988, and Eichenbaum and Hansen, 1990). This
research program is still very active, with some success in improving the models fit with
data. Using Epstein-Zin preferences, Bansal and Yaron (2004) proposed a model where
consumption and dividend growth contain a small predictable long-run component, and
consumption volatility is time-varying. Given these preferences and dynamics, Bansal and
Yaron were able to generate a stochastic discount factor m that can justify the observed equity
premium, the risk-free rate and the return volatility and that also generates dividend-yield
predictability. This approach has been quite influential and has led to a number of follow-up
studies, including one from Hansen, Heaton and Li (2008).
A second approach to modify preferences has been to introduce habits into the utility function
(Deaton, 1992) and make consumer utility not just dependent on the absolute level of
consumption, but also sensitive to changes in consumption levels. Thus, Sundaresan (1989),
Constantinides (1990) and Abel (1990) included habit formation in the CCAPM framework,
and showed that habits can increase the volatility of the stochastic discount factor. In a highly
cited study, Campbell and Cochrane (1999) were able to explain the equity premium puzzle in
a model where an external habit (which makes agents care about changes in aggregate, and
not only individual consumption) is added to the standard power-utility framework.
A third approach that has also met with some success, is to consider heterogeneity in investor
preferences. In particular, if investors have different attitudes toward risk, the stochastic
18
See also Weil (1989).

30

discount factor m that appears as a result of market trading will be influenced not just by
aggregate consumption but also by its distribution. To understand the equity premium, for
example, it might be more relevant to test the CCAPM implications using data from the sub-
group of investors actually owning significant amounts of stock. Indeed, it turns out that the
consumption of individual stockholders fluctuates more than does aggregate consumption, a
difference that at least goes part of the way toward explaining the pricing puzzles and
confirms investor heterogeneity as a fruitful hypothesis (see, e.g., Malloy, Moskowitz and
Vissing-J orgensen, 2009). A large number of studies of market incompleteness against
individual risks show that wealth heterogeneity that is a result of individual wage shocks
generates heterogeneity in risk attitudes (for early contributions, see Mankiw, 1986, Heaton
and Lucas, 1992, Huggett, 1993, Telmer, 1993, and Constantinides and Duffie, 1996). An
important finding reported in this literature is that individual wage risk is countercyclical,
which helps explain the pricing puzzles further.
A common feature of most of the models discussed here is the assumption that the consumer
does not only process information in a rational and efficient manner, but also knows the true
data generating process. In joint work with Thomas Sargent (e.g., Hansen and Sargent, 2001,
and Cagetti et al., 2002), Hansen has investigated the consequences of assuming that the
representative agent is uncertain about the true model and follows a policy of robust control
across a set of alternative models. Hansen and Sargent showed that model uncertainty can be
seen as an extra risk factor; the fear of a worst outcome makes the risk aversion of agents
effectively larger and can account for a higher price of risk than in an equivalent standard
model.

6. Excess volatility and predictability: behavioral-finance approaches
6.1 Robert Shiller and behavioral finance
The findings of excess volatility and predictability are challenging for the notion that prices
incorporate all available information or for standard asset-pricing theory or for both. Based
on his early findings, Shiller (1981b) argued that the excess volatility he documented seemed
difficult to reconcile with the basic theory and instead could be indicative of fads and
overreaction to changes in fundamentals. In his 1984 paper, entitled Stock Prices and Social
Dynamics, he developed these arguments further. This paper became an important starting
31

point for a growing research literature in behavioral finance, for which Shiller became one
of the most influential proponents.
19

In his paper, Shiller outlined a number of arguments that were followed and developed by
subsequent researchers. First, he argued that the lack of (risk-adjusted) price predictability
does not preclude the existence of irrational investors. The trading of such investors could
make prices excessively volatile and noisy, which should make deviations from a random
walk very hard to detect over short horizons (especially if rational investors would tend to
eliminate the most obvious mispricings). In subsequent work, Shiller and Perron (1985) and
Summers (1986) argued more formally that the power of short-run predictability tests is likely
to be very low.
20

Second, Shiller reviewed some of the psychology literature showing that individuals are
subject to decision biases, such as Tversky and Kahnemans (1974) finding of people
overreacting to superficially plausible evidence without any statistical basis. Shiller argued
that stock prices are particularly vulnerable to psychological biases because of the ambiguity
in the true value of a stock, due to the lack of an accepted valuation model (i.e., investors face
Knightian uncertainty rather than risk). These psychological biases are reinforced and
exacerbated by social movements because investors are subject to group psychology
dynamics, such as peer pressure. Hence, one investors opinion of the value of a stock is
likely to be affected by the opinions of others. As a result, as opinions diffuse throughout the
population, stock prices will fluctuate in a way similar to what would be caused by fads or
fashions. Shiller also reviewed informal evidence that supported the idea that fads and
fashions had contributed to past market booms and busts.
Finally, to illustrate his argument more formally, Shiller posited a simple model economy
populated by ordinary investors, whose demand does not respond to expected returns, and
smart money investors who respond rationally to expected returns but are limited by their
wealth. In such a model, the trades of ordinary investors will lead to temporary deviations of
19
Other early work that met similar challenges with behavioral explanations was done by Slovic (1972), Miller (1977),
Harrison and Kreps (1978), and Modigliani and Cohn (1979). Slovic (1972) argued that the (then recent)
psychological evidence on heuristics and biases can be applied to finance. Miller (1977) argued that differences of
opinion among investors, together with the fact that stocks are difficult to sell short, will lead to over-optimistic
stock prices. Harrison and Kreps (1978) made a similar argument in a dynamic setting, and showed that differences
in opinion and short-sales constraints can lead to speculative bubbles. Modigliani and Cohn (1979) argued that the
negative relation between expected inflation and stock prices (documented by Fama and Schwert, 1977) could be
explained by investors suffering from money illusion, i.e., an inability to distinguish real from nominal returns.
20
Shiller and Perron (1985) derived power functions for tests aiming to reject the random walk hypothesis in a runs
test (e.g., Fama, 1965) when the true asset return process is mean reverting. Apart from showing that these tests have
low power when the time span of the data is short, they also showed that increasing the sampling frequency for a
given time span does not increase power only increasing the length of the span does.

32

stock prices from fundamental values, and these deviations can generate overreaction to
dividend news, excess volatility, and mean reverting stock prices, consistent with the finding
that high dividend yields predict lower stock prices.
In subsequent work, Shiller has continued to argue for the importance of social psychology,
using evidence from investor surveys (Shiller, 1987, 1988, 1989, Shiller and Pound, 1989).
He has also extended his analysis of fads and bubbles to housing markets (Case and Shiller,
1987, 1989, 2003).
Following Shillers original work, many researchers turned to psychological evidence on
individual behavior and biases, including prospect theory (Kahneman and Tversky, 1979),
overconfidence (Oskamp, 1965) and mental accounting (Kahneman and Tversky, 1984, and
Thaler, 1985). This psychology-based work has derived new asset-pricing models that could
explain the documented asset-pricing anomalies.
21
Some models attribute under- and
overreaction to information due to overconfidence and/or bounded rationality, which leads to
excess volatility, momentum and mean reversion in asset prices (Barberis et al., 1998, Daniel
et al., 1998, Hong and Stein 1999). Other models modify preferences based on psychological
evidence such as prospect theory (Benartzi and Thaler, 1995, Barberis et al., 2001), ambiguity
aversion (Epstein and Wang, 1994, Epstein and Schneider, 2008, Cagetti et al., 2002) or
disappointment aversion (Routledge and Zin, 2010) to explain excess volatility, predictability
and the equity premium. Many of these papers can be thought of as modifying the preference
assumptions in rational-agent models, such as the CCAPM, which illustrates a convergence
between rational and behavioral models in recent research.
In his 1984 paper, Shiller also addressed a serious criticism against behavioral explanations
(often attributed to Friedman, 1953), namely that even if some (or most) investors are
irrational, the (perhaps few) rational investors in the market could make money as long as
there were arbitrage opportunities. Such arbitrage trades would lead the irrational investors to
lose money and be forced out of the market, ultimately eliminating any mispricing. Shiller
argued that rational investors control too little wealth for this to work in practice. Another
early argument for limits to arbitrage is the difficulty of short-selling overpriced stocks
(Miller, 1977). Subsequently, a number of researchers provided more rigorous theoretical
models explaining the limited ability of rational investors to make markets informationally
efficient. A common approach is to model rational arbitrageurs as financial intermediaries
(e.g., hedge funds), whose capital is withdrawn by investors in case they experience persistent
21
See Shleifer (2000) and Barberis and Thaler (2003) for overviews of this literature.

33

losses (DeLong et al., 1990a, Shleifer and Vishny, 1997). Because of such withdrawals, the
rational arbitrageurs may not be able to trade against substantial market mispricing, or they
may even find it optimal to trade in the opposite direction if the mispricing is expected to
increase in the short-run, thus increasing the mispricing rather than decreasing it (DeLong et
al., 1990b, Abreu and Brunnermeier, 2002, 2003).

6.2. Further work in behavioral finance
Shillers early studies stimulated a large body of empirical research aimed at backing up the
behavioral-finance arguments with empirical evidence.
22
Many of these studies have focused
on apparent anomalies in the cross-section of stock returns, rather than mispricing in the
market as a whole, and are reviewed in section 7.
An early study that found the evidence hard to reconcile with informational efficiency was
Rolls (1984) investigation of the orange juice futures market. Even though weather is the
most obvious and significant influence on orange crops, Roll found that weather surprises
explain only a small fraction of the variability in futures prices.
A number of studies have documented deviations from the law of one price in financial
markets and argue that these deviations are indicative of irrational market sentiment. Froot
and Dabora (1999) studied twin stock companies with stocks traded in more than one
location. They found that the prices of these twins frequently differ across trading locations,
and that a twin's relative price rises when the market on which it is traded more intensively
rises. This suggests that stock prices are driven to some extent by local investor sentiment,
rather than simply by changes in fundamental values.
Another group of studies analyze the so-called closed-end fund puzzle originally discovered
by Zweig (1973), i.e., the finding that closed-end equity funds typically trade at values
different from the market value of their underlying stock portfolio, and frequently at a
22
This discussion focuses on the behavioral finance literature, which seeks to explain asset-pricing patterns and
anomalies. Parallel to this literature, a vast behavioral literature has also emerged that analyzes the impact of
psychology on financial decisions of individuals. See Barberis and Thaler (2003), section 7, for an overview. This
literature has been influential in providing practical policy advice, e.g., in the area of individual pension saving
schemes (see Thaler and Benartzi, 2004).

34

discount to their net asset value.
23
Lee, Shleifer and Thaler (1991) argued that the discounts
on closed-end funds can be interpreted as a measure of irrational investor sentiment. They
showed that the discounts across closed-end funds with very different asset portfolios exhibit
significant co-movement over time, and also co-move with returns on small stocks, a market
segment where individual investors are also dominant. Baker and Wurgler (2007) reviewed
the investor sentiment literature and showed that a sentiment index (including closed-end
fund discounts and other variables) is highly correlated with aggregate stock returns.
Another group of papers documented apparent limits to arbitrage and its effect on stock
prices. Starting with Shleifer (1986), a number of papers have shown that the price of a stock
tends to increase when it is included in a market index (such as the S&P 500), consistent with
buying pressure from index funds. Wurgler and Zhuravskaya (2002) showed that the index
inclusion effect is stronger among stocks without close substitutes, which makes mispricings
harder to arbitrage away.
Shiller developed his arguments further in his book Irrational Exuberance (2000), which had
a considerable impact on the popular debate. In this book, he used a combination of statistical
evidence, survey data, and a review of psychological and sociological studies to argue that
fads and feedback loops have contributed to past stock market booms. Notably, he argued that
the dramatic rise in stock prices, technology stocks in particular, in the late 1990s was driven
by fads an argument made only a few months before the significant stock market decline in
20002001.
24

A number of studies following Shillers book documented several anomalies in the pricing of
technology stocks during the recent stock market boom. For example, Mitchell, Pulvino and
Stafford (2002) and Lamont and Thaler (2003) used partial spin-offs (or carve-outs) of tech
companies to show that the valuation of these spinoff stocks was irrationally high. In
particular, comparing the value of the spin-off with the value of the parent company, which
still held a partial stake in the spun-off company, the valuation of the spin-off implied that the
parents remaining assets had negative value.
25
They also reported evidence that short-sales
constraints, together with the fact that mispricings often increased before they eventually
23
Closed-end funds are typically owned by individual investors and are similar to mutual funds, with the exception
that investors cannot redeem their fund shares for cash but have to sell their shares in the market to get their money
back.
24
In the second edition of his book, published in 2005, Shiller extended his analysis to real estate, arguing that the
real estate market was similarly irrationally overvalued, and he predicted large problems for financial institutions with
the eventual burst of the real estate market bubble.
25
Also, Ofek and Richardson (2003) linked the high prices of internet stocks to short-sale constraints and lock-ups
preventing insiders from selling their shares after an IPO.

35

disappeared, made it difficult and risky for an arbitrageur to profit from these mispricings.
Along the same lines, Brunnermeier and Nagel (2004) showed that sophisticated investors,
such as hedge funds, preferred to acquire tech stocks and ride the bubble in the late 1990s
rather than shorting these stocks. More recently, Xiong and Yu (2011) documented a bubble
in the pricing of Chinese warrants in the late 2000s, which they showed traded far above their
fundamental value, and they argued that short-sales constraints again prevented the mispricing
from being arbitraged away.
26

In recent years, Shiller has continued to explore the impact of psychological factors on
financial markets in popular books such as Shiller (2008) and Akerlof and Shiller (2010).

7. What determines differences in expected returns across assets?
The research reviewed so far primarily has been focused on the time-series pattern of asset
prices. A related question concerns the cross-sectional pattern of prices, in particular for
stocks. Why is a particular stock more highly valued than another one at the same point in
time? According to equation (1), the answer depends on expected future cash flows,
,+
,
and the discount rates (return requirements),
+
. The discount rates should reflect time
preferences in the economy as well as risk premia. Since investors are generally risk-averse,
they should demand a higher expected return for more risky assets. A central insight, dating
back to the portfolio model of Markowitz (1959), is that investors should only demand
compensation for systematic risk, i.e., risk that cannot be eliminated by holding a well-
diversified portfolio. But which systematic risks drive stock returns, and to what extent are
investors compensated for them in terms of higher expected returns? Alternatively, to the
extent that fads and investor irrationality affect stock prices, as Shiller (1984) suggested, how
would this affect differences in expected returns across stocks?

7.1 Early tests of the Capital Asset Pricing Model (CAPM)
The CAPM was developed by Sharpe (1964), Lintner (1965) and Mossin (1966). William
Sharpe was awarded the 1991 Prize for his contribution to developing the CAPM, which still
26
A body of literature also focuses on the asset-pricing bubbles in an experimental setting. See, for example, Smith,
Suchanek and Wiliams (1988).

36

remains a fundamental asset-pricing model taught to students. According to the static CAPM,
the expected return
of a given financial asset i is given by

),
where
is the risk-free rate,
is the expected return on the market portfolio (i.e., a

portfolio of all assets in the economy), and
is the key measure of systematic risk which

should be compensated by a higher rate of return equal to the covariance of asset i with the
market portfolio (the beta of the stock). In the mid-1960s, this model provided a promising
explanation of asset prices, but it had not yet been empirically tested in a rigorous way.
How good is the CAPM at explaining the cross-section of asset prices? After its development
in the mid-1960s, economists set out to test the model empirically. These tests started from
time-series regressions of stock returns on index returns to generate estimates of stock-
specific beta coefficients,
. Assuming that market expectations are rational so that

observed returns
,
are equal to expected returns plus a random error
,
CAPM can be
tested based on this equation:

,
=
0,
+
1,

+
,
.
If CAPM is correct, then
0
=
, the risk-free rate, and

1,
=
, the expected
return on the market in excess of the risk-free rate. Early tests by Douglas (1969), Black and
Scholes (1973), and Black, J ensen and Scholes (1972) used time-series stock data to estimate
in a first step, and then cross-sectional data to regress the return on beta in a second step.
Results typically yielded a positive relation in accordance with theory, but the estimated
coefficient implied an implausibly high value for the riskless rate of return. In addition, these
studies did not account for the strong cross-sectional correlation in stock returns that is caused
by common shocks that affect groups of stocks at the same time. Not accounting for such
correlation leads to biased inference. Typically, the estimated standard errors are downward
biased.
27

Fama and MacBeth (1973) suggested an alternative approach to test the CAPM, and their
approach has become a standard method for testing cross-sectional asset-pricing models.
Their simple but powerful insight was that lack of predictability, with constant expected
returns over time, implies that stock returns are uncorrelated over time, even though they are
27
For example, Black, Jensen and Scholes (1972) noticed that their CAPM tests gave unreasonably high t values.

37

correlated across stocks at a given time. Based on this insight, Fama and MacBeth (1973)
presented a two-step approach for dealing with the problem of cross-sectional correlation.
The first step estimates a sequence of cross-sectional regressions say, month by month of
stock returns on the characteristics that should determine expected returns according to the
asset-pricing model. In the case of the CAPM, each cross-sectional test regresses stock returns
on an estimated beta (which in turn had been estimated using data from, say, the previous five
years). The second step calculates the time-series average of the coefficients from the cross-
sectional regressions, and tests whether these averages deviate significantly from the expected
values according to theory. The coefficients from each cross-sectional regression can be
interpreted as returns on portfolios weighted by these characteristics (an interpretation
developed by Fama, 1976, chapter 9). These returns should be serially uncorrelated under the
joint hypothesis that forecast errors are unpredictable and that the first-stage regressors
include all relevant determinants of expected returns. The correct standard error for the test
can be calculated from the time-series variability of the coefficients from the cross-sectional
regressions.
Using their methodology, Fama and MacBeth found that CAPM betas seemed to explain
differences in expected returns across stocks. They found, however, that the intercept (
0,
) in
the regression was larger than the risk-free rate, which is inconsistent with the Sharpe-
Lintner-Mossin CAPM, but possibly consistent with the zero-beta version of the CAPM
due to Black (1972).
The Fama-MacBeth two-step approach quickly became widely used in empirical asset-pricing
research, due to both its simplicity in implementation and its robustness. Even though CAPM
was refuted in later tests, as discussed below, the Fama-MacBeth procedure is still a standard
method for testing multi-factor cross-sectional asset-pricing models, and has been used in
thousands of applications.
28

28
An econometric issue not explicitly addressed in the original Fama-MacBeth study is that the betas used in the
second step suffer from estimation error from the first step. Shanken (1992) and Jagannathan and Wang (1998)
accounted for this and derived correct asymptotic standard errors.

38

7.2 CAPM anomalies
Although the early CAPM tests seemed promising, the empirical support for the model was
increasingly questioned towards the end of the 1970s.
29

First, in an influential paper, Roll (1977) criticized tests of the CAPM, showing that any valid
CAPM test presupposed complete knowledge of the market portfolio. According to the
CAPM theory, the market portfolio contains every individual asset in the economy, including
human capital, and is therefore inherently unobservable. Using a stock market index as a
proxy for the market portfolio, which previous tests had done, would therefore lead to biased
and misleading results.
30

Second, numerous studies tested for the determinants of cross-sectional differences in returns,
using the methodologies developed in the earlier tests. These tests led to the discovery of a
number of CAPM anomalies, where stock-specific characteristics seemed related to
differences in returns. A consistent finding was that various versions of (the inverse of)
scaled stock price, such as the earnings/price (E/P) ratio (Basu, 1977, 1983), the book-to-
market ratio (i.e., the book equity value divided by the market equity value; Statman, 1980,
Rosenberg, Reid and Landstein, 1985), and the debt/equity ratio (Bhandari, 1988), were
positively related to expected returns, even after the CAPM beta had been controlled for.
Furthermore, DeBondt and Thaler (1985) showed that stocks that had overperformed over
longer horizons, such as the last three to five years, tended to underperform over subsequent
years (and vice versa). Finally, stocks of firms with a smaller market value of equity were
shown to have higher expected returns (Banz, 1981) than stocks of larger firms, the so-called
size effect.
31
To make matters worse for the CAPM, several studies indicated that CAPM
beta did not seem very successful in explaining returns as the sample period of the earlier tests
was extended (Reinganum, 1981, and Lakonishok and Shapiro, 1986) or when controlling for
other macroeconomic factors (Chen, Roll and Ross, 1986).
32

29
Fama (1991) provided an extensive review of this research.
30
Later, Gibbons, Ross and Shanken (1989) developed a test of the CAPM that addressed Rolls critique. Using the
insight that the CAPM implies that the market portfolio is mean-variance efficient, they derived a test of the ex-ante
mean-variance efficiency of a given portfolio.
31
A number of papers also documented seasonality in stock returns, most prominently the so-called January
effect, indicating that stocks outperform in the month of January (Rozeff and Kinney, 1976). Keim (1983) showed
that the January effect is essentially also a small-firm or size effect.
32
As an alternative to the standard CAPM, the CCAPM predicts that a stocks return covariation with aggregate
consumption, its consumption beta, should be related to expected return differences across assets. The CCAPM
tests did not fare better in explaining anomalies, however, and the consumption beta was even shown to be
dominated by the standard CAPM beta (Mankiw and Shapiro, 1986, Breeden, Gibbons and Litzenberger, 1989).

39

Most of these results were integrated in the widely cited paper by Fama and French (1992),
which convincingly established that the CAPM beta has practically no additional explanatory
power once book-to-market and size have been accounted for.
33

7.3 The Fama-French three-factor model
The body of work discussed above was synthesized into the three-factor model of Fama and
French (1993). Building on the rejection of the simple version of CAPM in their earlier paper
(Fama and French, 1992), the paper presented a model which added two new factors to
CAPM and suggested a methodology for constructing and testing such factors, building on
Fama and Macbeth (1973). The two factors, small-minus-big market value (SMB) and
high-minus-low book-to-market ratio (HML), are based on portfolios of stocks sorted
according to the two characteristics that had been found to correlate with expected returns,
size and book-to-market value. Each factor is equivalent to a zero-cost arbitrage portfolio that
takes a long position in high book-to-market (small-size) stocks and finances this with a short
position in low book-to-market (large-size) stocks. Fama and French showed that the SMB
and HML factors, apart from explaining differences in expected returns across stocks, also
explain a significant amount of variation in the time-series, i.e., stocks with a similar exposure
to these factors move together. Hence, they argued, SMB and HML are priced risk factors and
the three-factor model should be interpreted as a multi-factor model in the sense of Merton
(1973) and Ross (1976).
34

Over the following years, Fama and French extended this work in a number of papers. For
example, in a study published in 1996, they found that the three-factor model captures the
return differences from other anomalies, including E/P, leverage and the return reversal of
DeBondt and Thaler (1985). They also showed that HML in particular has similar explanatory
power for international stock returns (Fama and French, 1998) and is present in U.S. data
earlier than that used for their original study (Davis, Fama and French, 2000).

33
The fact that the beta on the stock market does not relate to expected returns in the cross-section is particularly
striking because the stock market factor explains substantial variation in the time series for individual stocks. Partly
because of this time series significance, the stock market beta is typically included in multi-factor models, despite its
weak explanatory power in the cross-section.
34
In addition, Fama and French (1993) showed that two other factors, related to maturity and default risk, capture
much of the time-series and cross-sectional variation in bond returns, and that stock and bond market returns are
linked through shared variation in the stock and bond factors.

40

7.4 Rational and behavioral explanations for the cross section of stock returns
Empirically, the Fama-French approach has provided an effective way to simplify and unify
the vast literature on the cross section of stock returns, and their method has been widely used
both as a reference model for academic research and as a practical guide for professional
investors.
35
A weakness of the three-factor model is that it is primarily an empirical model
that describes stock returns, but it is silent on the underlying economic reasons for why these
risk factors have nonzero prices.
As shown by Ball (1978), characteristics such as book-to-market and size are essentially the
inverse of a scaled stock price, and can thus be thought of as proxies for the stochastic
discount factor for the stock, as evident from the simple present-value relationship.
Consequently, scaled price variables should be related to expected returns when added to any
misspecified asset-pricing model.
36

In their original work, Fama and French developed a rational, multi-factor interpretation of
their results, arguing that HML and SMB are capturing fundamental risk factors for which
investors demand compensation. In support of this argument, Fama and French (1995)
showed that high book-to-market predicts lower earnings, and argued that the excess return to
HML therefore should be interpreted as compensation for distress risk.
In contrast, other researchers interpreted the significance of HML and SMB as capturing the
effects of market mispricing and investor irrationality along the lines of Shiller (1984).
Lakonishok, Shleifer and Vishny (1994) argued that excess return to high book-to-market
stocks, or value stocks, is due to the fact that they are underpriced by investors, while low
book-to-market stocks are overpriced glamor stocks that subsequently underperform the
market.
Although the size and book-to-market effects could be consistent with models of investor
mispricing and psychological biases, recent research has found considerable co-movement
among stocks with similar book-to-market ratios, i.e., value (low book-to-market) versus
growth (high book-to-market) stocks (see, e.g., Campbell and Vuolteenaho, 2004), and that
value stocks co-move with value strategies in other asset classes, such as fixed income and
currencies (see Asness et al., 2013), which is consistent with attributing the higher excess
35
Later, Carhart (1997) suggested adding momentum as a fourth factor, which is now commonly added to the Fama-
French benchmark model.
36
Also see Berk (1995).

41

return to value stocks to a common risk factor.
37
Campbell and Vuolteenaho (2004),
Campbell. Polk, and Vuolteenaho (2009), and Campbell et al. (2012) argue that the book-to-
market effect can be explained by an intertemporal CAPM model (in the sense of Merton,
1973) in which investors care more about permanent cash-flow-driven movements than about
temporary discount-rate-driven movements in the aggregate stock market. In their model, the
required return on a stock is determined not by its overall beta with the market, but by two
separate betas, one with permanent cash-flow shocks to the market (to which high book-to-
market value stocks are more sensitive), and the other with temporary shocks to market
discount rates (to which low book-to-market growth stocks are more sensitive). Recently,
Campbell et al. (2012) found that the same argument can explain a large part of cross-
sectional returns for other assets as well, such as equity index options and corporate bonds.
A more serious challenge to informational efficiency is J egadeesh and Titmans (1993)
discovery of momentum in stock prices.
38
They found that an investment strategy that buys
stocks that have performed well and sells stocks that have performed poorly over the past 3-
to 12-month period generates significant excess returns over the following year. The fact that
winner stocks keep on winning and losers keep losing is consistent with a story where
relevant information only gradually disseminates into prices, and this pattern seems unlikely
to be explained by changes in risk, given the relatively short horizon. Moreover, unlike many
of the other anomalies, momentum is not captured by the Fama-French three-factor model.
Based on these findings, a number of behavioral-finance papers have built theories based on
investor psychology to explain both the book-to-market and momentum effects, e.g., based on
investor underreaction to news in the short-run (leading to momentum) and overreaction in
the longer run (leading to reversals, or book-to-market effects). Examples have been
presented by Daniel, Hirshleifer and Subrahmanyam (1998), Barberis, Shleifer and Vishny
(1998), and Hong and Stein (1999). Moreover, consistent with limits to arbitrage, momentum
has been shown to be particularly pronounced among smaller, more illiquid stocks, and
momentum strategies entail substantial risk due to the high correlation within industries
(Moskowitz and Grinblatt, 1999).
Another strand of the literature has retained (or remained agnostic to) the standard assumption
of rational investors with standard preferences, but instead introduced financial market
37
Although Campbell and Vuolteenaho (2004) also found similar co-movement with respect to size, the excess
returns to small stocks seem to be much weaker, or even have largely disappeared in recent data.
38
In the words of Eugene Fama: Of all the potential embarrassments to market efficiency, momentum is the
primary one.(Fama and Litterman, 2012).

42

frictions to explain asset-pricing patterns. One group of papers has introduced market
segmentation, which implies limited risk-sharing among investors. This in turn leads to
downward-sloping demand curves for assets in the short term, and mean reversion due to
slow-moving capital across markets (Duffie, 2010). Another group of papers has focused on
frictions due to financial intermediaries that are restricted by regulation and/or agency
problems in their trading of financial assets. These frictions can lead to fire sales of assets
when the capital or liquidity of intermediaries becomes scarce (Brunnermeier and Pedersen,
2009, and Coval and Stafford, 2007). Finally, a number of studies have focused on liquidity
and its impact on asset pricing. In these models, investors demand an additional risk premium
for holding illiquid assets that cannot be easily sold when investors need liquidity, e.g., for
consumption (Amihud and Mendelson, 1986, Pastor and Stambaugh, 2003, and Acharya and
Pedersen, 2005). Models based on financial frictions and liquidity have been shown to have
explanatory power during the recent financial crisis (Brunnermeier, 2009).

8. Influences on market practice
Asset pricing is one of the fields in economics where academic research has had the most
impact on non-academic practice. Even though there is still no broad consensus regarding the
interpretation of some results, the research initiated by Fama, Shiller and Hansen has
produced a body of robust empirical findings, which have important practical implications:
1. In the short term, predictability in stock returns is very limited, which is consistent with stock
prices quickly reflecting new public information about future cash flows. To the extent that short-
term return predictability can be found, it is too small to profit from because of transaction costs.
2. In the longer term, there is economically significant predictability in stock returns, indicative of
variations in expected returns or discount rates. In particular, expected returns in good times (at
the peak of the business cycle, when measures of relative valuation such as price/dividend ratios
are high) are lower than expected returns in bad times.
3. In the cross-section of stocks, a number of factors such as book-to-market predict differences in
expected returns. Stocks with a similar exposure to these factors co-move, implying that the higher
returns come with higher risk.

43

The early findings on the lack of short-term predictability in stock prices had considerable
practical impact. One implication is that it should be extremely hard for asset managers to
generate excess risk-adjusted returns. In one of the first studies on this issue, J ensen (1968)
evaluated mutual fund performance and found that the majority of funds did not generate any
excess risk-adjusted returns. Subsequent studies of mutual-fund performance generally have
failed to find positive excess performance (and often found negative excess performance)
after fund fees. Recently, Fama and French (2010) have documented that only the extreme
tails of active mutual funds generate significant (negative and positive, respectively) risk-
adjusted excess returns before fees, and that the aggregate portfolio of active mutual funds in
fact is close to the market portfolio. The latter means that the sector as a whole gives negative
excess returns to investors.
Inspired by the work of Fama, J ensen and others, so-called index funds started to emerge in
the early 1970s.
39
Today, passively managed funds, such as index funds and Exchange Traded
Funds (ETFs), exist for a large variety of indexes and asset classes, including size and book-
to-market. In 2012, these funds had over $3.6 trillion (U.S.) under management and accounted
for 41% of the worldwide flows into mutual funds.
The research on market predictability and on cross-sectional return differences across
financial assets has also had considerable practical impact and has contributed to the growth
of quantitative investment management, where investors use quantitative factors and
statistical modeling to make investment decisions. For example, many professional investors
use factor models such as the Fama-French model to guide their portfolio decisions, and long-
term institutional investors commonly use variables that have been shown to predict medium-
run stock market returns to adjust the fraction of equity relative to bonds in their portfolios.
The academic work on the determinants of cross-sectional returns has also had a large impact
on the practice of portfolio performance measurement. Given that a portfolio manager can get
a higher rate or return on her portfolio simply by investing in assets with higher risk, she
needs to be evaluated based on risk-adjusted returns. J ensen (1968) introduced a measure of
risk-adjusted performance, the so-called J ensens alpha, which is essentially the intercept of
a regression of excess returns on risk factors, such as the Fama-French three factors.
40

Intuitively, since an investor can achieve a high return simply by investing in assets with high
39
The worlds first index fund was started at the U.S. bank Wells Fargo in 1971, managed on behalf of Samsonite
Corporation (the Samsonite Luggage Fund). A few years later in 1975, Vanguard launched the first index fund
directed towards retail investors. See Bernstein (2005)
40
In Jensens original study, he measured alpha relative to the market model, i.e., controlling for the CAPM beta.

44

loadings on the Fama-French factors, a portfolio manager should only be rewarded on excess
performance relative to these factors, i.e., on alpha. Alpha has become a standard tool for
evaluating portfolio managers and mutual funds (used, e.g., by Morningstar). Moreover,
following the work of Fama and French, it has become standard to evaluate performance
relative to size and value benchmarks, rather than simply controlling for overall market
returns.
Research findings from empirical asset pricing have also had practical impact outside the
investment management industry. The event-study methodology of Fama, Fisher, J ensen and
Roll (1969) has become an important tool in legal practice for assessing damages in lawsuits,
for example in securities-fraud cases (Mitchell and Netter, 1994). Event studies have also
been used by competition authorities, to evaluate the competitive effect of mergers by looking
at the stock-price reaction of a merger on the other firms in the industry (e.g., Beverley,
2007).
Another area of practical impact is the measurement of asset returns and price indexes. The
CRSP data set created at the University of Chicago was the first comprehensive stock market
database in existence. It has had a profound impact not only on academic research but also on
quantitative investment strategies used by the industry.
Beyond stock prices, Case and Shiller (1987) constructed the first systematic, high-quality
indexes of U.S. house prices. The S&P Case-Shiller index is now the standard real estate price
index in the U.S., widely used by practitioners and policymakers. Shillers interest in index
construction was motivated by the insight that the volatility of house prices constitutes a
major risk for many households. In his 1991 book Macro Markets, Shiller highlighted the fact
that major risks in society, like house-price risks, are uninsurable despite their importance. He
argued that developing markets for derivative contracts based on price indexes would help
households to hedge against such risks. In particular, such contracts would allow households
to go short in the housing market. But it would also allow households to speculate against an
overvalued housing market. Shiller has also translated these insights into practice and has
helped setting up a market in cash-settled house-price futures at the Chicago Mercantile
Exchange based on the S&P Case-Shiller indexes.

45

9. Conclusions
Eugene Fama, Lars Peter Hansen, and Robert Shiller have developed empirical methods and
used these methods to reach important and lasting insights about the determination of asset
prices. Their methods have shaped subsequent research in the field and their findings have
been highly influential both academically and practically. The waves of research following
the original contributions of the Laureates constitute a landmark example of highly fruitful
interplay between theoretical and empirical work.
We now know that asset prices are very hard to predict over short time horizons, but that they
follow movements over longer horizons that, on average, can be forecasted. We also know
more about the determinants of the cross-section of returns on different assets. New factors
in particular the book-to-market value and the price-earnings ratio have been demonstrated
to add significantly to the prior understanding of returns based on the standard CAPM.
Building on these findings, subsequent research has further investigated how asset prices are
fundamentally determined by risk and attitudes toward risk, as well as behavioral factors.

46

REFERENCES

Abel, A.B. (1990), Asset pricing under habit formation and catching up with the J oneses,
American Economic Review 80, 38-42.
Abreu, D. and M. Brunnermeier (2002), Synchronization risk and delayed arbitrage,
Journal of Financial Economics 66(2-3), 341-360.
Abreu, D. and M. Brunnermeier (2003), Bubbles and crashes, Econometrica 71(1), 173-
204.
Acharya, V. and L.H. Pedersen (2005), Asset pricing with liquidity risk, Journal of
Financial Economics 77, 375-410.
Akerlof, G. and R. Shiller (2010), Animal Spirits: How Human Psychology Drives the
Economy, and Why It Matters for Global Capitalism. Princeton University Press.
Alexander, S.S. (1961), Price movements in speculative markets: Trends or random walks,
Industrial Management Review 2, 7-26.
Alexander, S.S. (1964), Price movements in speculative markets: Trends or random walks.
No. 2. In: Cootner, Paul (ed.), The random character of stock prices. Cambridge:
M.I.T.
Amihud, Y. and H. Mendelson (1986), Asset pricing and the bid-ask spread, Journal of
Arellano, M., and S. Bond (1991), Some tests of specification for panel data: Monte Carlo
evidence and an application to employment equations, Review of Economic Studies
58, 277-298.
Asness, C.S., T.J . Moskowitz, and L.H. Pedersen (2013), Value and momentum
everywhere, Journal of Finance 68, 929-985.
Asquith P. and D.W. Mullins (1986), Signalling with dividends, stock repurchases, and
equity issues, Financial Management Autumn, 27-44.
Bachelier, L. (1900), Thorie de la Speculation, Paris.
Baker, M. and J . Wurgler (2007), Investor sentiment in the stock market, Journal of
Economic Perspectives 21(2), 129-151.
Ball, R. (1978), Anomalies in relationships between securities yields and yield-surrogates,
Journal of Financial Economics 6, 103-126.
Ball, R. and P. Brown (1968), An Empirical Evaluation of Accounting Income Numbers,
Journal of Accounting Research 6, 159-178.
Bansal, R. and A. Yaron (2004), Risks for the long run: A potential resolution of asset
pricing puzzles, Journal of Finance 59(4), 1481-1509.
Banz, R.W. (1981), The relationship between return and market value of common stocks,
Barberis, N., M. Huang, and J . Santos (2001), Prospect theory and asset prices, Quarterly
Journal of Economics 116, 1-53.
Barberis, N., A. Shleifer, and R. Vishny (1998), A model of investor sentiment, Journal of
47

Barberis, N. and R. Thaler (2003), A survey of behavioral finance. In: Constantinides,
G.M., M. Harris, and R.M. Stulz (ed.), Handbook of the Economics of Finance, vol. 1,
ch. 18, 1053-1128, Elsevier.
Basu, S. (1977), Investment performance of common stocks in relation to their price-
earnings ratios: A test of the efficient markets hypothesis, Journal of Finance 32,
663-682.
Basu, S. (1983), The relationship between earnings yield, market value, and return for NYSE
common stocks: Further evidence, Journal of Financial Economics 12, 129-156.
Benartzi, S. and R.H. Thaler (1995), Myopic loss aversion and the equity premium puzzle,
Quarterly Journal of Economics 110(1), 73-92.
Berk, J .B. (1995), A critique of size-related anomalies, Review of Financial Studies 8 (2),
275-286.
Bernstein, P.L. (2005), Capital ideas: The improbable origins of modern Wall Street. J ohn
Wiley and Sons.
Beverley, L. (2007), Stock market event studies and competition commission inquiries,
Mimeo, U.K. Competition Commission.
Bhandari, L.C. (1988), Debt/Equity ratio and expected common stock returns: empirical
evidence, Journal of Finance 43, 507-528.
Black, F. (1972), Capital market equilibrium with restricted borrowing, Journal of Business
45, 444-455.
Black F., M.C. J ensen, and M. Scholes (1972), The capital asset pricing model: Some
empirical tests. In: M. J ensen, ed., Studies in the theory of capital markets. Praeger.
Black, F. and M. Scholes (1973), The pricing of options and corporate liabilities, Journal of
Political Economy 81, 637-654.
Blundell, R.W. and S. Bond (1998), Initial conditions and moment restrictions in dynamic
panel data models, Journal of Econometrics 87, 115-143.
Breeden, D.T. (1979), An intertemporal asset pricing model with stochastic consumption and
investment opportunities, Journal of Financial Economics 7, 265-296.
Breeden, D.T., M.R. Gibbons, and R.H. Litzenberger (1989), Empirical tests of the
consumption-oriented CAPM, Journal of Finance 44, 231-262.
Brunnermeier, M. (2009), Deciphering the liquidity and credit crunch of 2007-2008,
Journal of Economic Perspectives 23(1), 77-100.
Brunnermeier, M. and S. Nagel (2004), Hedge funds and the technology bubble, Journal of
Finance 59, 2013-2040.
Brunnermeier, M., and L.H. Pedersen (2009), Market liquidity and funding liquidity,
Review of Financial Studies 22(6), 2201-2238.
Cagetti, M., L.P. Hansen, T.J . Sargent, and N. Williams (2002), Robustness and pricing with
uncertain growth, Review of Financial Studies 15(2), 363-404.
Campbell, J .Y. (1987), Stock returns and the term structure, Journal of Financial
Economics 18(2), 373-399.
Campbell, J .Y. (1993), Intertemporal asset pricing without consumption data, American
Economic Review 83(3), 487-512.
48

Campbell, J .Y and J .H. Cochrane (1999), By force of habit: a consumption-based
explanation of aggregate stock market behavior,, Journal of Political Economy 107,
205-251.
Campbell, J .Y., C. Polk, and T. Vuolteenaho (2009), Growth or glamour: Fundamentals and
systematic risk in stock returns, Review of Financial Studies 23(1), 305-344.
Campbell, J .Y., S. Giglio, C. Polk, and R. Turley (2012), An intertemporal CAPM with
stochastic volatility, NBER Working Paper 18411.
Campbell, J .Y. and R.J . Shiller (1987), Cointegration and tests of present value models,
Journal of Political Economy 95, 1062-1088.
Campbell, J .Y. and R.J . Shiller (1988a), The dividend-price ratio and expectations of future
dividends and discount factors, Review of Financial Studies 1, 195-227
Campbell, J .Y. and R.J . Shiller (1988b), Stock prices, earnings, and expected dividends,
Journal of Finance 43, 661-676.
Campbell, J .Y. and R.J . Shiller (1991), Yield spreads and interest rate movements: A bird's
eye view, Review of Economic Studies 58, 495-514.
Campbell, J .Y, and T. Vuolteenaho (2004), Bad beta, good beta, American Economic
Review 94(5), 1249-1275.
Carhart, M.M. (1997), On persistence of mutual fund performance, Journal of Finance
52(1), 57-82.
Case, K.E. and R.J . Shiller (1987), Prices of single family homes since 1970: New indexes
for four cities, New England Economic Review September/October, 45-56.
Case, K.E. and R.J . Shiller (1989), The efficiency of the market for single-family homes,
American Economic Review 79(1), 125-137.
Case, K.E. and R.J . Shiller (2003), Is there a bubble in the housing market? Brookings
Papers on Economic Activity.
Chen, H., R. Roll, and S.A. Ross (1986), Economic forces and the stock market, Journal of
Business 56, 383-403.
Cochrane, J .H. (2001), Asset Pricing, Princeton University Press, 2001.
Cochrane, J .H. (2007), The dog that did not bark: a defense of return predictability, Review
of Financial Studies 21, 1533-1575.
Cochrane, J .H. (2011), Presidential address: Discount rates, Journal of Finance 66(4),
1047-1108.
Conrad, J . and G. Kaul (1988), An anatomy of trading strategies, Review of Financial
Studies 11, 489-519.
Constantinides, G.M. (1990), Habit formation: A resolution of the equity premium puzzle,
Constantinides, G.M., and D. Duffie (1996), Asset pricing with heterogeneous consumers,
Cootner, P. (1962), Stock prices: Random vs. systematic changes, Industrial Management
Review 3, 24-45.
Cootner, P. (1964). The Random Character of Stock Prices. Cambridge: M.I.T.
49

Coval, J . and E. Stafford (2007), Asset fire sales (and purchases) in equity markets, Journal
of Financial Economics 86, 479-512.
Cowles, A. (1933), Can stock market forecasters forecast? Econometrica 1, 309-324.
Daniel, K., D. Hirshleifer, and A. Subrahmanyam (1998), Investor psychology and security
market under- and over-reactions, Journal of Finance 53(6), 1839-1884.
Dann, L.V. (1981), Common stock repurchases: An analysis of returns to bondholders and
stockholders, Journal of Financial Economics 9, 113-138.
Davis, J .L, E.F. Fama, and K.R. French (2000), Characteristics, covariances, and expected
returns: 1928-1997, Journal of Finance 55, 389-406.
Deaton, A. (1992), Understanding Consumption, Princeton University Press.
DeBondt, W.F.M. and R. Thaler (1985), Does the stock market overreact? Journal of
Finance 40, 793-805.
DeLong, B., A. Shleifer, L. Summers, and R. Waldmann (1990a), Noise trader risk in
financial markets, Journal of Political Economy 98, 703-738.
DeLong, B., A. Shleifer, L. Summers, and R. Waldmann (1990b), Positive feedback
investment strategies and destabilizing rational speculation, Journal of Finance 45,
375-395.
Dolley, J .C. (1933), Characteristics and procedure of common stock split-ups, Harvard
Business Review 11, 316-326.
Douglas, G.W. (1969), Risk in equity markets: an empirical appraisal of market efficiency,
Yale Economic Essays
Duffie, D. (2010), Asset price dynamics with slow-moving capital, Journal of Finance 65,
1237-1267.
Eichenbaum, M.S. and L.P. Hansen (1990), Estimating models with intertemporal
substitution using aggregate time-series data, Journal of Business and Economic
Statistics 8(1), 53-69.
Eichenbaum, M.S., L.P. Hansen, and K.J . Singleton (1988), A time-series analysis of
representative agent models of consumption and leisure choice under uncertainty,
Quarterly Journal of Economics 103(1), 51-78.
Engle, R.F. and C.W.J . Granger (1987), Co-integration and error correction: Representation,
estimation, and testing, Econometrica 55, 251-276.
Epstein, L.G. and M. Schneider (2008), Ambiguity, information quality, and asset pricing,
Journal of Finance 63, 197228.
Epstein, L.G. and T. Wang (1994), Intertemporal asset pricing under Knightian uncertainty,
Econometrica 62, 283-322.
Epstein, L.G., and S.E. Zin (1989), Substitution, risk aversion, and the temporal behavior of
consumption and asset returns: A theoretical framework, Econometrica 57(4), 937-
969.
Fama, E.F. (1963), Mandelbrot and the stable Paretian hypothesis, Journal of Business
36(4), 420-429.
Fama, E.F. (1965), The behavior of stock market prices, Journal of Business 38, 34-105.
50

Fama, E.F. (1970), Efficient capital markets: a review of theory and empirical work,
Fama, E.F. (1976), Foundations of Finance: Portfolio Decisions and Securities Prices, Basic
Books, New York.
Fama, E.F. (1984), Forward and spot exchange rates, Journal of Monetary Economics 14,
319-338.
Fama, E.F. (1991), Efficient capital markets II, Journal of Finance 46, 1575-1618.
Fama, E.F. (1998), Market efficiency, long-term returns, and behavioral finance, Journal of
Fama, E.F. and R.R. Bliss (1987), The information in long-maturity forward rates,
Fama, E.F. and M.E. Blume (1966), Filter rules and stock market trading, Journal of
Business 39:II, 226-241.
Fama, E.F., L. Fisher, M. J ensen and R. Roll (1969), The adjustment of stock prices to new
information, International Economic Review 10, 1-21.
Fama, E.F. and K.R. French (1988a), Dividend yields and expected stock returns, Journal
of Financial Economics 22, 3-26.
Fama, E.F. and K.R. French (1988b), Permanent and temporary components of stock prices,
Fama, E.F. and K.R. French (1989), Business conditions and expected returns on stocks and
bonds, Journal of Financial Economics 25, 23-50.
Fama, E.F. and K.R. French (1992), The cross-section of expected stock returns, Journal of
Finance 47, 427-466.
Fama, E.F. and K.R. French (1993), Common risk factors in the returns on stocks and
bonds, Journal of Financial Economics 33, 3-56.
Fama, E.F. and K.R. French (1995), Size and book-to-market factors in earnings and
returns, Journal of Finance 50, 131-156.
Fama, E.F. and K.R. French (1996), Multifactor explanations for asset pricing anomalies,
Journal of Finance 51(1), 55-84.
Fama, E.F. and K.R. French (1998), Value versus growth: The international evidence,
Fama, E.F. and K.R. French (2010), Luck versus skill in the cross-section of mutual fund
returns, Journal of Finance 65(5), 1915-1947.
Fama, E.F., and R. Litterman (2012), An experienced view on markets and investing,
Financial Analysts Journal 68(6), 1-5.
Fama, E.F. and J .D. MacBeth (1973), Risk, return and equilibrium: empirical tests, Journal
of Political Economy 81, 607-636.
Fama, E.F. and G.W. Schwert (1977), Asset returns and inflation, Journal of Financial
Economics 5, 115-146.
French, K.R. and R. Roll (1986), Stock return variances: The arrival of information and the
reaction of traders, Journal of Financial Economics 17, 5-26.
51

Friedman, M. (1953), The case for flexible exchange rates. In: Essays in Positive
Economics, Chicago University Press.
Froot, K., and E. Dabora (1999), How are stock prices affected by the location of trade?
Gibbons, M.R., S.A. Ross, and J . Shanken (1989), A test of the efficiency of a given
portfolio, Econometrica 57(5), 1121-1152.
Grossman, S.J . and R.J . Shiller (1981), The determinants of the variability of stock market
prices, American Economic Review 71, 222-227.
Grossman, S.J . and J .E. Stiglitz (1980), On the impossibility of informationally efficient
markets, American Economic Review 70, 393-408.
Hansen, B.E. and K.D. West (2002), Generalized method of moments and
macroeconomics, Journal of Business and Economic Statistics 20(4), 460-469.
Hansen, L.P. (1982), Large sample properties of generalized method of moments
estimators, Econometrica 50, 1029-1054.
Hansen, L.P. (1985), A method for calculating bounds on the asymptotic covariance matrices
of generalized method of moments estimators, Journal of Econometrics 30, 203-238.
Hansen, L.P., J .C. Heaton, and N. Li (2008), Consumption strikes back? Measuring long-run
risk, Journal of Political Economy 116, 260-302.
Hansen, L.P., J .C. Heaton, and M. Ogaki (1988), Efficiency bounds implied by multiperiod
conditional moment restrictions, Journal of the American Statistical Association 83
Hansen, L.P., J .C. Heaton, and A. Yaron (1996), Finite-sample properties of some alternative
GMM estimators, Journal of Business & Economic Statistics 14.
Hansen, L.P. and J .A. Scheinkman (1995), Back to the future: generating moment
implications for continuous-time Markov processes, Econometrica 63, 767-804
Hansen, L.P. and R.J . Hodrick (1980), Forward exchange rates as optimal predictors of
future spot rates: an econometric analysis, Journal of Political Economy 88(5), 829-
853.
Hansen, L.P. and R. J agannathan (1991), Implications of security market data for models of
dynamic economies, Journal of Political Economy 99, 225-262.
Hansen, L.P. and R. J agannathan (1997), Assessing specification errors in stochastic
discount factor models, Journal of Finance 52(2), 557-590.
Hansen, L.P., and S.F. Richard (1987), The role of conditioning information in deducing
testable restrictions implied by dynamic asset pricing models, Econometrica 55, 587-
613.
Hansen, L.P., and T.J . Sargent (1980), Formulating and estimating dynamic linear rational
expectation models, Journal of Economic Dynamics and Control 2, 7-46.
Hansen, L.P. and T.J . Sargent (2001), Robust control and model uncertainty, American
Economic Review 91(2), 60-66.
Hansen, L.P. and T.J . Sargent (2008), Robustness, Princeton Universty Press.
Hansen, L.P. and K.J . Singleton (1982), Generalized instrumental variable estimation of
nonlinear rational expectations models, Econometrica 50(5), 1269-1286.
52

Hansen, L.P. and K.J . Singleton (1983), Stochastic consumption, risk aversion, and the
temporal behavior of asset prices, Journal of Political Economy 91(2), 249-265.
Hansen, L.P. and K.J . Singleton (1984) Erratum: Generalized instrumental variable
estimation of nonlinear rational expectations models, Econometrica 52(1), 267-268.
Harrison, J .M. and D.M. Kreps (1978), Speculative investor behavior in a stock market with
heterogeneous expectations, Quarterly Journal of Economics 92(2), 323-336.
Harrison, J .M. and D.M. Kreps (1979), Martingales and arbitrage in multiperiod securities
markets, Journal of Economic Theory 20, 381-408.
Heaton, J . and D. Lucas, (1992), The effects of incomplete insurance markets and trading
costs in a consumption-based asset pricing model, Journal of Economic Dynamics
and Control 16, 601-620.
Hong, H. and J . Stein (1999), A unified theory of underreaction, momentum trading, and
overreaction in asset markets, Journal of Finance 54(6), 2143-2184.
Huggett, M. (1993), The risk-free rate in heterogeneous-agent incomplete-insurance
economies, Journal of Economic Dynamics and Control 17, 953-969.
J agannathan, R., G. Skoulakis, and Z. Wang (2002), Generalized method of moments:
Applications in finance, Journal of Business and Economic Statistics 20(4), 470-481.
J agannathan, R. and Z. Wang (1998), An asymptotic theory for estimating beta-pricing
models using cross-sectional regression, Journal of Finance 53(4), 1285-1309.
J arrow, R.A. and M. Larsson (2012), The meaning of market efficiency, Mathematical
Finance 22(1), 1-30.
J egadeesh, N. and S. Titman (1993), Returns to buying winners and selling losers:
Implications for stock market efficiency, Journal of Finance 48(1), 65-91.
J ensen, M.C. (1968), The performance of mutual funds in the period 1945-1964,Journal of
Finance 23(2), 389-416.
Kahneman, D. and A. Tversky (1979), Prospect theory: an analysis of decision under risk,
Kahneman, D. and A. Tversky (1984), Choices values and frames, American Psychologist
39, 341-350.
Keim, D.B. (1983), Size-related anomalies and stock return seasonality, Journal of
Keim, D.B. and R.F. Stambaugh (1986), Predicting returns in the stock and bond markets,
Kendall, M.G. (1953), The analysis of economic time series, Part I: Prices, Journal of the
Royal Statistical Society 96, 11-25.
Kleidon, A.W. (1986), Variance bounds tests and stock price valuation models, Journal of
Political Economy 94(5), 953-1001.
Kothari, S.P., and J .B. Warner (2007), Econometrics of event studies. In: Eckbo, B.E.
(ed.),Handbook of Corporate Finance: Empirical Corporate Finance, Volume 1,
Elsevier, p. 3-36.
Kreps, D.M and E.L. Porteus (1978), Temporal resolution of uncertainty and dynamic choice
theory, Econometrica 46, 185-200
53

Lakonishok, J . and A.C. Shapiro (1986), Systematic risk, total risk and size as determinants
of stock market returns, Journal of Banking and Finance 10, 115-132.
Lakonishok, J ., A. Shleifer, and R.W. Vishny (1994), Contrarian investment, extrapolation,
and risk, Journal of Finance 49(5), 1541-1578.
Lamont, O., and R. Thaler (2003), Can the market add and subtract? Mis-pricing in tech
stock carve-outs, Journal of Political Economy 111(2), 227-268.
Lee, C.M.C., A. Shleifer, and R.H. Thaler (1991), Investor sentiment and the closed-end
fund puzzle, Journal of Finance 46(1), 75-109.
Lehmann, B.N. (1990), Fads, martingales, and market efficiency, Quarterly Journal of
Economics 105(1), 1-28.
LeRoy, S.F. and R. Porter (1981), The present-value relation: tests based on implied variance
bounds, Econometrica 49, 55-74.
Lettau, M. and S. Ludvigson (2001), Resurrecting the (C)CAPM: A cross-sectional test
when risk-premia are time-varying, Journal of Political Economy 109(6), 1238-1287.
Lintner, J . (1956), Distribution of incomes of corporations among dividends, retained
earnings, and taxes, American Economic Review 46(2), 97-113.
Lintner, J . (1965), The valuation of risk assets and the selection of risky investments in stock
portfolios and capital budgets, Review of Economics and Statistics 47, 13-37.
Lo, A.W. and A.C. MacKinlay (1988), Stock market prices do not follow random walks:
evidence from a simple specification test, Review of Financial Studies 1, 41-66.
Lucas, R.E. (1978), Asset prices in an exchange economy, Econometrica 46, 1429-1446.
MacKinlay, A.C. (1997), Event studies in economics and finance, Journal of Economic
Literature 35(1), 13-39.
Malloy, C.J ., T.J . Moskowitz, and A. Vissing-J orgensen (2009), Long-run stockholder
consumption risk and asset returns, Journal of Finance 64(6), 2427-2479.
Mandelbrot, B. (1963), The variation of certain speculative prices, Journal of Business 36,
394-419.
Mankiw, N.G. (1986), The equity premium and the concentration of aggregate shocks,
Mankiw, N.G., and M.D. Shapiro (1986), Risk and return: Consumption beta versus market
beta, Review of Economics and Statistics 48, 452-459.
Markowitz, H. (1959), Portfolio Selection: Efficient Diversification of Investments, Yale
University Press.
Marsh, T.A. and R.C. Merton (1986), Dividend variability and variance bounds tests for the
rationality of stock prices, American Economic Review 76(3), 483-498.
Mehra, R. and E. Prescott (1985), The equity premium puzzle, Journal of Monetary
Merton, R.C. (1973), An intertemporal asset pricing model, Econometrica 41(5), 867-887.
Miller, E. (1977), Risk, uncertainty, and differences of opinion, Journal of Finance 32(4),
1151-1168.
54

Mitchell, M.L. and J .M. Netter (1994), The role of financial economics in securities fraud
cases: Applications at the Securities and Exchange Commission, Business Lawyer
February 1994.
Mitchell, M., T. Pulvino, and E. Stafford (2002), Limited arbitrage in equity arkets, Journal
of Finance 57(2), 551-584.
Modigliani, F. and R.A. Cohn (1979), Inflation, rational valuation, and the market,
Financial Analysts Journal 35(2), 24-44.
Moskowitz, T.J . and M. Grinblatt (1999), Do industries explain momentum? Journal of
Finance 54(4), 1249-1290.
Mossin, J . (1966), Equilibrium in a capital asset market, Econometrica 34(4), 768-783.
Nelson, C.R. and M.J . Kim (1993), Predictable Stock Returns: The Role of Small Sample
Bias. Journal of Finance 48, 641661.
Neyman, J . and E.S. Pearson (1928), On the use and interpretation of certain test criteria for
purposes of statistical inference, Part II, Biometrika 20A, 263-294.
Ofek, E. and M. Richardson (2003), DotCom mania: The rise and fall of internet stock
prices, Journal of Finance 58(3), 1113-1137.
Osborne, M.F.M. (1959), Brownian motion in the stock market, Operations Research 7,
145-173.
Oskamp, S. (1965), Overconfidence in case-study judgments, Journal of Consulting
Psychology 29, 261-265.
Pastor, L., and R.F. Stambaugh (2003), Liquidity risk and expected stock returns, Journal
Pearson, K. (1894), Contributions to the mathematical theory of evolution, Philosophical
Transactions of the Royal Society of London, A 185, 71-110
Pearson, K. (1900), On the criterion that a given system of deviations from the probable in
the case of a correlated system of variables is such that it can be reasonably supposed
to have arisen from random sampling, Philosophical Magazine 5
th
Series 50, 157-
175.
Reinganum (1981), Misspecification of capital asset pricing: Empirical anomalies based on
earnings yields and market values, Journal of Financial Economics 9, 19-46.
Roberts, H.V. (1959), Stock-market patterns and financial analysis. Methodological
suggestions Journal of Finance 14, 1-10.
Roll, R. (1977), A critique of the asset pricing theory's tests Part I: On past and potential
testability of the theory, Journal of Financial Economics 4(2), 129-176.
Roll, R. (1984), Orange juice and weather, American Economic Review 74, 861-880.
Rosenberg, B., K. Reid, and R. Lanstein (1985), Persuasive evidence of market
inefficiency, Journal of Portfolio Management 11, 9-17.
Ross, S.A. (1976), The arbitrage pricing theory of capital asset pricing, Journal of
Economic Theory 13, 341-360.
Ross, S.A. (1978), A simple approach to the valuation of risky streams, Journal of Business
51(3), 453-475.
55

Routledge, B.R. and S.E. Zin (2010), Generalized disappointment aversion and asset prices,
Rozeff, M.S. and W.R. Kinney (1976), Capital market seasonality: The case of stock
returns, Journal of Financial Economics 3, 379-402.
Samuelson, P.A. (1965), Proof that properly anticipated prices fluctuate randomly,
Industrial Management Review 6, 41-49.
Sargan, J .P. (1958), The estimation of economic relationships using instrumental variables,
Shanken, J . (1992), On the estimation of beta-pricing models, Review of Financial Studies
5(1), 1-33.
Sharpe, W.F. (1964), Capital asset prices: A theory of market equilibrium under conditions
of risk, Journal of Finance 19(3), 425-442.
Sharpe, W.F. (1966), Mutual fund performance, Journal of Business 39, 119-138.
Shiller, R.J . (1979), The volatility of long term interest rates and expectations models of the
term structure, Journal of Political Economy 87, 1190-1219.
Shiller, R.J . (1981a), Do stock prices move too much to be justified by subsequent changes
in dividends? American Economic Review 71, 421-436.
Shiller, R.J . (1981b), The use of volatility measures in assessing market efficiency, Journal
of Finance 36(2), 291-304.
Shiller, R.J . (1982), Consumption, asset markets and macroeconomic fluctuations, Carnegie
Rochester Conference Series on Public Policy 17, 203-238.
Shiller, R.J . (1984), Stock prices and social dynamics, Carnegie Rochester Conference
Series on Public Policy, 457-510.
Shiller, R.J . (1987), Investor behavior in the 1987 stock market crash: Survey evidence,
NBER working paper no. 2446.
Shiller, R.J . (1988), Portfolio insurance and other investor fashions as factors in the 1987
stock market crash, NBER Macroeconomics Annual 3, 287-297.
Shiller, R.J . (1989), Market Volatility, MIT Press.
Shiller, R.J . (2000), Irrational Exuberance, Princeton University Press.
Shiller, R.J . (2008), The subprime solution: How todays global financial crisis happened,
and what to do about it, Princeton University Press.
Shiller, R.J ., J .Y. Campbell, and K.L. Schoenholtz (1983), Forward rates and future policy:
interpreting the term structure of interest rates, Brookings Papers on Economic
Activity 1, 173-223.
Shiller, R.J . and P. Perron (1985), Testing the random walk hypothesis: Power vs. frequency
of observation, Economics Letters 18, 381-386.
Shiller, R.J . and J . Pound (1989), Survey evidence on the diffusion of interest and
information among investors, Journal of Economic Behavior and Organization 12,
47-66.
Shleifer, A. (1986), Do demand curves for stocks slope down, Journal of Finance 41(3),
579-590.
56

Shleifer, A. (2000), Inefficient Markets: An Introduction to Behavioral Finance. Clarendon
Lectures in Economics, Oxford University Press.
Shleifer, A., and R. Vishny (1997), The limits of arbitrage, Journal of Finance 52(1), 35-
55.
Slovic, P. (1972), Psychological study of human judgement: Implications for investment
decision making, Journal of Finance 27(4), 779-801.
Smith, V.L., G.L. Suchanek, and A.W. Williams (1988), Bubbles, crashes, and endogenous
expectations in experimental spot markets, Econometrica 56(5), 1119-1151.
Stambaugh, R.F. (1999), Predictive regressions, Journal of Financial Economics 54, 375-
421.
Statman, D. (1980), Book values and stock returns, The Chicago MBA: A Journal of
Selected Papers 4, 25-45.
Summers, L.H. (1986), Does the stock market rationally reflect fundamental values?
Sundaresan, S.M. (1989), Intertemporally dependent preferences and the volatility of
consumption and wealth, Review of Financial Studies 2, 73-89.
Telmer, C. (1993), Asset-pricing puzzles and incomplete markets, Journal of Finance 48,
1803-1832.
Thaler, R.H. (1985), Mental accounting and consumer choice, Marketing Science 4, 199-
214.
Thaler, R.H. and S. Benartzi (2004), Save More Tomorrow: using behavioral economics
to increase employee saving, Journal of Political Economy 112 (supplement), S164-
S187.
Tirole, J . (1985), Asset bubbles and overlapping generations, Econometrica 53, 1499-1528.
Tversky, A., and D. Kahneman (1974), J udgement under uncertainty: Heuristics and biases,
Science 185, 1124-1131.
Weil, P. (1989), The equity premium puzzle and the risk-free rate puzzle, Journal of
Monetary Economics 24, 401-421.
Wurgler, J ., and E. Zhuravskaya (2002), Does arbitrage flatten demand curves for stocks?
Journal of Business 75(4), 583-608.
Xiong, W., and J . Yu (2011), The Chinese warrants bubble, American Economic Review
101, 2723-2753.
Zweig, M.E. (1973), An investor expectations stock price predictive model using closed-end
fund premiums, Journal of Finance 28(1), 67-78.
15 OCTOBER 2012
Scientifc Background on the Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2012
STABLE ALLOCATI ONS AND
THE PRACTI CE OF MARKET DESI GN
1 Introduction
Economists study how societies allocate resources. Some allocation problems
are solved by the price system: high wages attract workers into a particu-
lar occupation, and high energy prices induce consumers to conserve energy.
In many instances, however, using the price system would encounter legal
and ethical objections. Consider, for instance, the allocation of public-school
places to children, or the allocation of human organs to patients who need
transplants. Furthermore, there are many markets where the price system
operates but the traditional assumption of perfect competition is not even
approximately satised. In particular, many goods are indivisible and het-
erogeneous, whereby the market for each type of good becomes very thin.
How these thin markets allocate resources depends on the institutions that
govern transactions.
This years prizewinning work encompasses a theoretical framework for
analyzing resource allocation, as well as empirical studies and actual redesign
of real-world institutions such as labor-market clearinghouses and school ad-
missions procedures. The foundations for the theoretical framework were
laid in 1962, when David Gale and Lloyd Shapley published a mathematical
inquiry into a certain class of allocation problems. They considered a model
with two sets of agents, for example workers and rms, that must be paired
with each other. If a particular worker is hired by employer A, but this
worker would have preferred employer B, who would also have liked to hire
this worker (but did not), then there are unexploited gains from trade. If
employer B had hired this worker, both of them would have been better o.
Gale and Shapley dened a pairing to be stable if no such unexploited gains
from trade exist. In an ideal market, where workers and rms have unre-
stricted time and ability to make deals, the outcome would always be stable.
Of course, real-world markets may dier from this ideal in important ways.
But Gale and Shapley discovered a deferred-acceptance procedure which
is easy to understand and always leads to a stable outcome. The procedure
species how agents on one side of the market (e.g., the employers) make
oers to those on the other side, who accept or reject these oers according
to certain rules.
The empirical relevance of this theoretical framework was recognized by
Alvin Roth. In a study published in 1984, Roth found that the U.S. market
for new doctors had historically suered from a series of market failures, but
a centralized clearinghouse had improved the situation by implementing a
1
procedure essentially equivalent to Gale and Shapleys deferred-acceptance
process. Roths 1984 article claried the tasks that markets perform, and
showed how the concept of stability provides an organizing principle which
helps us understand why markets sometimes work well, and why they some-
times fail to operate properly.
Subsequently, Roth and his colleagues used this framework, in combina-
tion with empirical studies, controlled laboratory experiments and computer
simulations, to examine the functioning of other markets. Their research has
not only illuminated how these markets operate, but has also proved useful
in designing institutions that help markets function better, often by imple-
menting a version or extension of the Gale-Shapley procedure. This has led
to the emergence of a new and vigorous branch of economics known as mar-
ket design. Note that in this context the term market does not presuppose
the existence of a price system. Indeed, monetary transfers are ruled out in
many important applications.
The work that is rewarded this year uses tools from both non-cooperative
and cooperative game theory. Non-cooperative game theory was the subject
of the 1994 Prize to John Harsanyi, John Nash and Reinhard Selten, and the
2005 Prize to Robert Aumann and Thomas Schelling. The starting point for a
non-cooperative analysis is a detailed description of a strategic problem faced
by individual decision makers. In contrast, cooperative game theory studies
how groups (coalitions) of individuals can further their own interests by
working together. The starting point for a cooperative analysis is therefore a
description of what each coalition can achieve. The person chiey responsible
for the development of cooperative game theory is Lloyd Shapley.
In many ways, the cooperative and non-cooperative approaches comple-
ment each other. Two properties of key importance for market design are
stability, which encourages groups to voluntarily participate in the market,
and incentive compatibility, which discourages strategic manipulation of the
market. The notion of stability is derived from cooperative game theory,
while incentive compatibility comes from the theory of mechanism design, a
branch of non-cooperative game theory which was the subject of the 2007
Prize to Leonid Hurwicz, Eric Maskin and Roger Myerson.
Controlled laboratory experiments are frequently used in the eld of mar-
ket design. Vernon Smith shared the 2002 Prize for his work in experimental
economics. Alvin Roth is another major contributor in this area.
The combination of game theory, empirical observations and controlled
experiments has led to the development of an empirical science with many
2
important practical applications. Evidence from the actual implementation
of newly designed or redesigned institutions creates an important interplay
and feedback eect: the discovery of a practical problem in implementation
may trigger theoretical elaboration, new experiments, and nally changes in
a design. Although these components form an integrated whole, we describe
them separately, starting with some basic theoretical concepts. We introduce
the idea of stability in Section 2. Then we describe some models of matching
markets in Section 3, with emphasis on the Gale-Shapley deferred-acceptance
procedure. In Section 4, we review how Alvin Roth recognized the real-world
relevance of the theory. Some real-world cases of market design are outlined
in Section 5. In Section 6, we note other important contributions of the two
laureates. Section 7 concludes.
2 Theory I: Stability
Gale and Shapley (1962) studied stable allocations in the context of a specic
model which will be described in Section 3. But rst we will consider the idea
of stability from the more general perspective of cooperative game theory.
2.1 Coalitional games with transferable utility
In this section we introduce some basic denitions from cooperative game
theory.
1
Consider a set N = f1; 2; :::; ng of n individuals (or players), for
example, traders in a market. A group of individuals who cooperate with
each other are said to form a coalition. A game in coalitional form with
transferable utility species, for each coalition S N, its worth v(S). The
worth is an economic surplus (a sum of money) that coalition S can generate
using its own resources. If coalition S forms, then its members can split the
surplus v(S) in any way they want, and each members utility equals her
share of the surplus; we call this transferable utility. The function v is
called the characteristic function. Two special coalitions are the singleton
coalition fig consisting only of player i 2 N, and the grand coalition N
consisting of all players.
Cooperative game theory studies the incentives of individuals to form
coalitions, given that any potential conicts of interest within a coalition
1
The formal apparatus of cooperative game theory was introduced in von Neumann
and Morgensterns (1944) classical work.
3
can be solved by binding agreements. These agreements induce the coalition
members to take actions that maximize the surplus of the coalition, and
this maximized surplus is what the coalition is worth. A diculty arises,
however, if the surplus also depends on actions taken by non-members. In
this case, the worth of a coalition can be determined in a consistent way by
assuming that the non-members try to maximize their own payos (Huang
and Sjstrm, 2003, Kczy, 2007).
In games with transferable utility, it is assumed that the players can freely
transfer utility among themselves, in eect by making side-payments. But
in some environments, side-payments are constrained and utility is not (per-
fectly) transferable. For example, in the National Resident Matching Pro-
gram discussed below, wages are xed before the market opens (Roth, 1984a).
In other situations, such as donations of human organs, side-payments are
considered repugnant (Roth, 2007). Cooperative game theory can han-
dle such situations, as it is very well developed for general non-transferable
utility games.
2.2 Stability and the core
Let x
i
denote individual is payo, and let x = (x
1
; x
2
; :::x
n
) denote the payo
vector. If the members of some coalition S can use their own resources to
make themselves better o, then we say that coalition S can improve upon
x, or block x. When utility is transferable, coalition S can improve upon x if
X
i2S
x
i
< v(S): (1)
Indeed, if inequality (1) holds, then S can produce v(S) and distribute this
surplus so as to make all its members strictly better o than they are under
x. The allocation x is then unstable.
An allocation is said to be stable if it cannot be improved upon by any
coalition.
2
Thus, with transferable utility, the payo vector x is stable if
X
i2S
x
i
v(S)
for every coalition S N. The set of all stable payo vectors is called the
core.
2
Stability has various denitions in the literature. Throughout this document we refer
solely to stability against any possible coalitional deviation.
4
Although we have introduced stability in the context of transferable util-
ity games, the denition extends in a straightforward way to general non-
transferable utility games. In general, an allocation is stable if no coalition
can improve upon it. That is, no coalition, by using its own resources, can
bring about an outcome that all its members prefer.
3
The idea of stability in cooperative game theory corresponds to the idea
of Nash equilibrium in non-cooperative game theory. In non-cooperative
game theory, a Nash equilibrium is a situation such that no individual can
deviate and make herself better o. In cooperative game theory, a stable
allocation is a situation such that no coalition can deviate and make its
members better o. From an economic point of view, stability formalizes an
important aspect of idealized frictionless marketplaces. If individuals have
unlimited time and ability to strike deals with each other, then the outcome
must be stable, or else some coalition would have an incentive to form and
make its members better o. This basic idea is due to Edgeworth (1881),
and is implicit in von Neumann and Morgensterns (1944) analysis of stable
set solutions. D.B. Gillies (1953a,b, 1959) and Shapley (1953c, 1955) were
the rst to explicitly consider the core as an independent solution concept.
Laboratory experiments, where subjects must reach an agreement without
any formalized procedure for making and accepting proposals, have provided
support for the prediction that the nal agreement will belong to the core
(Berl, McKelvey, Ordeshook and Winer, 1976).
The following example shows how stable allocations are identied and
that the core (i.e., the set of stable allocations) is sometimes quite large.
Example 1 A partnership consisting of one senior partner (Mary) and two
junior partners (Peter and Paul) generates earnings of 135. If Mary leaves
the partnership, she can earn 50 on her own: v(fMaryg) = 50. Any ju-
nior partner can earn 10 on his own: v(fPeterg) = v(fPaulg) = 10.
Mary and one junior partner together can earn 90, so v(fMary; Peterg) =
3
In non-transferable utility games there can be a distinction between weak and strong
improvement. The most common denition states that a coalition can improve upon an
allocation if all its members can be made strictly better o. However, the results of
Roth and Postlewaite (1977) suggest that it can sometimes be reasonable to use a weaker
requirement: a coalition can improve upon an allocation if some members can be made
strictly better o while no member is made strictly worse o. If improvement is dened
in this weaker sense, then some coalition members may be indierent with respect to
participating in the coalition, but they are still assumed to participate.
5
v(fMary; Paulg) = 90. The two juniors together can earn 25, so that
v(fPeter; Paulg) = 25. The grand coalition is worth 135 and utility is trans-
ferable, so they are free to divide up the 135 in any way they want. What is
the maximum and minimum payo Mary can get in a stable allocation?
Mary must get at least 50, and each junior partner must get at least 10, to
induce them to participate. Thus, stability requires
x
Mary
50; x
Peter
10; x
Paul
10:
Two-player coalitions must also be taken into account: the two junior partners
can improve on the allocation if they together get strictly less than 25, while
a coalition of Mary and one junior partner can improve if they together get
strictly less than 90. Thus, stability also requires
x
Peter
+x
Paul
25; x
Mary
+x
Peter
90; x
Mary
+x
Paul
90: (2)
These inequalities, together with the partnerships budget restriction x
Mary
+
x
Peter
+ x
Paul
= 135, imply that Marys minimum payo is 50, and the
maximum is 110.
4
2.3 Do stable allocations always exist?
Example 1 shows that the core may be quite large. In other instances, the
situation may be quite the opposite and the core may even be empty. To
illustrate this, suppose Example 1 is modied so that the surplus generated
by the grand coalition is only 101. This yields the budget restriction x
Mary
+
x
Peter
+x
Paul
= 101. But if we add the three inequalities in (2), which must
still be satised, we nd that stability requires 2 (x
Mary
+x
Peter
+x
Paul
)
205. Thus, the surplus of 101 is too small to allow a stable allocation. In
general, if there is not enough surplus available, it may be impossible to divide
it up in a stable way. Bondareva (1963) and Shapley (1967) independently
derived an exact formula for how much surplus must be available in order for
the core to be non-empty in games with transferable utility. Their result was
extended to games without transferable utility by Scarf (1967) and Billera
4
The most important single-valued solution concept in cooperative game theory is the
Shapley value (Shapley 1953a). Marys Shapley value is 80, the midpoint of the interval
[50; 110] and arguably a reasonable compromise.
6
(1970). Shapley (1971) showed that the core is always non-empty if the game
is convex (in the sense that the value of a players marginal contribution to
a coalition is increased if other players join the coalition).
5
2.4 Core and competitive equilibrium
Edgeworth (1881) was the rst to argue that if some traders are not satised
with what they receive on the market, then they can recontract, i.e., withdraw
from the market and trade among themselves (not necessarily at prevailing
market prices). The contract curve is the set of outcomes that cannot be
destabilized by recontracting. As Shubik (1959) noted, Edgeworths contract
curve corresponds to the core of the economy. Edgeworth conjectured that
in markets with suciently many traders, the contract curve would approxi-
mately equal the competitive equilibrium, and he veried this conjecture for
the special case of two goods and two types of traders. Debreu and Scarf
(1963) veried Edgeworths conjecture under more general assumptions: if
the economy is replicated so the number of traders of each type becomes
very large, then the core approximately equals the set of competitive equi-
libria (see also Shubik, 1959). Thus, without having to specify the precise
rules that govern trade, the core provides a key theoretical foundation for
competitive equilibrium.
But many environments dier considerably from the perfectly competitive
benchmark. Examples include collective-choice problems, such as choosing
the level of a public good, and the matching markets which will be described
in the next section. In the non-cooperative approach to such problems, in-
stitutions are analyzed in detail, and a solution concept such as Nash equi-
librium is applied. The cooperative approach, on the other hand, can make
predictions which are independent of the ne details of institutions. Specif-
ically, if agents have unrestricted contracting ability, then the nal outcome
must be stable, for any unstable outcome will be overturned by some coali-
tion that can improve upon it. We will now describe how Shapley and his
colleagues applied this idea to various models of matching.
5
Example 1 is an example of a convex game.
7
3 Theory II: Matching Markets
In many markets, goods are private but indivisible and heterogeneous, and
the traditional assumption of perfect competition cannot be maintained. Im-
portant examples include markets for skilled labor. Since no two workers have
exactly the same characteristics, the market for each particular bundle of la-
bor services can be quite thin. In such markets, the participants must be
appropriately matched in order to trade with each other.
3.1 Two-sided matching
Consider a market with two disjoint sets of agents such as buyers and sell-
ers, workers and rms, or students and schools that must be matched with
each other in order to carry out transactions. Such two-sided matching mar-
kets were studied by Gale and Shapley (1962). They ruled out side-payments:
wages (and other match characteristics) are not subject to negotiation.
Stable matchings To be specic, suppose one side of the market consists
of medical students and the other of medical departments. Each department
needs one intern and each medical student wants one internship. A matching
is an assignment of internships to applicants. Naturally, the students have
preferences over departments, and the departments have preferences over
students. We assume for convenience that preferences are strict (i.e., no
ties). A matching is said to be unacceptable to an agent if it is worse than
remaining unmatched.
In general, a matching is stable if no coalition can improve upon it. In this
particular model, a stable matching must satisfy the following two conditions:
(i) no agent nds the matching unacceptable, and (ii) no department-student
pair would prefer to be matched with each other, rather than staying with
their current matches. Condition (i) is an individual rationality condition and
condition (ii) is pairwise stability. The two conditions imply that neither any
singleton coalition, nor any department-student pair, can improve on the
matching. (These are the only coalitions we need to consider in this model.)
6
6
Gale and Shapley (1962) dened a matching to be stable if no coalition consisting of
one agent from each side of the market could improve on it (i.e., pairwise stability). Given
the special structure of their model, this was equivalent to nding a matching in the core.
8
The Gale-Shapley algorithm Gale and Shapley (1962) devised a deferred-
acceptance algorithm for nding a stable matching. Agents on one side of the
market, say the medical departments, make oers to agents on the other side,
the medical students. Each student reviews the proposals she receives, holds
on to the one she prefers (assuming it is acceptable), and rejects the rest. A
crucial aspect of this algorithm is that desirable oers are not immediately
accepted, but simply held on to: deferred acceptance. Any department whose
oer is rejected can make a new oer to a dierent student. The procedure
continues until no department wishes to make another oer, at which time
the students nally accept the proposals they hold.
In this process, each department starts by making its rst oer to its
top-ranked applicant, i.e., the medical student it would most like to have as
an intern. If the oer is rejected, it then makes an oer to the applicant it
ranks as number two, etc. Thus, during the operation of the algorithm, the
departments expectations are lowered as it makes oers to students further
and further down its preference ordering. (Of course, no oers are made to
unacceptable applicants.) Conversely, since students always hold on to the
most desirable oer they have received, and as oers cannot be withdrawn,
each students satisfaction is monotonically increasing during the operation of
the algorithm. When the departments decreased expectations have become
consistent with the students increased aspirations, the algorithm stops.
Example 2 Four medical students (1, 2, 3 and 4) apply for internships in
four medical departments: surgery (S), oncology (O), dermatology (D) and
pediatrics (P). All matches are considered acceptable (i.e., better than re-
maining unmatched). The students have the following preference orderings
over internships:
1: S O D P
2: S D O P
3: S O P D
4: D P O S
Thus, surgery is the most desirable internship, ranked rst by three of the
students. Each medical department needs one intern. They have the following
preference orderings over students:
S: 4 3 2 1
O: 4 1 3 2
D: 1 2 4 3
P: 2 1 4 3
9
Internships are allocated using the Gale-Shapley algorithm, with the depart-
ments making proposals to the students. Each department makes its rst
oer to its top-ranked applicant: student 1 gets an internship oer from D,
student 2 gets one from P, and student 4 gets oers from S and O. Student
4 prefers O to S, so she holds on to the oer from O and rejects the oer
from S. In the second round, S oers an internship to student 3. Now each
student holds an internship, and the algorithm stops. The nal assignment
is:
1 !D; 2 !P; 3 !S; 4 !O:
Gale and Shapley (1962) proved that the deferred-acceptance algorithm is
stable, i.e., it always produces a stable matching. To see this, note that in Ex-
ample 2, the algorithm allocates student 2 to her least preferred department,
P, the only one to make her an oer. Now note that departments D, S and O
have been assigned interns they think are preferable to student 2 this must
be the case, otherwise they would have oered 2 an internship before making
oers to their assigned interns. Thus, even if they could replace their assigned
interns with student 2, they would not want to do so. By this argument, any
department which a student prefers to her assignment will not prefer her to
its assigned intern, so the match is pairwise stable. Individual rationality
holds trivially in Example 2, since there are no unacceptable matches, but it
would hold in general as well, because students would reject all unacceptable
oers, and departments would never make oers to unacceptable applicants.
The case where each department wants one intern corresponds to Gale
and Shapleys (1962) marriage model. The case where departments may
want more than one intern is their college admissions model. Gale and
Shapley (1962) showed how the results for the marriage model generalize
to the college admissions model. In particular, a version of the deferred-
acceptance algorithm produces stable matchings even if departments want
to hire more than one intern.
7
The algorithm provides an existence proof for this type of two-sided
matching problem: since it always terminates at a stable matching, a sta-
ble matching exists. In fact, more than one stable matching typically exists.
Gale and Shapley (1962) showed that interests are polarized in the sense that
7
In the college admissions model, Gale and Shapley (1962) did not formally specify the
employers preferences over dierent sets of employees, but this was done in later work
(see Roth, 1985).
10
dierent stable outcomes favor one or the other side of the market.
8
This
leads to a delicate issue: for whose benet is the algorithm operated?
Who gains the most? In Example 2, the nal assignment favors the
medical departments more than the students.
9
In general, the employer-
proposing version of the algorithm, where employers propose matches as in
Example 2, produces an employer-optimal stable matching: all employers
agree it is the best of all possible stable matchings, but all applicants agree
it is the worst. The symmetric applicant-proposing version of the algorithm
instead leads to an applicant-optimal stable matching (which all applicants
agree is the best but all employers agree is the worst). This illustrates how
the applicants interests are opposed to those of the employers, and how
stable institutions can be designed to systematically favor one side of the
market.
Example 3 The preferences are as in Example 2, but now the students have
the initiative and make the proposals. Students 1, 2 and 3 start by making
proposals to S, while student 4 makes a proposal to D. Since S prefers student
3, it rejects 1 and 2. In the second round, 1 makes a proposal to O and 2
makes a proposal to D. Since D prefers 2 to 4, it rejects 4. In the third round,
4 proposes to P. Now each department has an intern, and the algorithm stops.
The nal assignment is:
1 !O; 2 !D; 3 !S; 4 !P:
It can be checked that the students strictly prefer this assignment to the as-
signment in Example 2, except for student 3 who is indierent (she is assigned
to S in both cases). The departments are strictly worse o than in Example
2, except for S which gets student 3 in either case.
A social planner could conceivably reject both the applicant-optimal
and employer-optimal stable matchings in favor of a stable matching that
satises some fairness criterion, or perhaps some version of majority rule
8
This insight follows from the more general fact that the set of stable matchings has
the mathematical structure of a lattice (Knuth, 1976).
9
Departments P, D and O get their most preferred intern, while S gets the intern it
ranks second. The only student who is assigned to her favorite department is student 3.
Students 1 and 4 are allocated to departments they rank third, while student 2 is assigned
to the department she considers the worst.
11
(Grdenfors, 1975). In practice, however, designs have tended to favor the
applicants. In the context of college admissions, Gale and Shapley (1962)
argued in favor of applicant-optimality based on the philosophy that colleges
exist for the sake of the students, not the other way around.
Incentive compatibility Can the Gale-Shapley algorithm help partici-
pants in real-world markets nd stable matchings? An answer to this ques-
tion requires a non-cooperative analysis, that is, a detailed analysis of the
rules that govern the matching process and the incentives for strategic be-
havior, to which we now turn.
Above, the deferred-acceptance algorithm was explained as a decentral-
ized procedure of applications, oers, rejections and acceptances. But in
practice, the algorithm is run by a clearinghouse in a centralized fashion.
Each applicant submits her preference ordering, i.e., her personal ranking of
the employers from most to least preferred. The employers submit their pref-
erences over the applicants. Based on these submitted preferences, the clear-
inghouse goes through the steps of the algorithm. In the language of mecha-
nism design theory, the clearinghouse runs a revelation mechanism, a kind of
virtual market which does not suer from the problems experienced by some
real-world markets (as discussed later, these include unraveling and conges-
tion). The revelation mechanism induces a simultaneous-move game, where
all participants submit their preference rankings, given a full understanding
of how the algorithm maps the set of submitted rankings into an allocation.
This simultaneous-move game can be analyzed using non-cooperative game
theory.
A revelation mechanism is (dominant strategy) incentive compatible if
truth-telling is a dominant strategy, so that the participants always nd it op-
timal to submit their true preference orderings. The employer-proposing al-
gorithm viewed as a revelation mechanism is incentive compatible for the
employers: no employer, or even coalition of employers, can benet by mis-
representing their preferences (Dubins and Freedman, 1981, Roth, 1982a).
10
However, the mechanism is not incentive compatible for the applicants. To
see this, consider the employer-proposing algorithm of Example 2. Suppose
all participants are truthful except student 4, who submits D P S O,
10
More precisely, no coalition of employers can improve in the strong sense that every
member is made strictly better o (see Roth and Sotomayor, 1990, Chapter 4, for a
discussion of the robustness of this result).
12
which is a manipulation or strategic misrepresentation of her true prefer-
ences D P O S. The nal matching will be the one that the
applicant-proposing algorithm produced in Example 3, so student 4s manip-
ulation makes her strictly better o.
11
This proves that truth-telling is not
a dominant strategy for the applicants. Indeed, Roth (1982a) proved that
no stable matching mechanism exists for which stating the true preferences
is a dominant strategy for every agent. However, notice that despite student
4s manipulation, the nal matching is stable under the true preferences.
Moreover, it is an undominated Nash equilibrium outcome. This illustrates
a general fact about the Gale-Shapley algorithm, proved by Roth (1984b):
all undominated Nash equilibrium outcomes of the preference manipulation
game are stable for the true preferences.
12
The usefulness of Roths (1984b) result is limited by the fact that it may
be dicult for applicants to identify their best responses, as required by the
denition of Nash equilibrium. For example, if student 4 knows that the
other applicants are truthful but not what their true preferences are, then
student 4 will not be able to foresee the events outlined in Footnote 11.
Therefore, she cannot be sure that this particular manipulation is protable.
This argument suggests that in large and diverse markets, where participants
have very limited information about the preferences of others, the scope for
strategic manipulation may be quite limited. Roth and Rothblum (1999)
verify that when an applicants information is suciently limited, she cannot
gain by submitting a preference ordering which reverses her true ordering of
two employers. However, it may be protable for her to pretend that some
acceptable employers are unacceptable.
11
Consider how Example 2 would be dierent if student 4s preferences were D P
S O. Then, when student 4 receives simultaneous oers from O and S, she rejects O.
In the second round, O would make an oer to student 1 who holds an oer from D but
prefers O and therefore rejects D. In the third round, D makes an oer to 2 who holds an
oer from P but prefers D and therefore rejects P. In the fourth round, P makes an oer
to 1 who holds an oer from O and rejects P. In the fth round, P makes an oer to 4
who holds an oer from S and now rejects S. In the sixth round, S makes an oer to 3.
The algorithm then stops; the nal assignment is
1 !O; 2 !D; 3 !S; 4 !P:
12
If we believe coalitions can jointly manipulate their reports, we can restrict attention
to undominated Nash equilibria that are strong or rematching proof (Ma, 1995). Roths
(1984b) result applies to these renements as well.
13
Roth and Peranson (1999) used computer simulations with randomly gen-
erated data, as well as data from the National Resident Matching Program,
to study the gains from strategic manipulation of a deferred-acceptance algo-
rithm. Their results suggested that in large markets, very few agents on either
side of the market could benet by manipulating the algorithm. Subsequent
literature has claried exactly how the gains from strategic manipulation
vanish in large markets (Immorlica and Mahdian, 2005, Kojima and Pathak,
2009).
A related question is the following: if participants have incomplete infor-
mation, does there exist a Bayesian-Nash equilibrium (not necessarily truth-
telling) such that the outcome is always stable for the true preferences? Roth
(1989) proved that this cannot be true for any mechanism, assuming both
sides of the market behave strategically.
13
However, the applicant-proposing
deferred-acceptance mechanism is incentive compatible for the applicants, so
if the employers do not behave strategically, then truthtelling is a Bayesian-
Nash equilibrium which produces a stable matching.
14
Even if both sides
of the market are strategic, the lack of incentive-compatibility is less serious
in large markets where, as Roth and Peranson (1999) discovered, the poten-
tial gains from strategic manipulation are limited. Under certain conditions,
truthful reporting by both sides of the market is an approximate equilibrium
for the applicant-proposing deferred-acceptance mechanism in a suciently
large market (Kojima and Pathak, 2009).
Adjustable prices and wages Shapley and Shubik (1971) considered a
transferable-utility version of the Gale-Shapley model called the assignment
game. When employers are matched with workers, transferable utility means
that match-specic wages are endogenously adjusted to clear the market.
Shapley and Shubik (1971) showed that the core of the assignment game
is non-empty, and that competition for matches puts strict limits on the set of
core allocations. With transferable utility, any core allocation must involve
a matching which maximizes total surplus. Generically, this matching is
13
This negative result applies to all mechanisms, not just revelation mechanisms. How-
ever, Roths (1989) proof relies on the revelation principle, which states that without loss
of generality, we can restrict attention to incentive compatible revelation mechanisms.
14
The revelation mechanism which selects the applicant-optimal stable matching is
(dominant strategy) incentive compatible for the applicants also in the college admis-
sions model, where employers make multiple hires (Roth, 1985a).
14
unique. However, wages are not in general uniquely determined, thereby cre-
ating a polarization of interests similar to the Gale-Shapley model. Employer-
optimal and applicant-optimal stable allocations exist and are characterized
by the lowest and highest possible market-clearing wages. The core of the
assignment game captures a notion of free competition reminiscent of tradi-
tional competitive analysis. In fact, in this model there is an exact corre-
spondence between core and competitive equilibria.
Shapley and Shubik (1971) did not provide an algorithm for reaching sta-
ble allocations when utility is transferable, but Crawford and Knoer (1981)
showed that a generalized Gale-Shapley algorithm accomplishes this task
(see also Demange, Gale and Sotomayor, 1986). In the employer-proposing
version, each employer starts by making a low salary oer to its favorite ap-
plicant. Any applicant who receives more than one oer holds on to the most
desirable oer and rejects the rest. Employers whose oers are rejected con-
tinue to make oers, either by raising the salary oer to the same applicant,
or by making an oer to a new applicant. This process always leads to the
employer-optimal stable allocation. Kelso and Crawford (1982) and Roth
(1984c, 1985b) generalized these results still further. Specically, Kelso and
Crawford (1982) introduced the assumption that employers, who have pref-
erences over sets of workers, consider workers to be substitutes. Under this
assumption, an employer-proposing deferred-acceptance algorithm still pro-
duces the employer-optimal stable allocation, while an applicant-proposing
version produces the applicant-optimal stable allocation (Kelso and Craw-
ford, 1982, Roth, 1984c).
15
When side-payments are available, the deferred-acceptance algorithm can
be regarded as a simultaneous multi-object English auction, where no object
is allocated until bidding stops on all objects. As long as the objects for sale
are substitutes, this process yields the bidder-optimal core allocation. Roth
and Sotomayor (1990, Part III) discuss the link between matching and auc-
tions, a link which was further strengthened by Hateld and Milgrom (2005).
Varian (2007) and Edelman, Ostrovsky and Schwarz (2007) showed that the
assignment game provides a natural framework for analyzing auctions used
by Internet search engines to sell space for advertisements.
15
If employers want to hire more than one worker, the employer-proposing algorithm
lets them make multiple oers at each stage. To see why substitutability is required for
this algorithm to work as intended, note that this assumption guarantees that if an oer
is rejected by a worker, the employer does not want to withdraw any previous oers made
to other workers.
15
3.2 One-sided matching
Shapley and Scarf (1974) studied a one-sided market, where a set of traders
exchange indivisible objects (such as plots of land) without the ability to use
side-payments. Each agent initially owns one object. Abdulkadiro glu and
Snmez (1999) later generalized the model to allow for the possibility that
some agents do not initially own any objects, while some objects have no
initial owner.
Shapley and Scarf (1974) proved that the top-trading cycle algorithm,
which they attributed to David Gale, always produces a stable allocation.
The algorithm works as follows. Starting from the initial endowment, each
agent indicates her most preferred object. This can be described in a di-
rected graph indicating, for each agent, whose object this agent would pre-
fer. There must exist at least one cycle in the directed graph, i.e., a set
of agents who could all obtain their preferred choices by swapping among
themselves. These swaps occur, and the corresponding agents and objects
are removed from the market. The process is repeated with the remaining
agents and objects, until all objects have been allocated. The algorithm is
illustrated in the following example.
Example 4 There are four agents, 1, 2, 3, and 4, and four objects, A, B,
C, and D. The agents have the following preferences:
1: A B C D
2: B A D C
3: A B D C
4: D C A B.
Given two alternative initial endowment structures, Figure 1 indicates the
implied preferences with arrows.
16
Figure 1: Top Trading Cycles for dierent endowment structures.
On the left-hand side of Figure 1 the initial allocation (endowment) is DCBA,
that is, agent 1 owns object D; agent 2 owns object C; etc. In stage 1, agent
1 indicates that her favorite object is owned by 4, while 4 indicates that her
favorite object is owned by 1. Thus, agents 1 and 4 form a cycle. They swap
objects and are removed together with their objects D and A. Now agents 2
and 3 remain, with their endowments C and B. In the second stage, both
2 and 3 indicate that 3s object is their favorite among the two objects that
remain. Therefore, the process terminates with the nal allocation ACBD.
This allocation is stable: no coalition of traders can reallocate their initial
endowments to make all members better o. The right-hand side of the g-
ure shows that, had the initial endowment been ABCD, then no trade would
have occurred.
Roth and Postlewaite (1977) show that if preferences over objects are
strict, and if stability is dened in terms of weak improvements (see Footnote
3), then for any given initial endowment there is a unique stable allocation.
For example, if the initial endowment in Example 4 is DCBA, then the
coalition consisting of agents 1 and 4 can obtain their favorite objects simply
17
by swapping their endowments. Hence, any stable outcome must give A to
1 and D to 4: Object C cannot be given to agent 3, because she can block
this by refusing to trade. Hence the unique stable allocation is ACBD. In
contrast, when the initial endowment is ABCD, agents 1, 2 and 4 would
block any outcome where they do not obtain their favorite objects (which
they already own), so ABCD is the only stable allocation.
The revelation mechanism that chooses the unique stable allocation, com-
puted by the top-trading cycle algorithm from submitted preference orderings
and given endowments, is dominant strategy incentive compatible for all par-
ticipants (Roth, 1982b). In fact, this is the only revelation mechanism which
is Pareto ecient, individually rational and incentive compatible (Ma, 1994).
Important real-world allocation problems have been formalized using Shap-
ley and Scarfs (1974) model. One such problem concerns the allocation of
human organs, which will be discussed in Section 5.3. Another such problem
concerns the allocation of public-school places to children. In the school-
choice problem, no initial endowments exist, although some students may
be given priority at certain schools. Abdulkadiro glu and Snmez (2003)
adapted the top-trading cycle to the school choice problem, but another ap-
proach is to incorporate the schools preferences over students via the Gale-
Shapley algorithm. This will be discussed in Section 5.2.
4 Evidence: Markets for Doctors
The work on stable allocations and stable algorithms was recognized as an
important theoretical contribution in the 1960s and 1970s, but it was not
until the early 1980s that its practical relevance was discovered. The key
contribution is Roth (1984a), which documents the evolution of the market
for new doctors in the U.S. and argues convincingly that a stable algorithm
improved the functioning of the market. This work opened the door to Roths
participation in actual design, which began in the 1990s. Roth also conducted
empirical studies of other medical markets, documenting and analyzing how
several regions in the U.K. had adopted dierent algorithms (Roth, 1991a).
These further strengthened the case for stable algorithms. The overall evi-
dence provided by Roth was pivotal.
Centralized matching mechanisms, such as the one in the U.S. market
for new doctors, have well-dened rules of the game known to both the
participants themselves and the economists who study the market. Knowl-
18
edge of these rules makes it possible to test game-theoretic predictions, in
the eld as well as in laboratory experiments. Moreover, the rules can be
redesigned to improve the market functioning (see Section 5). Accordingly,
these types of matching mechanisms have been studied in depth and are by
now well understood. Other markets with clearly dened rules have also been
the subject of intensive studies; the leading example is auction markets. In
fact, matching and auction theory are closely linked, as mentioned above.
We begin this section by describing the U.S. market for new doctors,
and then turn to the U.K. regional medical markets. We also consider how
important evidence regarding the performance of matching algorithms have
been generated using laboratory experiments.
4.1 The U.S. market for new doctors
Roth (1984a) studied the evolution of the U.S. market for new doctors. Stu-
dents who graduate from medical schools in the U.S. are typically employed
as residents (interns) at hospitals, where they comprise a signicant part
of the labor force. In the early twentieth century, the market for new doc-
tors was largely decentralized. During the 1940s, competition for medical
students forced hospitals to oer residencies (internships) increasingly early,
sometimes several years before a student would graduate. This so-called
unraveling had many negative consequences. Matches were made before stu-
dents could produce evidence of how qualied they might become, and even
before they knew what kind of medicine they would like to practice. The
market also suered from congestion: when an oer was rejected, it was of-
ten too late to make other oers. A congested market fails to clear, as not
enough oers can be made in time to ensure mutually benecial trades. To
be able to make more oers, hospitals imposed strict deadlines which forced
students to make decisions without knowing what other opportunities would
later become available.
Following Roths (1984a) study, similar problems of congestion and unrav-
eling were found to plague many markets, including entry-level legal, business
school and medical labor markets in the U.S., Canada and the U.K., the mar-
ket for clinical psychology internships, dental and optometry residencies in
the U.S., and the market for Japanese university graduates (Roth and Xing,
1994). When indivisible and heterogeneous goods are traded, as in these
markets for skilled labor, oers must be made to specic individuals rather
than to the market. The problem of coordinating the timing of oers can
19
cause a purely decentralized market to become congested and unravel, and
the outcome is unlikely to be stable. Roth and Xing (1994) described how
market institutions have been shaped by such failures, and explained their
ndings in a theoretical model (see also Roth and Xing, 1997).
In response to the failures of the U.S. market for new doctors, a centralized
clearinghouse was introduced in the early 1950s. This institution is nowcalled
the National Resident Matching Program (NRMP). The NRMP matched
doctors with hospitals using an algorithm which Roth (1984a) found to be
essentially equivalent to Gale and Shapleys employer-proposing deferred-
acceptance algorithm. Although participation was voluntary, essentially all
residencies were allocated using this algorithm for several decades. Roth
(1984a) argued that the success of the NRMP was due to the fact that its al-
gorithm produced stable matchings. If the algorithm had produced unstable
matchings, doctors and hospitals would have had an incentive to bypass the
algorithm by forming preferred matches on the side (a doctor could simply
contact her favorite hospitals to inquire whether they would be interested in
hiring her).
16
When a market is successfully designed, many agents are persuaded to
participate, thereby creating a thick market with many trading opportuni-
ties. The way in which a lack of stability can create dissatisfaction and reduce
participation rates is illustrated by Example 2. An unstable algorithm might
assign student 1 to pediatrics. But if the dermatology department nds out
that their top-ranked applicant has been assigned to a department she likes
less than dermatology, they would have a legitimate reason for dissatisfaction.
A stable algorithm would not allow this kind of situation. It is thus more
likely to induce a high participation rate, thereby creating many opportuni-
ties for good matches which, in turn, induces an even higher participation
rate. Roth and his colleagues have identied this virtuous cycle in a number
of real-world markets, as well as in controlled laboratory experiments.
16
As in the original Gale and Shapley (1962) model, the only issue considered by the
NRMP is to nd a matching. Salaries are determined by employers before residencies are
allocated, so they are treated as exogenous to the matching process. Crawford (2008)
argues that it would be quite feasible to introduce salary exibility into the matching
process by using a generalized deferred-acceptance algorithm of the type considered by
Crawford and Knoer (1981). See also Bulow and Levin (2006).
20
4.2 Regional medical markets in the U.K.
Roth (1990, 1991a) observed that British regional medical markets suered
from the same kinds of problems in the 1960s that had a-icted the U.S.
medical market in the 1940s. Each region introduced its own matching al-
gorithm. Some were stable, others were not (see Table 1). Specically, the
clearinghouses in Edinburgh and Cardi implemented algorithms which were
essentially equivalent to the deferred-acceptance algorithm, and these oper-
ated successfully for decades. On the other hand, Birmingham, Newcastle
and Sheeld quickly abandoned their unstable algorithms.
Table 1. Reproduced from Roth (2002, Table 1).
4.3 Experimental evidence
The empirical evidence seems to support the hypothesis that stable match-
ing algorithms can prevent market failure (Roth and Xing, 1994, Roth, 2002,
Niederle, Roth, and Snmez, 2008). However, many conditions inuence the
success or failure of market institutions. The objective of market designers
is to isolate the role of the mechanism itself, and compare the performance
of dierent mechanisms under the same conditions. But this is dicult to
accomplish in the real world. For example, British regional medical markets
might dier in numerous ways that cannot be controlled by an economist.
21
Accordingly, market designers have turned to controlled laboratory experi-
ments to evaluate and compare the performance of mechanisms.
Kagel and Roth (2000) compared the stable (deferred-acceptance) algo-
rithm used in Edinburgh and Cardi with the unstable priority-matching
algorithm used in Newcastle.
17
In their experiment, a centralized match-
ing mechanism was made available to the subjects, but they could choose
to match in a decentralized way, without using the mechanism. When the
mechanism used priority matching, the experimental market tended to un-
ravel, and many matches were made outside the mechanism. The deferred-
acceptance mechanism did not suer from the same kind of unraveling. This
provided experimental evidence in favor of Roths hypothesis that the match-
ing algorithm itself and, in particular, its stability, contribute importantly to
the functioning of the market.
Two regions, Cambridge and London Hospital, presented an anomaly for
Roths hypothesis. In these regions, the matching algorithms solved a linear
programming problem which did not produce stable outcomes. Yet, these
markets did not appear to unravel, and the unstable mechanisms remained
in use (see Table 1). In experiments, the linear programming mechanisms
seem to perform no better than priority matching, which suggests that con-
ditions specic to Cambridge and London Hospital, rather than the intrinsic
properties of their matching algorithms, may have prevented unraveling there
(nver, 2005). Roth (1991a) argued that these markets are in fact so small
that social pressures may prevent unraveling.
In one U.S. medical labor market (for gastroenterology), a stable algo-
rithm was abandoned after a shock to the demand and supply of positions.
McKinney, Niederle and Roths (2005) laboratory experiments suggested that
this market failed mainly because, while employers knew about the exogenous
shock, the applicants did not. Shocks that both sides of the market knew
about did not seem to cause the same problems. This suggested that the al-
gorithm would fail only under very special conditions. Roth and M. Niederle
17
The unstable algorithms in Birmingham and Sheeld used a similar method as the
Newcastle algorithm. In priority-matching, an applicants ranking of an employer and the
employers ranking of the applicant jointly determine the applicants priority at that
employer. Thus, the highest priority matches are those where the two parties rank each
other rst. Apart from being unstable, such methods are far from incentive-compatible;
deciding whom to rank rst is a dicult strategic problem. A similar problem is discussed
below with regard to the Boston mechanism (which itself is a kind of priority-matching
algorithm).
22
helped the American Gastroenterology Association reintroduce a deferred-
acceptance matching algorithm in 2006. Niederle, Proctor and Roth (2008)
describe early evidence in favor of the reintroduced matching mechanism.
5 Market Design
The theory outlined in Sections 2 and 3 and the empirical evidence discussed
in Section 4 allow us to understand the functions that markets perform, the
conditions required for them to be performed successfully, and what can go
wrong if these conditions fail to hold. We now consider how these insights
have been used to improve market functioning. Of course, real-world markets
experience idiosyncratic complications that are absent in theoretical models.
Real-world institutions have to be robust to agents who make mistakes, do
not understand the rules, have dierent prior beliefs, etc. They should also
be appropriate to the historical and social context and, needless to say, re-
spect legal and ethical constraints on how transactions may be organized.
Given the constraints of history and prevailing social norms, small-scale in-
cremental changes to existing institutions might be preferred to complete
reorganizations.
This section deals with three sets of real-world applications: rst, the
market for doctors in the U.S.; second, the design of school-admission pro-
cedures; and third, a case of one-sided matching (kidney exchange).
5.1 Redesigning the market for new doctors
As described in Section 4.1, Roths work illuminated why the older, and more
decentralized, system had failed, and why the new (deferred-acceptance) al-
gorithm adopted by the NRMP performed so much better. However, as
described by Roth and Peranson (1999), the changing structure of the med-
ical labor market caused new complexities to arise which led the NRMP to
modify its algorithm. By the 1960s a growing number of married couples
graduated from medical school, and they often tried to bypass the algorithm
by contacting hospitals directly. A couple can be regarded as a composite
agent who wants two jobs in the same geographic location, and whose pref-
erences therefore violate the assumption of substitutability. Roth (1984a)
proved that in a market where some agents are couples, there may not exist
any stable matching. The design of matching and auction mechanisms in the
presence of complementarities is an important topic in the recent literature.
23
A need for reform: the Roth-Peranson algorithm In the 1990s, the
very legitimacy of the NRMP algorithm was challenged. Specically, it was
argued that what was primarily an employer-proposing algorithm favored
hospitals at the expense of students.
Medical-school personnel responsible for advising students about
the job market began to report that many students believed the
NRMP did not function in the best interest of students, and that
students were discussing the possibility of dierent kinds of strate-
gic behavior (Roth and Peranson, 1999, p. 749).
The basic theory of two-sided matching, outlined in Section 3.1, shows
that the employer-proposing algorithm is not incentive compatible for the
applicants, i.e., it is theoretically possible for them to benet by strategically
manipulating or gaming it. However, the applicant-proposing version is
incentive compatible for the applicants. The complexity of the medical la-
bor market, with complementarities involving both applicants and positions,
means that the basic theory cannot be applied directly. However, computa-
tional experiments show that the theory can provide useful advice even in
this complex environment (Roth and Peranson, 1999). Overall, there seemed
to be strong reasons to switch to an applicant-proposing algorithm.
In 1995, Alvin Roth was hired by the Board of Directors of the NRMP to
direct the design of a new algorithm. The goal of the design was to construct
an algorithm that would produce stable matchings as favorable as possible
to applicants, while meeting the specic constraints of the medical market
(Roth and Peranson, 1999, p. 751). The new algorithm, designed by Roth
and Elliott Peranson, is an applicant-proposing algorithm modied to accom-
modate couples: potential instabilities caused by the presence of couples are
resolved sequentially, following the instability-chaining algorithm of Roth and
Vande Vate (1990). Computer simulations suggested that the Roth-Peranson
algorithm would turn out to be somewhat better for the applicants than the
old NRMP employer-proposing algorithm (Roth and Peranson, 1999). The
simulations also revealed that, in practice, it would essentially be impossible
to gain by strategic manipulation of the new algorithm (Roth, 2002).
Since the NRMP adopted the new algorithm in 1997, over 20,000 doctors
per year have been matched by it (Roth and Peranson, 1999, Roth, 2002).
The same design has also been adopted by entry-level labor markets in other
professions (see Table 2). The empirical evidence suggests that the outcome
is stable despite the presence of couples.
24
Table 2. Reproduced from Roth (2008a, Table 1).
25
5.2 School admission
Many students simply attend the single school where they live. Sometimes,
however, students have potential access to many schools. A matching of
students with schools should take into account the preferences of the students
and their parents, as well as other important concerns (about segregation, for
example). Should schools also be considered strategic agents with preferences
over students? Some schools might prefer students with great attendance
records, others might be mainly concerned about grades, etc. If the schools,
as well as the applicants, are regarded as strategic agents, then a two-sided
matching problem ensues.
In the theoretical models of Balinski and Snmez (1999) and Abdulka-
diro glu and Snmez (2003), classroom slots are allocated among a set of
applicants, but the schools are not considered strategic agents. Insights from
two-sided matching models are still helpful, however. An applicant may be
given high priority at some particular school (for example, if she lives close
to the school, has a sibling who attends the school, or has a high score on a
centralized exam). In this case, the school can be said to have preferences
over students, in the sense that higher-priority students are more preferred.
Stability then captures the idea that if student 1 has higher priority than
student 2 at school S, and student 2 attends school S, then student 1 must
attend a school that she likes at least as well as S (perhaps S itself).
An important dierence between this model and the two-sided model of
Section 3.1 is that the priority ranking of students can be based on objectively
veriable criteria. In such instances, the problem of incentive-compatibility
does not necessarily arise on the part of the schools. Moreover, the priority
orderings do not have the same welfare implications as preference order-
ings usually have. These arguments suggest using the applicant-proposing
deferred-acceptance algorithm, which is not only fully incentive compatible
for the applicants, but also applicant optimal (i.e., every applicant prefers
it to any other stable match).
18
The New York City public high schools
started using a version of the deferred-acceptance algorithm in 2003, and the
Boston public school system started using a dierent version in 2005 (Roth,
2008b).
19
18
For experimental evidence on school-choice mechanisms, see Chen and Snmez (2006),
Featherstone and Niederle (2011) and Pais and Pintr (2008).
19
The problems the market designers faced in these two markets were somewhat dif-
ferent. In New York, the school-choice system is in eect a two-sided market where the
26
Prior to 2003, applicants to New York City public high schools were asked
to rank their ve most preferred schools and these preference lists were sent
to the schools. The schools then decided which students to admit, reject,
or wait-list. The process was repeated in two more rounds. Students who
had not been assigned to any school after the third round were assigned
via an administrative process. This process suered from congestion, as the
applicants did not have sucient opportunities to express their preferences,
and the schools did not have enough opportunities to make oers. The
market failed to clear: about 30,000 students per year ended up, via the
administrative process, at a school for which they had not expressed any
preference (Abdulkadiro glu, Pathak and Roth, 2005).
Moreover, the process was not incentive-compatible. Schools were more
likely to admit students who ranked them as number one. Therefore, if a
student was unlikely to be admitted to her favorite school, her best strategy
would be to list a more realistic option as her rst choice. In 2003, Roth
and his colleagues A. Abdulkadiro glu and P.A. Pathak helped re-design this
admissions process. The new system uses an applicant-proposing deferred-
acceptance algorithm, modied to accommodate regulations and customs of
New York City. This algorithm is incentive compatible for the applicants, i.e.,
it is optimal for them to report their preferences truthfully, and congestion
is eliminated. During the rst year of the new system, only about 3,000
students had to be matched with schools for which they had not expressed
a preference, a 90 percent reduction compared to previous years.
Prior to 2005, the Boston Public School system (BPS) used a clearing-
house algorithm known as the Boston mechanism. This type of algorithm
rst tries to match as many applicants as possible with their rst-choice
school, then tries to match the remaining applicants with their second-choice
school and so on (Abdulkadiro glu, Pathak, Roth and Snmez, 2005). Evi-
dently, if an applicants favorite school is very dicult to get accepted at,
with this type of mechanism it is best to list a less popular school as the
rst choice. This presented the applicants with a vexing strategic situa-
tion: to game the system optimally, they had to identify which schools were
realistic options for them. Applicants who simply reported their true prefer-
ences suered unnecessarily poor outcomes. Even if, miraculously, everyone
had found a best-response strategy, every Nash equilibrium would have been
schools are active players. In Boston, the schools are passive and priorities are determined
centrally.
27
Pareto dominated by the truthful equilibrium of the applicant-proposing
deferred-acceptance mechanism (Ergin and Snmez, 2006). Roth and his
colleagues, A. Abdulkadiro glu, P.A. Pathak and T. Snmez, were asked to
provide advice on the design of a new BPS clearinghouse algorithm. In 2005,
an applicant-proposing deferred-acceptance algorithm was adopted. Since it
is incentive-compatible for the applicants, the need for strategizing is elimi-
nated. Other school systems in the U.S. have followed New York and Boston
by adopting similar algorithms; a recent example is the Denver public school
system.
20
5.3 Kidney exchange
In important real-world situations, side-payments are ruled out on legal and
ethical grounds. For example, in most countries it is illegal to exchange
human organs, such as kidneys, for money. Organs have to be assigned to
patients who need transplants by some other method. Some patients may
have a willing kidney donor. For example, a husband may be willing to
donate a kidney to his wife. A direct donation is sometimes ruled out for
medical reasons, such as incompatibility of blood types. Still, if patient A
has a willing (but incompatible) donor A
0
, and patient B has a willing (but
incompatible) donor B
0
, then if A is compatible with B
0
and B with A
0
, an
exchange is possible: A
0
donates to B and B
0
to A. Such bilateral kidney
exchanges were performed in the 1990s, although they were rare.
Roth, Snmez and nver (2004) noted the similarity between kidney ex-
change and the Shapley-Scarf one-sided matching model described in Section
3.2, especially the version due to Abdulkadiro glu and Snmez (1999). One
important dierence is that, while all objects in the Shapley-Scarf model
can be assigned simultaneously, some kidney patients must be assigned to
a waiting list, in the hope that suitable kidneys become available in the
20
Many dierent kinds of matching procedures are used in various parts of the world.
Braun, Dwenger, Kbler and Westkamp (2012) describe the two-part procedure that the
German central clearinghouse uses to allocate admission to university medical studies and
related subjects. In the rst part, 20 percent of all available university seats are reserved
for applicants with very good grades, and 20 percent for those with the longest waiting
time since completing high school. These seats are allocated using the Boston mechanism.
In the second part, all remaining seats are allocated using a university-proposing deferred-
acceptance algorithm. The authors use laboratory experiments to study the incentives
to strategically manipulate this two-part procedure. More information on matching algo-
rithms in Europe can be found at http://www.matching-in-practice.eu/.
28
future.
21
Roth, Snmez and nver (2004) adapted the top-trading cycle
algorithm to allow for waiting-list options. The doctors indicate the most
preferred kidney, or the waiting-list option, for each patient. If there is a
cycle, kidneys are exchanged accordingly. For example, three patient-donor
pairs (A; A
0
), (B; B
0
) and (C; C
0
) may form a cycle, resulting in a three-way
exchange (A gets a kidney from B
0
, B from C
0
, and C from A
0
). The rules
allow for chains where, for example, A gets a kidney from B
0
while B is
assigned a high priority in the waiting list (and another patient can receive a
kidney from A
0
). Roth, Snmez and nver (2004) constructed ecient and
incentive-compatible chain selection rules.
A bilateral exchange between (A; A
0
) and (B; B
0
) requires a double co-
incidence of wants: A
0
must have what B needs while B
0
must have what
A needs. A clearinghouse with a database of patient-donor pairs that imple-
ments more complex multilateral exchanges can increase market thickness,
i.e., raise the number of possible transplants. This is especially important if
many highly sensitized patients are compatible with only a small number of
donors (Ashlagi and Roth, 2012). However, complex multilateral exchanges
may not be feasible due to logistical constraints. Roth, Snmez and nver
(2005b) showed how ecient outcomes with good incentive properties can
be found in computationally ecient ways when only bilateral exchanges are
feasible. But signicant gains can be achieved with exchanges involving three
patient-donor pairs (Saidman et al., 2006, Roth, Snmez, and nver, 2007).
A number of regional kidney exchange programs in the U.S. have in fact
moved towards more complex exchanges. The New England Program for
Kidney Exchange, founded by Roth, Snmez and nver, in collaboration with
Drs. Frank Delmonico and Susan Saidman, was among the early pioneers
(Roth, Snmez and nver, 2005a). Recently, interest has focused on long
chains involving altruistic donors, who want to donate a kidney but have no
particular patient in mind. Such chains suer less from logistical constraints,
because the transplants do not need to be conducted simultaneously (Roth
et al., 2006).
This work on kidney exchange highlights an important aspect of mar-
ket design. Specic applications often uncover novel problems, such as the
NRMPs couples problem, the priorities of school choice, or the waiting-list
21
The problem of kidney exchange is inherently more dynamic than the applications
discussed in Sections 5.1 and 5.2. Whereas residencies and public-school places can be
allocated once per year, there is no such obvious timing of kidney exchanges. This has led
to theoretical work on the optimal timing of transactions (nver, 2010).
29
and logistical problems of kidney exchange. These new problems stimulate
new theoretical research, which in turn leads to new applications, etc. Alvin
Roth has made signicant contributions to all parts of this iterative process.
6 Other Contributions
6.1 Lloyd Shapley
In non-cooperative game theory, Shapleys contributions include a number of
innovative studies of dynamic games. Aumann and Shapleys (1976) perfect
folk theorem shows that any feasible payo vector (where each player gets at
least the minimum amount he can guarantee for himself) can be supported
as a strategic equilibrium payo of a repeated game involving very patient
players. The theory of repeated games was generalized by Shapley (1953b),
who introduced the important notion of a stochastic game, where the actions
chosen in one period may change the game to be played in the future. This
has led to an extensive literature (e.g., Mertens and Neyman, 1981). Shapley
(1964) showed that a certain class of learning dynamics may not converge to
an equilibrium point, a result which has stimulated research on learning in
games. Shapley and Shubik (1977) is an important study of strategic market
games.
Lloyd Shapley is the most important researcher in the eld of cooper-
ative game theory. Shapley and Shubik (1969) characterized the class of
transferable-utility market games, and showed that such games have non-
empty cores. Shapley (1953a) introduced, and axiomatically characterized,
the main single-valued solution concept for coalitional games with transfer-
able utility, nowadays called the Shapley value. Shapley (1971) proved that
for convex games, the Shapley value occupies a central position in the core.
Harsanyi (1963) and Shapley (1969) extended the Shapley value to games
without transferable utility.
The Shapley value has played a major role in the development of cooper-
ative game theory, with a large variety of applications. Although originally
intended as a prediction of what a player could expect to receive from a
game, it is often given a normative interpretation as an equitable outcome,
for example, when costs are allocated by some administrative procedure (e.g.,
Young, 1994). The book by Aumann and Shapley (1974) contains extensions
of the major justications, interpretations and computations of the Shapley
30
value to games with innitely many players. This work has important appli-
cations to problems of cost allocation (e.g., Billera, Heath and Raanan, 1978).
The book also contains a version of Edgeworths conjecture: in certain large
markets, the Shapley value and the core both coincide with the competitive
equilibrium allocation. The Shapley value for coalitional political games is
known as the Shapley-Shubik power index (Shapley and Shubik, 1954). It
has been used, in particular, to evaluate power shifts caused by changes in
voting systems (e.g., Hosli, 1993).
6.2 Alvin Roth
The book by Roth and Sotomayor (1990) documents the state of two-sided
matching theory three decades ago, including many key results due to Roth
and coauthors. Among his other theoretical contributions, Roth (1977) char-
acterized the Shapley value as a risk-neutral utility function dened on the
space of coalitional games with transferable utility.
Roth (1991b) describes how laboratory experiments and eld observations
can interact with game theory, thereby establishing economics as a more sat-
isfactory empirical science. Through his own laboratory experiments, Alvin
Roth has greatly contributed to this research program. In a series of experi-
ments, Roth and his coauthors tested the predictions of cooperative bargain-
ing theory (Roth and Malouf, 1979, Roth, Malouf and Murnighan, 1981, Roth
and Murnighan, 1982, Murnighan, Roth and Schoumaker, 1988). Coopera-
tive bargaining models were found to correctly predict the qualitative eects
of changes in risk aversion. These tests were facilitated by a device of Roth
and Malouf (1979), who controlled for the subjects inherent risk-aversion
by using lottery tickets as rewards. By varying the information given to a
subject about another subjects payos, the experiments revealed the impor-
tance of focal-point eects and fairness concerns. A series of experiments by
Ochs and Roth (1989) tested the predictions of non-cooperative bargaining
models. This was followed by the important cross-cultural study of Roth,
Prasnikar, Okuno-Fujiwara and Zamir (1991) which investigated bargaining
behavior in four dierent countries.
Laboratory experiments often reveal that subjects change their behavior
over time. Roth and Erev (1995) developed a reinforcement learning model,
where players tend to repeat choices that produce good outcomes. This model
turned out to be consistent with actual behavior in a number of experimental
31
games. Slonim and Roth (1998) used this type of model to explain behavior
in a simple non-cooperative bargaining game, while Erev and Roth (1998)
showed that a reinforcement learning model can predict behavior ex ante
(rather than merely explaining it ex post). This inuential series of articles
has shown that the explanatory and predictive power of game theory can be
increased if realistic cognitive limitations are taken into account.
7 Conclusion
Lloyd Shapley has led the development of cooperative game theory. His work
has not only strengthened its theoretical foundations, but also enhanced the
theorys usefulness for applied work and policy making. In collaboration with
D. Gale, H. Scarf and M. Shubik, he created the theory of matching markets.
Launching the theory, Gale and Shapley (1962) expressed the hope that one
day it would have practical applications. This hope has been fullled by the
emerging literature on market design.
The work by Alvin Roth has enhanced our understanding of how markets
work. Using empirical, experimental and theoretical methods, Roth and
his coauthors, including A. Abdulkadiro glu, P.A. Pathak, T. Snmez and
M.U. nver, have studied the institutions that improve market performance,
thereby illuminating the need for stability and incentive compatibility. These
contributions led directly to the successful redesign of a number of important
real-world markets.
For further reading An elementary introduction to cooperative game
theory can be found in Moulin (1995), while Shubik (1984) oers a more
advanced treatment. Serranos (2009) survey emphasizes the core and the
Shapley value, while Maschler (1992) discusses alternative cooperative solu-
tion concepts. For introductions to matching theory, see Roth and Sotomayor
(1990) or the original article by Gale and Shapley (1962). For general as-
pects of market design, see Roth (2002) and (2008b). Roth (2008a) discusses
the history, theory and practical aspects of deferred-acceptance algorithms.
Snmez and nver (2011) provide a detailed technical survey of the design
of matching markets. For the most recent developments in market design,
see Alvin Roths blog, http://marketdesigner.blogspot.com/.
32
References
[1] Abdulkadiro glu, A. and T. Snmez. 1999. House allocation with exist-
ing tenants. Journal of Economic Theory 88: 233260
[2] Abdulkadiro glu, A. and T. Snmez. 2003. School choice: A mechanism
design approach. American Economic Review 93: 72947.
[3] Abdulkadiro glu, A., P.A. Pathak and A.E. Roth. 2005. The New York
City high school match. American Economic Review 95: 36467.
[4] Abdulkadiro glu, A., P.A. Pathak, A.E. Roth and T. Snmez. 2005. The
Boston public school match. American Economic Review 95: 36871.
[5] Ashlagi, I. and A.E. Roth 2012. New challenges in multi-hospital kidney
exchange. American Economic Review 102: 354-359.
[6] Aumann, R.J. and L.S. Shapley. 1974. Values of Non-atomic Games.
Princeton: Princeton University Press.
[7] Aumann, R.J. and L.S. Shapley. 1976. Long term competition - a game
theoretic analysis. Mimeo.
[8] Balinski, M. and T. Snmez. 1999. A tale of two mechanisms: student
placement. Journal of Economic Theory 84: 73-94.
[9] Berl, J.E., R.D. McKelvey, P.C. Ordeshook and M.D. Winer. 1976. An
experimental test of the core in a simple n-person cooperative nonsid-
epayment game. Journal of Conict Resolution 20: 453-479.
[10] Billera, L. 1970. Some theorems on the core of an n-person game with-
out side-payments. SIAM Journal on Applied Mathematics 18: 567-
579.
[11] Billera, L., D. Heath and J. Raanan. 1978. Internal telephone billing
rates - A novel application of non-atomic game theory. Operations Re-
search 26: 956-965.
[12] Bondareva, O. N. 1963. Some applications of linear programming to
the theory of cooperative games. Problemy Kibernetiki 10: 11939 [in
Russian]. English translation in Selected Russian Papers in Game The-
ory 19591965. Princeton: Princeton University Press, 1968.
33
[13] Braun, S., N. Dwenger, D. Kbler and A. Westkamp. 2012. Implement-
ing quotas in university admissions: An experimental analysis. Working
paper, Max Planck Institute for Tax Law and Public Finance, Munich.
[14] Bulow, J. and J. Levin. 2006. Matching and price competition. Amer-
ican Economic Review 96: 652-668.
[15] Chen, Y. and T. Snmez. 2006. School choice: An experimental study.
Journal of Economic Theory 127: 2002-2031.
[16] Crawford V.P. 2008. The exible-salary match: A proposal to increase
the salary exibility of the National Resident Matching Program. Jour-
nal of Economic Behavior and Organization 66: 149160.
[17] Crawford V.P. and E.M. Knoer. 1981. Job matching with heterogeneous
rms and workers. Econometrica 49: 437-450.
[18] Debreu, G. and H. Scarf 1963. A limit theorem on the core of an econ-
omy. International Economic Review 4: 234246.
[19] Demange, G., D. Gale and M. Sotomayor. 1986. Multi-item auctions.
Journal of Political Economy 94: 863-872.
[20] Dubins, L.E. and D.A. Freedman. 1981. Machiavelli and the Gale-
Shapley algorithm. American Mathematical Monthly 88: 485-494.
[21] Edelman, B., M. Ostrovsky and M. Schwarz. 2007. Internet advertising
and the generalized second-price auction: Selling billions of dollars
worth of keywords. American Economic Review 97: 242259.
[22] Edgeworth, F. Y. 1881. Mathematical Psychics. London: C. Kegan
Paul & Co.
[23] Erev, I. and A.E. Roth. 1998. Predicting how people play games: Rein-
forcement learning in experimental games with unique, mixed strategy
equilibria. American Economic Review 88: 848-881.
[24] Ergin, H. and T. Snmez. 2006. Games of school choice under the
Boston mechanism. Journal of Public Economics 90: 215-237.
34
[25] Featherstone, C. and M. Niederle. 2011. School choice mechanisms un-
der incomplete information: An experimental investigation. Working
Paper, Harvard Business School, Cambridge, MA.
[26] Gale, D. and L.S. Shapley. 1962. College admissions and the stability
of marriage. American Mathematical Monthly 69: 915.
[27] Grdenfors, P. 1975. Match making: assignments based on bilateral
preferences. Behavioral Science 20: 166173.
[28] Gillies, D.B. 1953a. Location of solutions. In Report of an Informal
Conference on the Theory of n-Person Games held at Princeton Uni-
versity March 20-21, 1953: 1112. Mimeo, Department of Mathemat-
ics, Princeton University.
[29] Gillies, D.B. 1953b. Some Theorems on n-Person Games. Ph. D. The-
sis, Department of Mathematics, Princeton University (June 1953).
[30] Gillies, D.B. 1959. Solutions to general non-zero sum games. In Contri-
butions to the Theory of Games IV. Edited by A.W. Tucker and R.D.
Luce. Princeton, N.J.: Princeton University Press.
[31] Harsanyi, J.C. 1963. A simplied bargaining model for the n-person
cooperative game. International Economic Review 4: 194220.
[32] Hateld, J. and P. Milgrom. 2005. Matching with contracts. American
Economic Review 95: 91335.
[33] Hosli, M. 1993. Admission of European Free Trade Association states
to the European Community: eects on voting power in the European
Community Council of Ministers. International Organization 47: 629-
643.
[34] Huang, C.Y. and T. Sjstrm. 2003. Consistent solutions for coopera-
tive games with externalities. Games and Economic Behavior 43:196-
213.
[35] Immorlica, N. and M. Mahdian. 2005. Marriage, honesty, and stabil-
ity. In Proceedings of the Sixteenth Annual ACM-SIAM Symposium on
Discrete Algorithms: 53-62. Philadelphia: Society for Industrial and
Applied Mathematics.
35
[36] Kagel, J.H. and A.E. Roth. 2000. The dynamics of reorganization in
matching markets: A laboratory experiment motivated by a natural
experiment. Quarterly Journal of Economics 115: 20135.
[37] Kelso, A.S. and V.P. Crawford. 1982. Job matching, coalition forma-
tion, and gross substitutes. Econometrica 50: 1483504.
[38] Knuth, D. 1976. Mariages Stables. Montral: Les Presses de
lUniversit de Montral.
[39] Kczy, L. 2007. A recursive core for partition function form games.
Theory and Decision 63: 41-51.
[40] Kojima, F. and P.A. Pathak. 2009. Incentives and stability in large
two-sided matching markets. American Economic Review 99: 60827.
[41] Ma, J. 1994. Strategy-proofness and the strict core in a market with
indivisibilities. International Journal of Game Theory 23: 75-83.
[42] Ma, J. 1995. Stable matchings and rematching-proof equilibria in a
two-sided matching market. Journal of Economic Theory 66: 352369.
[43] Maschler, M. 1992. The bargaining set, kernel, and nucleolus. In Hand-
book of Game Theory with Economic Applications, Vol. 1. Edited by
R. J. Aumann and S. Hart. Elsevier Science.
[44] McKinney, C.N., M. Niederle and A.E. Roth. 2005. The collapse of a
medical labor clearinghouse (and why such failures are rare). American
Economic Review 95: 878-889.
[45] Mertens, J.-F. and A. Neyman. 1981. Stochastic games. International
Journal of Game Theory 10: 53-66.
[46] Moulin, H. 1995. Cooperative Microeconomics. Princeton University
Press.
[47] Murnighan, J.K., A.E. Roth and F. Schoumaker. 1988. Risk aversion
in bargaining: An experimental study. Journal of Risk and Uncertainty
1: 101-124.
[48] Niederle, M., D. Proctor and A.E. Roth. 2008. The gastroenterology
fellowship match: The rst two years. Gastroenterology 135: 344-346.
36
[49] Niederle, M., A.E. Roth and T. Snmez. 2008. Matching and Mar-
ket Design. In New Palgrave Dictionary of Economics, second edi-
tion. Edited by S. N. Durlauf and L. E. Blume. Basingstoke: Palgrave
Macmillan.
[50] Ochs, J. and A.E. Roth. 1989. An experimental study of sequential
bargaining. American Economic Review 79: 355-384.
[51] Pais, J. and . Pintr. 2008. School choice and information: An experi-
mental study on matching mechanisms. Games and Economic Behavior
64: 303328.
[52] Roth, A.E. 1977. The Shapley value as a von Neumann- Morgenstern
utility. Econometrica 45: 657-664.
[53] Roth, A.E. 1982a. The economics of matching: Stability and incentives.
Mathematics of Operations Research 7: 617-628.
[54] Roth, A.E. 1982b. Incentive compatibility in a market with indivisibil-
ities. Economics Letters 9: 127-132.
[55] Roth, A.E. 1984a. The evolution of the labor market for medical in-
terns and residents: A case study in game theory. Journal of Political
Economy 92: 9911016.
[56] Roth, A.E. 1984b. Misrepresentation and stability in the marriage
problem. Journal of Economic Theory 34: 383387.
[57] Roth, A.E. 1984c. Stability and polarization of interests in job match-
ing. Econometrica 52: 47-57.
[58] Roth, A.E. 1985a. The college admissions problem is not equivalent to
the marriage problem. Journal of Economic Theory 36: 277288
[59] Roth, A.E. 1985b. Conict and coincidence of interest in job match-
ing: Some new results and open questions. Mathematics of Operations
Research 10: 379-389
[60] Roth, A.E. 1989. Two-sided matching with incomplete information
about others preferences. Games and Economic Behavior 1: 191209.
37
[61] Roth, A.E. 1990. New physicians: A natural experiment in market
organization. Science 250: 152428.
[62] Roth, A.E. 1991a. A natural experiment in the organization of entry
level labor markets: Regional markets for new physicians and surgeons
in the U.K. American Economic Review 81: 41540.
[63] Roth, A.E. 1991b. Game theory as a part of empirical economics. Eco-
nomic Journal 101: 107-114
[64] Roth, A.E. 2002. The economist as engineer: Game theory, experimen-
tal economics and computation as tools of design economics. Econo-
metrica 70: 134178.
[65] Roth, A.E. 2007. Repugnance as a constraint on markets. Journal of
Economic Perspectives 21: 37-58.
[66] Roth, A.E. 2008a. Deferred acceptance algorithms: History, theory,
practice, and open questions. International Journal of Game Theory
36: 537569.
[67] Roth, A.E. 2008b. What have we learned from market design? Eco-
nomic Journal 118: 285310.
[68] Roth, A.E. and I. Erev. 1995. Learning in extensive-form games: Ex-
perimental data and simple dynamic models in the intermediate term.
Games and Economic Behavior 8: 164-212.
[69] Roth, A.E. and M. Malouf. 1979. Game-theoretic models and the role
of information in bargaining. Psychological Review 86: 574-594
[70] Roth, A.E. and J.K. Murnighan. 1982. The role of information in bar-
gaining: An experimental study. Econometrica 50: 1123-1142.
[71] Roth, A.E. and E. Peranson. 1999. The redesign of the matching market
for American physicians: Some engineering aspects of economic design.
American Economic Review 89: 74879.
[72] Roth, A.E. and A. Postlewaite. 1977. Weak versus strong domination
in a market with indivisible goods. Journal of Mathematical Economics
4: 131-137.
38
[73] Roth, A.E. and U. Rothblum. 1999. Truncation strategies in matching
markets: In search of advice for participants. Econometrica 67: 21-43.
[74] Roth, A.E. and M. Sotomayor. 1990. Two-sided Matching: A Study
in Game-theoretic Modeling and Analysis. Econometric Society Mono-
graph Series, Cambridge: Cambridge University Press.
[75] Roth, A.E. and J.H. Vande Vate. 1990. Random paths to stability in
two-sided matching. Econometrica 58: 1475-1480.
[76] Roth, A.E. and X. Xing. 1994. Jumping the gun: Imperfections and
institutions related to the timing of market transactions. American
Economic Review 84: 992-1044.
[77] Roth, A.E. and X. Xing. 1997. Turnaround time and bottlenecks in
market clearing: Decentralized matching in the market for clinical psy-
chologists. Journal of Political Economy 105: 284329.
[78] Roth, A.E., M. Malouf and J.K. Murnighan. 1981. Sociological ver-
sus strategic factors in bargaining. Journal of Economic Behavior and
Organization 2: 153-177.
[79] Roth, A.E., T. Snmez and M.U. nver. 2004. Kidney exchange. Quar-
terly Journal of Economics 119: 45788.
[80] Roth, A.E., T. Snmez and M.U. nver. 2005a. A kidney exchange
clearinghouse in New England. American Economic Review 95: 376
80.
[81] Roth, A.E., T. Snmez and M.U. nver. 2005b. Pairwise kidney ex-
change. Journal of Economic Theory 125: 15188.
[82] Roth, A.E., T. Snmez and M.U. nver. 2007. Ecient kidney ex-
change: Coincidence of wants in markets with compatibility-based pref-
erences. American Economic Review 97: 82851.
[83] Roth, A.E., V. Prasnikar, M. Okuno-Fujiwara and S. Zamir. 1991.
Bargaining and market behavior in Jerusalem, Ljubljana, Pittsburgh,
and Tokyo: an experimental study. American Economic Review 81:
1068-1095.
39
[84] Roth, A.E., T. Snmez, M.U. nver, F.L. Delmonico and S.L. Said-
man. 2006. Utilizing list exchange and undirected Good Samaritan do-
nation through chain paired kidney donations. American Journal of
Transplantation 6: 2694705.
[85] Saidman, S.L., A.E. Roth, T. Snmez, M.U. nver, and F.L. Del-
monico. 2006. Increasing the opportunity of live kidney donation by
matching for two and three way exchanges. Transplantation 81: 773
82.
[86] Scarf, H. 1967. The core of an n-person game. Econometrica 35: 50-69.
[87] Serrano, R. 2009. Cooperative games: core and Shapley value. In En-
cyclopedia of Complexity and Systems Science. Edited by R. Meyers.
New York: Springer.
[88] Shapley, L.S. 1953a. A value for n-person games. In Contributions to
the Theory of Games, Vol. 2. Edited by H. W. Kuhn and A. W. Tucker.
Princeton, NJ: Princeton University Press.
[89] Shapley, L.S. 1953b. Stochastic games. Proceedings of the National
Academy of Sciences of the USA 39: 10951100.
[90] Shapley, L.S. 1953c. Open questions. In Report of an Informal Confer-
ence on the Theory of n-Person Games held at Princeton University
March 20-21, 1953: 15. Mimeo, Department of Mathematics, Prince-
ton University.
[91] Shapley, L.S. 1955. Markets as cooperative games. Rand Corporation
Memorandum P-629, Santa Monica, CA.
[92] Shapley, L.S. 1964. Some topics in two-person games. In Advances in
Game Theory. Edited by M. Dresher, L. S. Shapley, and A. W. Tucker.
Princeton, NJ: Princeton University Press.
[93] Shapley, L.S. 1967. On balanced sets and cores. Navel Research Logis-
tics Quarterly 9: 458.
[94] Shapley, L.S. 1969. Utility comparison and the theory of games. In
La dcision: Agrgation et dynamique des ordres de prfrence. Paris:
Editions du CNRS.
40
[95] Shapley, L.S. 1971. Cores of convex games. International Journal of
Game Theory 1: 11-26.
[96] Shapley, L.S. and H. Scarf. 1974. On cores and indivisibility. Journal
of Mathematical Economics 1: 2337.
[97] Shapley, L.S. and M. Shubik. 1954. A method of evaluating the dis-
tribution of power in a committee system. American Political Science
Review 48: 787-792
[98] Shapley, L.S. and M. Shubik. 1969. On market games. Journal of Eco-
nomic Theory 1: 925.
[99] Shapley, L.S. and M. Shubik. 1971. The assignment game I: The core.
International Journal of Game Theory 1: 11130.
[100] Shapley, L.S. and M. Shubik. 1977. Trade using one commodity as a
means of payment. Journal of Political Economy 85: 937-968.
[101] Shubik, M. 1959. Edgeworth market games. In Contributions to the
Theory of Games IV. Edited by R.D. Luce and A.W. Tucker. Princeton,
N.J.: Princeton University Press.
[102] Shubik, M. 1984. Game Theory in the Social Sciences. MIT Press.
[103] Slonim, R. and A.E. Roth. 1998. Learning in high stakes ultimatum
games: an experiment in the Slovak Republic. Econometrica 66: 569-
596.
[104] Snmez, T. and M.U. nver. 2011. Matching, allocation, and exchange
of discrete resources. In Handbook of Social Economics. Edited by J.
Benhabib, A. Bisin, and M. Jackson. The Netherlands: North-Holland.
[105] nver, M.U. 2005. On the survival of some unstable two-sided matching
mechanisms. International Journal of Game Theory 33: 239254.
[106] nver, M.U. 2010. Dynamic kidney exchange. Review of Economic
Studies 77: 372414.
[107] Varian, H. 2007. Position auctions. International Journal of Industrial
Organization 25:11631178.
41
[108] von Neumann, J. and O. Morgenstern. 1944. Theory of Games and
Economic Behavior. Princeton, N.J.: Princeton University Press.
[109] Young, H.P., 1994. Cost allocation. In Handbook of Game Theory with
Economic Applications, Vol. 2. Edited by R.J. Aumann and S. Hart.
Amsterdam: Elsevier Science.
42
10 OCTOBER 2011
Scientic Background on the Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2011
EMPIRICAL MACROECONOMICS
Empirical Macroeconomics
One of the main tasks for macroeconomists is to explain how macroeco-
nomic aggregates such as GDP, investment, unemployment, and ination
behave over time. How are these variables aected by economic policy and
by changes in the economic environment? A primary aspect in this analy-
sis is the role of the central bank and its ability to inuence the economy.
How eective can monetary policy be in stabilizing unwanted uctuations in
macroeconomic aggregates? How eective has it been historically? Similar
questions can be raised about scal policy. Thomas J. Sargent and Christo-
pher A. Sims have developed empirical methods that can answer these kinds
of questions. This years prize recognizes these methods and their successful
application to the interplay between monetary and scal policy and economic
activity.
In any empirical economic analysis based on observational data, it is
dicult to disentangle cause and eect. This becomes especially cumbersome
in macroeconomic policy analysis due to an important stumbling block: the
key role of expectations. Economic decision-makers form expectations about
policy, thereby linking economic activity to future policy. Was an observed
change in policy an independent event? Were the subsequent changes in
economic activity a causal reaction to this policy change? Or did causality
run in the opposite direction, such that expectations of changes in economic
activity triggered the observed change in policy? Alternative interpretations
of the interplay between expectations and economic activity might lead to
very dierent policy conclusions. The methods developed by Sargent and
Sims tackle these diculties in dierent, and complementary, ways. They
have become standard tools in the research community and are commonly
used to inform policymaking.
Background Prior to the 1970s, expectations played, at best, a rudimen-
tary role in the analysis of macroeconomic outcomes. Following the seminal
work by Milton Friedman, Robert Lucas, Edmund Phelps, and others, it
became necessary to systematically incorporate expectations not only into
macroeconomic theory, but also and more importantly into its empirical
implementation. This was a major obstacle, however. At the time, formal
methods were simply not available to identify and analyze exogenous shocks,
1
as a means of evaluating macroeconomic theories that feature active for-
mation of expectations.
Sargent and Sims have both made seminal contributions that allow re-
searchers to specify, empirically implement, and evaluate dynamic models of
the macroeconomy with a central role for expectations. Their subsequent
work, from the initial papers until today, has delivered many extensions,
renements, and powerful applications. The contributions by Sargent and
Sims have generated literatures of methodological research and applied stud-
ies within the research community as well as the policymaking community.
Prior to the formative research by Sargent and Sims, the predominant
empirical method in macroeconomics was to statistically estimate a large
linear system, typically built around the Keynesian macroeconomic model.
This estimated system was then used to interpret macroeconomic time se-
ries, to forecast the economy, and to conduct policy experiments. Such large
models were seemingly successful in accounting for historical data. However,
during the 1970s most western countries experienced high rates of ination
combined with slow output growth and high unemployment. In this era of
stagation, instabilities appeared in the large models, which were increas-
ingly called into question.
Sargent structural econometrics Sargent began his research around
this time, during the period when an alternative theoretical macroeconomic
framework was proposed. It emphasized rational expectations, the notion
that economic decisionmakers like households and rms do not make system-
atic mistakes in forecasting. This framework turned out to be essential in
interpreting the ination-unemployment experiences of the 1970s and 1980s.
It also formed a core of newly emerging macroeconomic theories.
Sargent played a pivotal role in these developments. He explored the
implications of rational expectations in empirical studies, by showing how
rational expectations could be implemented in empirical analyses of macro-
economic events so that researchers could specify and test theories using
formal statistical methods and by deriving implications for policymaking.
He also advanced and applied broader views on expectations formation, such
as gradual learning. Sargents contributions to rational-expectations econo-
metrics were purely methodological. Specically, his methods for character-
izing and structurally estimating macroeconomic models with microeconomic
foundations broke new ground in allowing researchers to uncover the deep
2
underlying model parameters and perform hypothesis testing. In a broader
perspective, Sargent also raised important points of immediate policy rele-
vance. For example, his early studies of linkages between scal and monetary
policy still guides policymakers still today.
Sims VARs Sims launched what was perhaps the most forceful critique
of the predominant macroeconometric paradigm of the early 1970s by fo-
cusing on identication, a central element in making causal inferences from
observed data. Sims argued that existing methods relied on incredible
identication assumptions, whereby interpretations of what causes what
in macroeconomic time series were almost necessarily awed. Misestimated
models could not serve as useful tools for monetary policy analysis and, often,
not even for forecasting.
As an alternative, Sims proposed that the empirical study of macroeco-
nomic variables could be built around a statistical tool, the vector autoregres-
sion (VAR). Technically, a VAR is a straightforward N-equation, N-variable
(typically linear) system that describes how each variable in a set of macro-
economic variables depends on its own past values, the past values of the
remaining N 1 variables, and on some exogenous shocks. Simss insight
was that properly structured and interpreted VARs might overcome many
identication problems and thus were of great potential value not only for
forecasting, but also for interpreting macroeconomic time series and conduct-
ing monetary policy experiments.
Over the last past three decades, the VAR methodology has been devel-
oped signicantly in various directions, and Sims himself remained at the
forefront. As a result, VARs are now in ubiquitous use, among empirical
academic researchers and policymaking economists alike. By now, VARs
also constitute a major research tool in many areas outside of monetary eco-
nomics.
Applications Both Sargent and Sims have used their own methods in in-
uential applications concerning the determinants and eects of monetary
policy. In a series of contributions, Sargent has analyzed episodes of very high
ination, or hyperination. He explored the high ination in the U.S. during
the 1970s and subsequent changes that brought about a rapid, and seemingly
permanent, fall in ination. In this analysis, Sargent found that learning
a departure from fully rational expectations is important in order to un-
3
derstand how ination came and went. In fact, the dening characteristic of
Sargents overall approach is not an insistence on rational expectations, but
rather the essential idea that expectations are formed actively, under either
full or bounded rationality. In this context, active means that expectations
react to current events and incorporate an understanding of how these events
aect the economy. This implies that any systematic change in policymaking
will inuence expectations, a crucial insight for policy analysis.
Sims has also carried out many applied studies, to some extent on the very
same topics, i.e., the extent and implications of monetary policy shifts in the
U.S. His main focus, however, has been on the identication of unexpected
policy changes and their eects on economic activity. With regard to the
crucial distinction between expected and unexpected, Simss method oers a
means of separating expected from unexpected policy changes as drivers of
macroeconomic variables. His method has gained broad acceptance and has,
for example, allowed us to establish how unexpected monetary policy changes
lead to immediate eects on some macroeconomic variables but only slow,
and hump-shaped, eects on others. Indeed, some of the most inuential
studies of this issue were undertaken by Sims himself.
Later work Sargent and Sims are actively pursuing new research aimed
at further understanding expectations formation and its role in the economy.
Sargents focus here is on exploring a class of mechanisms for expectations
formation based on robust control, which captures the idea that the decision-
maker has an imperfect understanding of how the economy works. Similarly,
Simss most recent work explores a parallel, new theory for expectations for-
mation based on rational inattention, which captures agents limited capacity
to process information.
Relations between the two strands of work Although Sargents and
Simss empirical methodologies certainly dier, they complement one an-
other and are often used in conjunction. In fact, the dynamic behavior
of a Sargent-style structural model with rational expectations can often be
represented as a Sims-style VAR. Identication of such a VAR would then
directly correspond to identication of the structural parameters estimated
along the lines of rational-expectations econometrics. A key aspect of VAR
methodology, so-called impulse-response analysis, describes how fundamen-
tal shocks propagate through the macroeconomy. It has become a leading
4
method for describing and analyzing structural macroeconomic models. Con-
versely, VAR identication is often made with specic reference to structural
models, although such structural VAR identication typically refers to
classes of models rather than a specic model. Which of these approaches
is followed in a specic application depends on the purpose. Structural es-
timation is straightforwardly implemented with modern computers and is
particularly useful for analyzing policy regimes. VAR analysis, which relies
on fewer and less specic theoretical assumptions, is mainly used for identify-
ing what policy shocks have occurred and their likely eects, absent a change
in policy regime. The methods developed by Sargent and Sims thus comprise
a methodological core in modern empirical analyses of macroeconomic policy
and economic activity.
Outline of this survey In what follows, Sargents and Simss key con-
tributions are described separately in Sections 1 and 2. Section 3 provides
an elaborate example of how the methods are interrelated, while also allow-
ing for a more precise description of Sargents methodology. Section 4 oers
a brief account of the aforementioned work on robust control and rational
inattention. Section 5 concludes.
1 Structural Estimation and Active Expecta-
tions: Contributions of Thomas J. Sargent
1.1 Historical context and impact on current work
Sargents research began at a time when a group of economists among them
the previous laureates Lucas, Phelps, and Prescott had upset the prevailing
macroeconomic paradigm based on reduced-form models. They proposed a
new macroeconomics where expectations would play a pivotal role. More
generally, they insisted on the need to rebuild macroeconomic theory and
empirical methodology. The new theory would be based on models with
microeconomic foundations, i.e., a theory of key economic decisions that
would be invariant to changes in policy. The new empirical methodology
would be based on estimation of the structural parameters in these mod-
els, for example parameters describing individual preferences and production
functions. Sargent was a highly inuential and distinguished member of this
rational-expectations group.
5
As a result of this new research program, macroeconomics changed course
rather drastically. Views on policy were transformed, and the awards to Lu-
cas and Phelps were largely motivated by the policy implications of their
work. Views on business cycles also changed, due to the contributions of
Lucas and Sargent as well as later work by Kydland and Prescott. More
recently, Keynesian ideas Keynesian ideas have been revived in a New Key-
nesian Macroeconomics, which builds directly on Kydland and Prescott,
with the addition of various frictions such as sticky prices and wages. Mod-
ern empirical macroeconomic research relies heavily on structural-estimation
methods, of which Sargent is the main architect.
1.2 Empirical methods and early applications
In the early to mid- 1970s, Sargent wrote a number of highly inuential
papers, where he showed how rational expectations implied a radical reinter-
pretation of empirical macroeconomic phenomena and rendered invalid con-
ventional statistical tests of macroeconomic relations, such as the Friedman-
Phelps hypothesis on the Phillips curve. Taken together, these papers had
a profound impact on central hypotheses about the role of monetary policy
and the Phillips-curve tradeo.
Compared to other researchers at the time, Sargent focused more on
actual data and on ways to evaluate theory by taking active expectations
formation into account. He was thus able to show why earlier tests had gone
wrong and how new, more accurate, tests could be constructed.
Sargents general approach was to formulate, solve, and estimate a struc-
tural macroeconomic model with microeconomic foundations, i.e., a system
where all parameters, except those describing policy, are invariant to policy
interventions. Once its parameters are estimated, the model can be used as a
laboratory for analyzing policy experiments. (See Section 3 for a detailed
description of how this is carried out in the context of a specic example.)
Empirical applications Sargent combined the development of general
methods with concrete empirical applications. In a series of papers, he helped
to construct what later became an indispensable empirical method in modern
macroeconomics.
In a very early contribution which predated Lucass work on the same
topic Sargent (1971) demonstrated the crucial role of expectations in econo-
metric studies of the Phillips curve. A vital question in such studies had been
6
whether the long-run Phillips curve is vertical or has some (negative) slope,
as does the short-run curve. Sargent demonstrated that the usual econo-
metric tests relied critically on regarding expectations as passive, and that
the forward-looking nature of rational expectations implied that expectations
themselves depend on the slope of the long-run Phillips curve. This made
earlier rejections of a vertical Phillips curve invalid and indicated that the
curve could well be vertical. Since Sargents article, other eorts have been
made to pin down the slope of the long-run Phillips curve; see, e.g., Gal and
Gertler (1999) for a discussion.
Sargent (1973) carried out the rst successful econometric estimation un-
der rational expectations. It was based on a simple but complete model of the
U.S. economy, which embedded Irving Fishers theory that nominal interest
rates should increase one for one with expected ination. Sargent showed
that a test of Fishers theory also amounted to a test of the natural-rate hy-
pothesis. His econometric evidence implied rejection of the theory (although
with rather low statistical signicance). This paper served as a model for all
empirical applications to follow.
Sargent (1976) constructed and estimated an econometric model of the
U.S. economy subject to both real and nominal shocks. The estimation
results in this study gave the rst indications that models with real shocks,
to be studied later by Kydland and Prescott, might be empirically successful.
Further important contributions by Sargent in the area of macroeco-
nomics include Sargent (1978a) and the joint paper with Hansen and Sargent
(1980). In the area of forecasting, Sargent and Sims (1977) developed in-
dex modeling, later extended by Quah and Sargent (1993). Sargent (1989)
showed how ltering methods could be used to estimate linear rational-
expectations models in the presence of error-ridden data.
Among Sargents most recent applied contributions, Sargent and Surico
(2011) is noteworthy in showing how standard tests of the quantity theory of
money a pillar of the monetarist view are sensitive to the past monetary
policy regime. Using structural estimation as well as VARs, the authors argue
that apparent breakdowns of the quantity theory can instead be explained
by changes in the policy regime.
Many of the principal methods are collected in Sargents two monographs
Macroeconomic Theory (1979) and Dynamic Macroeconomic Theory (1987),
and in Rational Expectations and Econometric Practice (1981), a volume co-
edited by Lucas and Sargent. These publications have constituted required
reading for generations of macroeconomic researchers.
7
1.3 Policy implications
From the perspective of macroeconomic theory, the nature of expectations is
crucial for the eectiveness of various forms of economic policy. An important
part of Sargents work explores the restrictions rational expectations place
on policymakers. Sargent and Wallace (1976) and a series of papers that
followed, showed how merely replacing adaptive expectations with rational
expectations dramatically altered the policy implications of then-standard
macroeconomic models. Other work in this series include Sargent and Wal-
lace (1973, 1975). As these papers were cast in traditional macroeconomic
models, they added signicantly to Lucass (1972, 1973) seminal work on ex-
pectations and monetary policy, which had departed from traditional macro-
economics in its aim to develop a new theory of ination-output correlations.
Sargent and Wallace (1981) explored the connections between scal pol-
icy and monetary policy. They argued that monetary and scal policy were
inexorably linked, thereby demonstrating how Friedmans assertion that in-
ation is always and everywhere a monetary phenomenon can be quite
misleading. As the paper shows, scal policy may force monetary policy
to become highly inationary. The basic argument is that monetary pol-
icy generates seigniorage, i.e., real revenue that contributes to government
nancing, and that this seigniorage may become necessary in the wake of
large budget decits. Thus, current scal decits may require higher future
ination in order for the intertemporal budget to balance.
1
Sargents analysis of how monetary and scal policy inuences real ac-
tivity has also guided empirical work, where researchers have had to deal
with the Lucas critique, i.e., how the eects of policy change can be studied
using historical data. Lucas argued that this would require identifying deep
structural parameters, which traditional macroeconometric techniques did
not allow for. In response to this argument, Sargent and Sims have pursued
dierent, but complementary paths. Sargent has focused precisely on iden-
tifying structural parameters, while Sims has focused on ways of isolating
the eects of policy shocks without estimating deep structural parameters.
1
The argument is closely related to recent suggestions, by e.g., Woodford (1994) and
Sims (1994), that scal policy can also be a determinant of the current level of prices,
again through the government budget constraint. The scal theory of the price level
argues that the nominal price level adjusts so that the real value of scal authorities
initial nominal debt clears the intertemporal government budget (without a need for ina-
tion/seigniorage nance). While the unpleasant monetarist arithmetic is broadly accepted,
the scal theory of the price level remains controversial.
8
(These issues are further discussed in the context of the simple example in
Section 3.)
1.4 Episodes of high ination
Sargent has pursued important research on episodes with high ination, es-
pecially hyperination. This work includes an early paper The Ends of Four
Big Inations (1983), which analyzes historical hyperinations in Europe.
It may also be found in his book, The Conquest of American Ination (2001),
which studies how ination in the United States rose in the 1970s and then
gradually fell, as illustrated in Figure 1. The book emphasizes learning and
less than fully rational expectations (adaptive expectations are considered as
well and turn out to be important). It builds on Sargents earlier joint work
with Albert Marcet, e.g., Marcet and Sargent (1989a,b).
2
2
In the 1980s, a series of papers by Sargent and Marcet, one of Sargents students,
explored learning in macroeconomic analyses. Learning embodies two elements: (i) agents
incomplete knowledge of some model parameters, and (ii) a specication of how agents
learn about these parameters, based on observations of evolving time-series data. Marcet
and Sargent made reference to a literature in economic theory on how economic agents
learn under dierent circumstances.
Their contribution was to specify a plausible learning mechanismtypically least-
squares learning, by running OLS regressionsand explore its implications. Learning can
thus be viewed as a model of how expectations are formed. A central question becomes
whether such endogenous formation of expectations naturally tends toward rationality, i.e.,
full knowledge of the model. A subsequent literature developed to explore this question
in a variety of settings (see Evans and Honkapohja, 2001 for an overview). Learning has
not yet become a standard part of modern macroeconomic models, but there is increasing
recognition of its importance and the number of applications is growing.
9
Figure 1. Ination in the U.S., 1950-1995. The y axis
displays ination in percent and the x axis displays years.
A recent contribution by Sargent, Williams, and Zha (2006) is a good
illustration of Sargents views and methods for analyzing the dynamics of
ination. This paper considers a monetary authority that explicitly maxi-
mizes consumer welfare and has specic beliefs about the Phillips curve. In
particular, the central bank does not believe in an expectations-augmented
Phillips curve, but in a time-varying curve, while the private sector has ra-
tional expectations, so the true curve is indeed expectations-augmented.
The authors estimate this model structurally using a Bayesian Markov chain
Monte Carlo method and nd that it ts the data quite well. Their esti-
mates suggest that the central bank was initially fooled by an incorrect
belief about the Phillips curve, which led to a gradual increase in the ina-
tion rate. But the sequence of shocks in the 1970s, along with a revision
of the central banks beliefs, generated a subsequent fall in ination. Quite
10
surprisingly, the models forecasting ability outperforms that of advanced
atheoretical forecasting models (Bayesian VARs). What really occurred dur-
ing the 1970s and how the lessons from these experiences can be exploited
in modern policymaking remains a subject of considerable debate. Despite
this, Sargents historical interpretations are important benchmarks against
which current research has to measure up.
2 The Analysis of Macroeconomic Shocks:
Contributions of Christopher A. Sims
2.1 Historical context and impact on current work
At the time when Sims launched vector autoregression (VAR) methods, the
predominant empirical approach in macroeconomics was to estimate a large
system of equations built around a Keynesian macroeconomic model. Such
estimated systems were used for interpreting the time series, forecasting,
and conducting policy experiments. In a landmark contribution, Sims (1980)
argued that the resulting interpretations, forecasts, and policy conclusions all
rested on very shaky ground, because the estimation of these linear systems
generally relied on incredible identication assumptions.
To appreciate the problem of identication, suppose we consider the coee
market and try to explain movements in the quantity and price of coee. A
traditional approach is to isolate a variable that is believed to solely inuence
either supply or demand. One such variable is weather. Bad weather may
reduce the amount of coee produced at all prices, i.e., it shifts the supply
curve inward. If the demand curve for coee is not aected, a change in
the weather will lower the equilibrium quantity of coee and raise its price.
Variations in weather therefore allow us to trace out to identify the shape
of the demand curve. However, is the assumption that weather does not
inuence the demand curve plausible? Even if peoples taste for coee does
not depend directly on the weather, as Sims pointed out coee buyers know
that weather is variable and may stock up when adverse weather variations
arise. Thus, expectations about weather (and other varying determinants of
supply and/or demand) are likely to aect both supply and demand, in such
a way that weather changes may not have the expected consequences.
Even though it is well known that econometric identication is dicult
in general, Sims (1980) highlighted specic problems in macroeconomics
11
mostly, but not only, in the context of monetary economics owing to the
expectations of consumers and rms. In particular, it is hard to nd variables
that only aect either demand or supply of some macroeconomic variable
(such as consumption, or investment, or money), because expectations about
macroeconomic outcomes are in all likelihood based on all available variables.
Thus, identication in macroeconomic systems based on standard demand-
supply arguments is unlikely to work.
Sims (1980) did not only criticize the predominant macroeconometric
practice at the time. The paper also oered an identication strategy that
relied on an entirely dierent logic than the estimation of large-scale Key-
nesian models. A leading idea is to exploit the fact that the solution to a
macroeconomic system with active expectations formation can be expressed
as a VAR. This VAR can then be used to explore dierent ways of identify-
ing model parameters. Sims (1980) proposed a specic recursive scheme. A
number of alternative VAR identication strategies have subsequently been
proposed by Sims as well as by other researchers. Thus, by placing identi-
cation at the center of attention in macroeconomics, Simss work made it a
focal point of scientic discussion.
To illustrate howthe proposed approach could be applied in practice, Sims
(1980) estimated VARs for the U.S. and German economies, each based on
quarterly time series for six macroeconomic variables (money, GNP, unem-
ployment, wages, price level, and import prices). He then used the estimated
and identied VAR-systems to analyze the dynamic eects of shocks, via
impulse-response analysis and variance decomposition (see below).
Since this rst paper, Sims has continued to push the frontiers of macro-
economic VAR analysis through methodological as well as substantive contri-
butions. To give a few examples, Sims (1986) was one of the very rst papers
to discuss alternative identication schemes that are more structural than the
recursive one applied in Sims (1980). Sims, Stock and Watson (1990) showed
how to do estimation and inference in VAR systems with nonstationary time
series, including the case of cointegrated series rst analyzed by Engle and
Granger (1987). Doan, Litterman and Sims (1986) was one of the crucial
contributions to the forecasting with VARs estimated with Bayesian meth-
ods. Sims (1992) thoroughly discussed the eects of monetary policy on the
macroeconomy, drawing on the results from six-variable VARs estimated on
monthly time series for each one of the ve largest economies.
According to Simss view, VARs would be useful when interpreting time
series, in forecasting, and for understanding the eects of policy changes.
12
The ensuing literature conrms this most strongly for interpretation and
forecasting. As for policy experiments, VARs have become a main tool of
analysis for understanding the eects of temporary variations in policy but
not at least not yet for analyzing long-lasting changes in policy.
2.2 VAR analysis
VAR analysis can be described in simple terms as a method for extracting
structural macroeconomic shocks, such as unexpected exogenous shocks to
the central banks main policy instrument (e.g., the federal funds rate in
the U.S.) or unexpected exogenous changes in productivity, from historical
data and then analyzing their impact on the economy. Thus, this analysis is a
tool for (i) estimation of a forecasting model, by separating unexpected move-
ments in macroeconomic variables from expected movements; (ii) identica-
tion, by breaking down these unexpected movements into structural shocks,
i.e., shocks that can be viewed as fundamental causes of macroeconomic
uctuations; (iii) impulse-response analysis, by tracing out the dynamic im-
pact of these shocks on subsequent movements in all of the macroeconomic
variables. These three steps in the procedure are described in the following
subsections.
2.2.1 The forecasting model
Simss VAR approach is typically based on linearity and a rather unrestricted
specication with enough macroeconomic variables so that the system can
capture the key dynamics of the macroeconomy. A prerequisite for the two
remaining steps in the VAR method is a model that provides reasonable
forecasts, which amounts to a built-in assumption that the agents in an
economy (rms, households, etc.) make their forecasts actively, i.e., in a
forward-looking manner and in response to how the economy develops over
time.
Consider a vector x of dimension N denoting the macroeconomic variables
of interest. Given this vector, a reduced-form vector autoregression of order
p is a process such that
x
t
= H
1
x
t1
+ + H
P
x
tP
+ u
t
; (1)
where u
t
is uncorrelated with x
s
, s 2 ft1; : : : ; tPg and E(u
t
u
0
t
) = V . The
idea is to chose P large enough so that u becomes uncorrelated over time. A
13
large enough p will allow the VAR to approximate any covariance-stationary
processthus, the specication in (1) is rather general.
3
It is straightforward to identify H
p
; p = 1; :::; P as well as the covariance
matrix of the forecast errors V using standard regression techniques. Specif-
ically, the parameters H
p
can be estimated using ordinary least squares,
equation by equation. Estimated in this way, the VAR can be used for
forecasting. The shocks in (1), u
t
, are forecast errors. They constitute dier-
ences between the realization of x
t
and the best forecast, given information
on previous realizations of x. Thus, they are unpredictable.
Typically, these forecast errors for dierent components of x
t
are corre-
lated with each other. Therefore, they cannot be regarded as fundamental,
or structural, shocks to the economy. Instead, they should be viewed as a
function in practice, a linear combination of these fundamental shocks.
For example, suppose one of the variables in x is the interest rate. Then,
the corresponding element of u
t
cannot be interpreted as a pure interest-rate
shock, unexpectedly engineered by the central bank. Specically, part of the
interest-rate forecast error may be due to other shocks if the central banks
interest rate responds to other variables in the system within a given quar-
ter (a typical time period in macroeconomic models). Since policy variables
tend to react systematically to macroeconomic developments, this quali-
cation is very important. Thus, the system (1) cannot be used directly to
infer how interest-rate shocks aect the economy. The breakdown of forecast
errors into fundamental shocks is the identication part of VAR analysis to
be discussed below.
Forecasting with VARs Even with a limited number of variables and
without attempting to disentangle structural shocks from forecast errors,
VARs can be used directly for forecasting. Such forecasting is now quite
common. In a survey article, Stock and Watson (2001, p. 110) describe the
state of the art as follows: Small VARs like our three-variable system have
become a benchmark against which new forecasting systems are judged.
Forecasts with VARs have been compared with simple alternatives, such as
forecasting based on univariate models or pure random walks, and have of-
3
Nonstationary processes, and unit-root processes in particular, require separate analy-
ses, but they can also be studied using VAR methods with appropriate transformations.
This extension is related to the fundamental contributions of the 2003 laureate Clive W.
J. Granger.
14
ten been shown to outperform these techniques. Small VAR systems may
not be entirely stable, however, and may thus not be stable predictors of fu-
ture variables. As a result, state-of-the-art VAR forecasting tends to include
more than three variables and allow for time-varying coecients. The added
generality quickly increases the number of parameters to be estimated. One
common approach to this problem is to use BVARs, i.e., vector autoregres-
sions estimated using a Bayesian prior (see Litterman, 1986 and Sims, 1993).
Of course, the precise prior matters for the results, and many studies use the
so-called Minnesota prior (Litterman, 1986, and Doan, Litterman, and Sims,
1986) or a variant thereof.
Nowadays, a new approach is gaining ground, where the prior is based
on modern macroeconomic theory. That is, restrictions are formed based ex-
plicitly on how the econometrician a priori thinks the world works, expressed
in the form of a model. Early examples of this approach include Del Negro
and Schorfheide (2004). An alternative approach to forecasting is to rely on
structurally estimated full (perhaps nonlinear) models, and how to judge the
relative performance of dierent available forecasting approaches is still an
open issue.
2.2.2 Identication of structural shocks
Denote the fundamental shocks hitting the economy at time t by a vector "
t
;
dierent from u
t
: By denition, the components of vector "
t
are independent
variables, and their respective variances are normalized to unity. Like the
elements of u
t
, they are independent over time and therefore unpredictable.
Moreover, by construction they can be viewed as exogenous shocks: the fun-
damental economic shocks which cause subsequent macroeconomic dynam-
ics. Each element of the "
t
vector therefore has an interpretation, such as an
interest-rate rise generated by a surprise central-bank action, a sudden tech-
nology improvement, an unanticipated drop in oil prices, or an unexpected
hike in government spending.
The mapping from structural shocks to forecast errors is assumed to be
linear.
4
Thus, we can write
u
t
= G"
t
; with GG
0
= V ;
4
Linearity is not a necessary component of the specication but greatly simplies the
analysis.
15
where V is a variance-covariance matrix. The identication task now is to
impose appropriate restrictions on G. This requires knowledge of how the
economy works and a method for making use of such knowledge. Some of
the main identication schemes are briey discussed below.
Recursive identication The most common identication method, and
the one Sims (1980, 1989) used, is a so-called recursive scheme. The idea here
is to order the elements of x in such a way that the G matrix can plausibly
be argued to be lower triangular. In a simple three-variable case, this would
amount to a matrix of the form
G =
_
_
g
11
0 0
g
21
g
22
0
g
31
g
32
g
33
_
_
: (2)
It is necessary to determine the order of the variables. In the example, vari-
able 1 does not respond to the fundamental shocks in the other variables 2
and 3, variable 2 responds to shocks in variable 1 but not to shocks in vari-
able 3, and variable 3 responds to shocks in both of the other variables. The
ordering assumed is based on how rapidly dierent variables can react. For
example, it may be argued that most shocks cannot inuence government
spending contemporaneously (if the time period is short, such as a quarter
or a month), as most government activity is planned in advance and imple-
mented rather sluggishly. Stock prices, on the other hand, move very quickly
and are arguably inuenced by shocks to all contemporaneous variables, even
within a short time period.
In terms of this type of discussion, recursive identication is based on
economic theory, but only in a rudimentary sense. It is sucient to under-
stand how economic variables are dened and what decisions lie behind them,
without needing a specic structural theory of exactly how the variables are
linked. Therefore, recursive identication may often be more robust and
credible than identication schemes that rely on more detailed theoretical
assumptions of how the economy works. In partial recursive identication,
only a partial ordering is used, whereby some shocks can be identied and
others cannot. This method may be practical when an a priori ordering of
all variables in a larger system is dicult, and the focus is on a small set of
shocks, such as the monetary policy shocks. See, e.g., Sims and Zha (2006)
for such an approach.
16
Given an ordering, the elements of G can easily be computed from an
estimate of V (assuming that the estimate

V is positive denite). Recursive
identication amounts to a particular way of decomposing the matrix V;
which is called a Cholesky decomposition.
5
If the variables can convincingly
be ordered on a priori grounds in this manner, the identication task is
solved.
Other schemes for identication An alternative and more structural
scheme is long-run identication, rst proposed by Blanchard and Quah
(1989). Here, economic theory is used to make assumptions about which
shocks aect the economy in the long run. Blanchard and Quah took the
view that Keynesian-style demand shocks have no eect on output in the
long run, even though they may certainly inuence output in the short run.
But other shocks like technological or institutional change do exercise a
potential inuence on long-run output. Formally, the long-run shock variance
of the x vector would be expressed as a function of the individual shocks "
t
by
iterating forward using the VAR system (1). Restricting one element in the
matrix of long-run variances to zero amounts to an identifying assumption.
Other early approaches to structural identication include Bernanke (1986)
and Sims (1986).
Still other identication schemes may combine short-run and long-run
restrictions. They may also involve more elaborate assumptions from theory.
One approach followed by Faust (1998), Canova and De Nicol (2002) and
Uhlig (2005) is to use mild assumptions in the form of sign restrictions.
Here, certain (usually short-run) impulses or cross-correlations are assumed
to have a certain sign, whereas others are left unrestricted. Even if such
restrictions may not pin down the G matrix uniquely, they may still rule out
many possibilities. For example, the assumption that positive interest-rate
shocks cannot raise contemporaneous ination implies that such shocks must
lower contemporaneous output not by a determinate magnitude but in a
qualitative sense.
The usefulness of sign restrictions and agnostic identication depends on
the context. Consider an example (the one used later in Section 3) with three
variables ination, output, and the interest rate where each variable has
a contemporaneous inuence on every variable. A recursive identication
scheme is not consistent with this structure, since no variable is inuenced
5
The Cholesky decomposition, for a given ordering of variables, is unique.
17
only by its own structural shock, or by only two structural shocks. But the
structural shocks can still be consistently identied by sign restrictions and
agnostic procedures, since these rely precisely on theoretical restrictions on
how dierent variables are interrelated.
VAR identication based on classes of structural models rather than on a
very specic structure is common in applied work and a very active research
area. This research has led to a series of new stylized facts on how the
macroeconomy behaves, some of which are briey described in the context
of the monetary-policy example in Section 3.
2.2.3 Impulse-response analysis
With the structural shocks in hand, one can proceed to another central ele-
ment of VAR methodology: impulse-response analysis. An impulse-response
function describes how a given structural shock aects an element of the x
vector over time: the impulse (cause) and its propagation (eect).
It is straightforward to obtain the impulse responses from a VAR rep-
resentation. Using L to denote the lag operator (i.e., L
p
x
t
x
tp
), the
structural version of the VAR in (1) becomes
x
t
= [I H
1
L H
2
L
2
H
P
L
P
]
1
G"
t
:
In other words, x
t
can be described solely in terms of the entire history of
structural shocks. At the same time, the weight on L
p
in the square brackets
reects how a shock at t p inuences x
t
: An impulse response is thus
obtained by inspecting typically in a plot how the H elements of this
sum vary with p.
The estimation of VAR coecients and their accompanying standard er-
rors is rather straightforward. But providing error bands and condence in-
tervals for impulse-response functions is more complicated, and in the early
days of VAR analysis, error bands were not always provided. Nowadays,
however, they are routinely computed and displayed. Simss own approach,
along with a signicant part of the literature, is to use Bayesian methods,
but it is also common to provide classical condence intervals.
Impulse-response analysis has also become a very valuable tool for com-
paring models with data. This approach was initiated in Sims (1989). Since
estimated impulse responses provide stylized facts of a new variety, those
who formulate theoretical models of the macroeconomy commonly simulate
the model counterparts of estimated impulse responses. In developing new
18
models, common impulse-response estimates from the VAR literature are
thus used as reference points.
Another common element of the VAR methodology closely related to
impulse-response analysis is to perform so-called variance decomposition.
This amounts to computing how large a share of the variance of each vari-
able at dierent time horizons is explained by dierent types of structural
shocks. Thus, it may be concluded how dierent shocks have dierent ef-
fects at dierent horizons but combined account for the full dynamics of the
macroeconomic variables.
2.2.4 VAR applications
VAR analysis is used in a very broad set of contexts, including areas outside
macroeconomics such as nancial economics. Some of the macroeconomic
applications are as follows.
What are the eects of monetary policy? VARs have arguably been
most important in monetary economics. In particular, VARs have been used
to establish a set of facts regarding the eects of monetary policy. Monetary
shocks changes in the interest rate controlled by the central bank (the
federal funds rate in the U.S., or the xed repo rate in the Euro area)
have a signicant impact on both monetary and real variables, even though
these eects appear rather slowly and tend to display a hump-shaped pattern;
see the example in Section 3. As mentioned above, Sims (1992) discussed
the eects of monetary policy across ve dierent economies, nding several
common features but also some dierences.
What are the eects of scal policy? In the recent 2008-2009 reces-
sion, a central question for policymakers concerned the economys response to
temporary government spending (or temporary tax cuts). This is a complex
issue and clearly a reasonable answer should involve how, say, a spending
increase comes about (how it is nanced), as well as what kind of spend-
ing is raised. But are there any general lessons to be learned from looking
at historical data? As in the other contexts discussed so far, it is very im-
portant here to separate expected and unexpected changes, i.e., structural
(endogenous) shocks and (endogenous) responses, in government spending.
Various methods have been proposed to achieve this. One is to consider
military spending, which arguably has an important exogenous component.
19
Another is the so-called narrative approach, pioneered in Romer and Romer
(1989). However, some of the most commonly cited studies employ a VAR
methodology to identify how the economy reacts to spending. This is the
case in an inuential study by Blanchard and Perotti (2002), who formulate
a VAR with three variables: government spending, taxes, and output. Their
identication relies on knowledge of how total taxes react to changes in in-
come for a given tax schedule, while leaving other changes to be interpreted
as fundamental scal policy shocks. Blanchard and Perottis estimates imply
that positive government spending shocks and negative tax shocks inuence
output positively, with economically signicant eects. The sizes of these
responses (often labeled scal multipliers) have since been the subject of
extensive recent research with VAR methods.
What causes business cycles? Another important set of results obtained
through the use of VARs concerns the perennial question as to what drives
business cycles. In particular, researchers have used VAR methods to exam-
ine Kydland and Prescotts arguments that technology (productivity) move-
ments are essential drivers. Using a variety of identication schemes, tech-
nology shocks have been compared to other shocks such as monetary policy
shocks. Simss own rst studies on this topic, in particular his 1972 pa-
per Money, Income, and Causality, had an important impact. He found
that monetary movements cause movements in income (money Granger-
causes income), in the sense of Granger (1969), thus lending some support
to a monetarist view.
6
However, variance decomposition shows that a rather
small fraction of the total movements in output are accounted for in this way,
especially in the longer run. This has given rise to a large and active sub-
sequent literature, with studies inspired by both real-business-cycle theories
and Keynesian theories.
Along these lines, Gal (1999) examined technology shocks versus other
shocks through a VAR analysis, based on long-run restrictions for identica-
tion. In a very simple 2x2 VAR with productivity and total hours worked,
Gal reached the conclusion that technology shocks have relatively limited,
and somewhat counterintuitive, short-run impacts. This has been followed
by many other studies and has generated a debate that is still in progress,
6
Heuristically, a variable r Granger-causes another variable j if information about the
prior realizations of r makes it possible to arrive at better forecasts of future realizations
of j than if only past realizations of j were observed.
20
since it represents an alternative method for calibrating models when com-
paring central theories of business-cycle uctuations. The identication of
technology shocks which are arguably hard to measure directly through
long-run identifying restrictions based on VAR models has become an im-
portant subeld in recent empirical business-cycle analysis.
3 An Example: Monetary Policy and Macro-
economic Activity
In order to illustrate the methods developed by Sargent and Sims, let us intro-
duce a simple but commonly studied three-variable models of the macroecon-
omy, with ination, output, and the nominal interest rate. The interest rate
represents the monetary-policy variable whereas ination and output are de-
termined by the private agents in the economy households and rms. First,
we use this simple description of the economy to illustrate how identication
of monetary policy shocks is usually performed with the VAR methodology
developed by Sims. Impulse-response analysis then allows us to trace the
dynamic eects of unexpected changes in the interest rate engineered by
the monetary authority. Second, we use Sargents approach to describe the
analysis of a change in the policy regime, i.e., a systematically dierent way
of choosing the interest rate. This requires us to formulate a more elabo-
rate model of the economy in which structural estimation can be applied to
identify deep, policy-invariant parameters. The structural model can then
be used as a laboratory for assessing dierent interest-setting rules. The ex-
ample in this section also illustrates how the methods of Sargent and Sims
relate to each other.
3.1 A monetary VAR analysis
Let us now follow the three-step method developed by Sims, letting the VAR
be
_
_

t
y
t
i
t
_
_
= F
_
_

t1
y
t1
i
t1
_
_
+
_
_
u
;t
u
y;t
u
i;t
_
_
,
where is the ination rate, y output, and i the nominal interest rate. Con-
sider all of these variables as departures from trend: output, for example,
21
really stands for the output gap, a measure of capacity utilization.
7
In prac-
tice, VARs rely on a number of lags, but here we adhere to a single lag to
economize on notation.
As explained in Section 2.2 above, the VAR constitutes a forecasting sys-
tem that can be estimated using standard methods based on historical data.
The time series of errors obtained, the vector u
t
, is therefore unpredictable
by construction. It is a prediction error, but does not reveal the fundamental
shocks to the economy. In particular, u
i
cannot be interpreted as exogenous,
or central-bank-engineered, shocks to the interest rate.
Then, how are the fundamental shocks identied? The ordering of the
variables in the system above reects recursive identication, commonly used
in the monetary literature. The assumptions are thus: (i) a current shock
to ination is the only structural shock that inuences ination contempo-
raneously; (ii) contemporaneous output is aected by the ination shock
as well as an output shock; (iii) the interest rate can respond to all three
fundamental shocks in the system, including the fundamental shock to the
interest rate itself. Although rather stringent, these assumptions are usually
viewed as reasonable by applied macroeconomic researchers when the data
are monthly (or even quarterly). Concretely, the assumptions may reect the
inherent relative sluggishness of the dierent variables, due to informational
dierences among dierent actors in the economy, as well as adjustment costs.
The assumptions give us
_
_
u
t
u
yt
u
it
_
_
= G
_
_
"
t1
"
y
t1
"
i
t1
_
_
,
where " is the vector of structural shocks and G has the diagonal structure
described in equation (2). Since G is invertible, it is easy to compute the
structural shocks from the forecast errors obtained in the VAR estimation.
The G matrix is obtained uniquely from the relation V = G
0
G, since this
matrix equation amounts to six equations in six unknowns.
What do the estimated time series of monetary policy shocks imply in
practice? As an example, let us consider the results reported in a study of
7
Despite measuring the three variables as deviations from trend, some of them may
still not be stationary in applications. Among others, the 2003 economics laureate Clive
Granger has shown how non-stationary variables can be handled in VAR models. On this,
see also Sims, Stock and Watson (1990).
22
the U.S. economy by Christiano, Eichenbaum, and Evans (1999). This study,
which relies on recursive identication in a medium-scale VAR model, ts our
illustration because it focuses on output, ination and interest rates. It is also
appropriate in that the authors make an explicit connection to a structural
macroeconomic model, which is an extended version of the model discussed
below. The Christiano-Eichenbaum-Evans study derives time paths for the
key macroeconomic shocks and variables that are reasonably robust to the
identication scheme. The patterns, moreover, appear quite similar across
dierent countries.
A selection of estimated impulse-response functions is depicted in Figure
2, which shows how an increase in the federal-funds rate aects U.S. output
and ination. The top graph shows how an increase (by one standard devia-
tion) in the short-run interest rate gradually fades away, over approximately
six quarters. Such a shock leads to a smooth hump-shaped output contrac-
tion (the middle graph), no immediate response for ination but a delayed
decrease in the price level (the bottom graph). These responses are entirely
conditional and can be viewed as constituting an event study, where the
event is an unanticipated increase in the short-run interest rate by the central
bank.
Findings such as those in Figure 2 underlie the common practice by
ination-targeting central banks of setting their interest rate with a view
to ination one to two years down the line. As can be seen from the bottom
graph, a current interest-rate shock has little eect on ination in the rst
four quarters. The results also suggest a clear tradeo in the pursuit of con-
tractionary monetary policy: gains in terms of lower future ination have to
be weighed against losses in terms of lower output in the more immediate
future. Analogously, expansionary interest-rate policy has to trade o higher
output in the immediate future against higher ination in the more distant
future.
23
Figure 2: Estimated dynamic response to a monetary policy shock. The y axes
display percentage response of the interest rate, output, and the price level,
respectively, to a one-standard-deviation (0.72 point) increase in the Fed funds
rate, while the axes display quarters. Source: Christiano, Eichenbaum and Evans
(1999).
24
Findings such as those in Figure 2 have also been used extensively in the
development of new macroeconomic theories. Any new theory of monetary
and real variables has to be capable of reproducing the VAR evidence. The
connection between theory and empirics has implied that theories are often
studied in their (approximate) VAR representations, since this allows for easy
comparison with empirical VAR studies.
3.2 Analysis of changes in the monetary policy regime
After having arrived at estimates of historical unexpected changes in mone-
tary policy and their subsequent impact on the economy, the harder question
remains how systematic changes in monetary policy would alter the dynamic
behavior of ination, output, and the interest rate. The inherent challenge is
that such policy changes would likely inuence the estimated VAR coecients
(F and V ), since the workings of the economy in particular households and
rms expectations of future policy will be altered. This is where Sargents
method of structural estimation is required. By specifying how the economy
works in more detail, the method allows us to disentangle exactly how the
VAR coecients will change.
Consider a simple structural model of the economy. To keep matters
as simple as possible, the coecients in the model example are not derived
from microeconomic foundations, although examples of such derivations are
hinted at in footnotes below.
Model formulation Maintaining the earlier notation, the nominal side of
the private economy is described by an equation that determines the path of
ination:
t
= a
E
E
t
t+1
+ a
t1
+ a
y
y
t
+ "
;t
: (3)
The rst term on the right-hand side captures the forward-looking aspect
of price formation E
t
z
t+1
denotes the private sectors expectations, at t;
of any variable z
t+1
.
8
Including ination expectations in this equation may
reect forward-looking price setting by individual rms, for example, because
they may not change their prices in every time period and therefore care
about future demand conditions and the prices of other rms. The second
term captures a backward-looking component of ination, for instance due to
indexation of contracts. The output gap, y, also inuences prices (the third
8
For simplicity, this and the following equations are shown without constants.
25
term), typically because high output is associated with higher marginal costs
of production. Finally, "
is an exogenous (often called cost-push) shock, i.e.,

a random variable.
9
In the second equation of the model, output is determined by
y
t
= (1 b
y
)E
t
y
t+1
+ b
y
y
t1
+ b
r
(i
t
E
t
t+1
) + "
y;t
: (4)
The rst term in the expected output (gap) appears because output, similar
to prices, is partly determined in a forward-looking way. This may capture
smoothing over time, especially of consumption. The second, backward-
looking term (with b
y
> 0) may be due to adjustment costs or habits in
consumer preferences. According to the third term, output reacts negatively
to a higher real interest rate (b
r
< 0); i is the nominal interest rate and iE
is the real interest rate. The idea here is that a high real interest rate makes
current consumption less attractive relative to future consumption. Output
also contains a random element "
y
; for example reecting uctuations in
consumers attitudes toward saving.
10
9
To give a hint regarding the microeconomic foundations of (3), suppose rms are
monopolistic competitors each sells one out of a variety of imperfectly substitutable
consumption goods. Their production function includes a random labor-augmenting tech-
nology component, is common across goods, and has decreasing returns to labor input.
Several models of price stickiness imply Phillips-curve relations as in equation (3). For
instance, suppose that in its price setting, each rm is able to adjust its price with prob-
ability 1 0, but when a price adjustment is possible, it is entirely forward-looking and
maximizes the expected present value of prots (this setting relies on Calvo, 1983). Then,
a version of equation (3) with a
= 0 holds as a linearization around a zero-ination

steady state, where is the appropriate aggregate price index for the variety of consump-
tion goods. In this case, the parameters of the equation a
E
and a
y
can be written
as explicit functions of underlying primitives ,, c, , and 0, where , represents the con-
sumers subjective discount factor (which is used in the rms present-value calculation), c
the elasticity of output with respect to labor input, and the contemporaneous elasticity
of substitution across the dierent consumption goods. The case where a
6= 0 refers to
a more elaborate setting, either with a more complex adjustment assumption for prices
or an assumption that some price contracts are indexed to ination. The shock -
can
represent exogenous movements in the rms price markup over marginal costs.
10
Ignoring the taste shock and setting /
y
= 0, equation (4) can easily be derived from the
representative consumers optimal-saving condition. This so-called Euler equation sets the
marginal utility value of consumption today equal to the discounted value of consumption
tomorrow times the return on saving: n
0
(c
t
) = ,1
t
[n
0
(c
t+1
)(
1+it
1+t+1
)]. Here, n is the
utility function and
1+it
1+t+1
the realized gross real interest rate between t and t + 1, and
, is the consumers discount factor. With a power utility function, n(c) = c
1
,(1 o),
26
The simple model economy is completed with a common specication of
policy, where the central bank sets the nominal interest rate according to a
form of the Taylor rule:
i
t
= c
t
+ c
y
y
t
+ c
i
i
t1
+ "
i;t
; (5)
where "
i
is a monetary policy shock. The three response coecients in the
policy rule are all positive: c
> 0 means that the central bank raises the

interest rate when ination goes up, c
y
> 0 that it raises the interest rate
in the wake of higher output, and c
i
0 that it prefers smooth changes in
interest rates.
To describe macroeconomic uctuations, suppose that all parameters of
the model the coecient vectors a, b, and c are given. Random " shocks
are realized over time, thereby causing uctuations in output and ination
as well as in monetary policy. In the example, the c coecients are policy
parameters, while the a and b coecients reect deep microeconomic para-
meters in preferences, technology, and other policy-invariant details of the
environment. As suggested in footnotes 9 and 10, these coecients may de-
pend on a small number of such parameters. The " vector is assumed to be
unpredictable.
It may seem intuitively clear that a change in the policy rule of the
monetary authority say, an increased sensitivity of the interest rate to the
ination rate, a higher value of a
will inuence the time-series properties

of the economy. But how exactly? The diculty in analyzing this economy
over time is that current events depend on expectations about the future, at
least if E
t
reects some amount of forward-looking, purposeful behavior, i.e.,
knowledge of how the economy works. For example, if private rms at t form
expectations of prices at t + 1 and expect the price formation equation to
hold also at t + 1, their expectations should reect that knowledge. In this
example, rational expectations mean precisely that expectations are formed
with full knowledge of equations (3)(5). Events at t thus depend on events
o governing the intertemporal elasticity of substitution in consumption, one can linearize
the logarithm of the Euler equation. Since consumption must equal output in a closed
economy without investment, the real interest rate has to be set so that saving equals
zero. This condition implies equation (4). In this way, /
r
in equation (4) can be derived
as a function of o. As for the second term in (4), one obtains /
y
6= 0 if the utility function
has a form of consumption habits, where past consumption inuences current utility. The
rst two coecients, 1 /
y
and /
y
, sums to unity due to an assumption that the utility is
jointly homothetic in current and past consumption.
27
at t + 1, which in turn depend on events at t + 2, and so on. Prior to the
rational-expectations literature, expectations were treated as exogenous or
mechanically related to past values. For example, expected ination could
be assumed constant or equal to current ination. This meant that it was an
easy, mechanical, task to nd a solution to an equation system like (3)(5).
But the underlying assumption was that a change in the policy regime did
not trigger any response of private-sector expectations. This was clearly an
incredible assumption, especially in the case of monetary policy.
Model solution and cross-equation restrictions How did Sargent pro-
ceed in solving such a system?
11
An important rst step was to generally
characterize the solution to a typical macroeconomic model of this sort. In
early contributions, Sargent (1977, 1978a, 1978b) used the fact that the struc-
ture of equations such as (3)-(5) allowed a forward solution. Sargent thus
expressed current variables as a(n innite) weighted sum of current and ex-
pected future shocks " and predetermined variables
t1
, y
t1
and i
t1
with
weights that all depend on the primitive parameter vectors a, b, and c. As
a second step, given explicit assumptions on the distributions for the shocks
for instance, that they be normally distributed he appealed to the sta-
tistical time-series literature, which shows how general processes of this kind
can be estimated using maximum-likelihood methods. The goal here was to
estimate the vectors a, b, and c, the deeper determinants of the general time-
series process for i, , and y: Sargent showed that this could be accomplished
in a relatively straightforward way.
12
A key insight here is that the structural parameters appear in dierent
equations of the system and thus imply cross-equation restrictions. Sargent
called these restrictions the hallmark of rational-expectations econometrics
and such restrictions are still central elements of modern estimation. Sargent
also used recursive methods for linear macroeconomic systems of the type
described in equations (1)(3).
To illustrate the method and compactly show how cross-equation restric-
tions appear, it is useful to use a recursive representation. Conjecture that
11
To be clear, Sargent did not solve and estimate the model in this particular example.
12
In terms of the parameters of utility and technology discussed in footnotes 9 and 10,
the goal was to estimate c, ,, , o, and 0.
28
the solution to equations (1)(3) for all t would take the form
_
_

t
y
t
i
t
_
_
= F
_
_

t1
y
t1
i
t1
_
_
+ G
_
_
"
;t
"
y;t
"
i;t
_
_
; (6)
where F and G are 3x3 matrixes. As before, the components of matrix F
and the stochastic properties of the residual G" can readily be estimated by
standard regression methods.
However, the task at hand is not to estimate the F and G matrices,
but instead the basic parameter vectors a, b, and c. This requires nding the
mapping between the two. Using x
t
to denote
_
y
t

t
i
t
0
and "
t
to denote
_
"
y;t
"
;t
"
i;t
0
, (6) becomes x
t
= Fx
t1
+ G"
t
. Moreover, the structural
equations (3)-(5) can be written as x
t
= Ax
t
+BE
t
x
t+1
+Cx
t1
+"
t
.
13
Given
the conjectured solution, it must be the case that E
t
x
t+1
= Fx
t
, since the
shock vector " is unpredictable. This implies x
t
= Ax
t
+BFx
t
+Cx
t1
+"
t
.
Assuming that I A BF is invertible, we must have
x
t
= [I A BF]
1
Cx
t1
+ [I A BF]
1
"
t
:
Thus, the conjectured solution is veried with [I A BF]
1
C = F: This
is indeed a cross-equation restriction, where the fundamental parameters in
A and B imply restrictions on the coecients in matrix F:
When estimating an equation like x
t
= Fx
t1
+ u
t
, where u is a re-
gression residual, it is necessary to recognize the restrictions which theory
imposes on the coecients in the F matrix. In the model example, this re-
striction involves nine equations in eight unknown parameters (contained
in A, B, and C). Moreover, residual G" has a variance-covariance ma-
trix with six distinct elements, which gives six additional equations through
GG
0
= [I A BF]
1
_
[I A BF]
1
_
0
. The variance-covariance matrix
of " has three distinct elements since the three shocks are stochastically
independent which adds three fundamental parameters to be estimated.
In total, the simple example has 15 equations and 11 unknown parameters.
Thus, the system is over-identied as long as certain conditions are met.
14
It
13
Here, =
_
_
0 a
y
0
0 0 /
r
c
c
y
0
_
_
, 1 =
_
_
a
E
0 0
/
r
1 0
0 0 0
_
_
and C =
_
_
a
0 0
0 /
y
0
0 0 c
i
_
_
.
14
These conditions involve local invertibility of the mapping between the (1, G) and the
(, 1, C) matrices.
29
can be estimated using maximum likelihood, and the cross-equation restric-
tions implied by the model can be tested explicitly by an over-identication
test. Given parameter estimates, (estimates of) the fundamental shocks "
are recovered.
Note also that the new equation system (6) is written as an autoregression
in vector form: it is a VAR. This insight extends quite generally in the sense
that structural models have VAR representations.
15
In the example, we can
translate the structural model into a VAR, using [I A BF]
1
to map the
true innovations " to the VAR residual u. The identication problem, in the
case of the structural model, is precisely to nd the right G or, equivalently,
to nd A, B, and F. Owing to this, nding the structural shocks in the VAR
is closely related to nding the deep parameters of the system in (3)(5).
Clearly, if we manage to nd A, B, and F; the identication problem of
nding the fundamental shocks " is solved. Thus, there is a tight connection
between nding the fundamental shocks in the VAR and nding the deep
parameters of the system in (3)(5).
Viewed in this light, Sargents method involved imposing a specic eco-
nomic model, including well-dened dynamics in the example, how the
lagged terms appear and then structurally estimating that particular model.
Simss proposed method is an alternative: rely less on a specic model for-
mulation and instead rely on a whole class of statistical models. Whereas
Sargents method requires a convincing model of the economy, Simss method
requires a convincing identication scheme.
In many macroeconomic models of current interest, the original microeco-
nomically founded equations are non-linear rather than linear. This makes
the cross-equation restrictions harder to characterize, and the estimation
harder to carry out than in the simple linear example above. Moreover, mod-
ern estimation is not always carried out using classical maximum likelihood
methods but rather by Bayesian techniques. Despite these complications, the
central insights from Sargents work continue to guide empirical work today.
15
In terms of the specic identication method used by Christiano, Eichenbaum, and
Evans (1999), recursive identication is not consistent with the simple structural model
in our example. As expectations operators appear on the right-hand side of the equa-
tions, there are no zeroes in the G matrix. The authors argue, however, that although
expectations are crucial determinants of economic activity, all economic participants do
not have the same information. By spelling out such informational dierences, they show
how recursive identication of the sort shown above is fully consistent with this extension
of the model.
30
Dierent types of policy analysis In the model discussed above, one
can evaluate the eects of monetary policy in two dierent ways. One way
would be to consider the eects of temporary policy shocks, in the sense of
specic realizations of "
i;t
random shocks, so-called control errors, to the
interest rate for a given policy rule (i.e., for given coecients a
i
; a
; and a
y
).
Policy analysis with the VAR methodology is the study of such policy shocks,
i.e., unexpected (by the private sector) movements in the short-term nominal
interest rate controlled by the central bank. Impulse-response diagrams are
then used to trace the immediate and future impacts of a policy shock on
macroeconomic variables.
The other way to evaluate monetary policy is to consider changes in policy
regimes, i.e., lasting changes in the systematic reaction of policy to economic
variables. After having structurally estimated the model economy and solved
for structural parameter vectors a and b, one may conduct counterfactual
policy experiments by varying the coecient vector c. Specically, once the
deep parameters in a and b have been identied, they are held constant as
c varies. However, the reduced form dened by F and G; which describes
the data, will indeed vary since these matrixes do indeed depend on c. This
is the essence of the celebrated Lucas critique (1976).
16
According to the
estimation procedure sketched above, these variations in F and G are clear
and predictable, as they arise only through the well-dened eects of c on
the reduced form. This implies that experiments, comparing dierent policy
regimes, can be performed in a way that is immune to the Lucas critique.
The structural macroeconometric approach to policy analysis and the
VAR approach were developed in parallel and reect dierent views on gov-
ernment versus the private sector. Simss view considers policy shocks as
deliberate actions, which are unpredictable by the private sector; see Sims
(1987). In other words, the information sets of the central bank and the
private sector do not coincide. Policy shocks can thus be seen as deliberate
actions from the perspective of the central bank, yet unpredictable from the
perspective of the private sector.
Questions regarding the publics view of monetary policy and the Federal
Reserves view of the workings of the U.S. economy have been addressed in
16
In words, Lucass point was that the researcher cannot hope to analyze policy ex-
periments if the equations describing the economy (1 and G) are taken as unaected by
policy (c) when they actually are. The solution is thus to identify deeper, policy-robust
parameters (a and /) that in turn, along with policy, determine the equations of the
economy.
31
recent related contributions by both Sargent and Sims. As mentioned above,
Sargent, Williams, and Zha (2006) explicitly model central-bank learning
in this process, and tentatively conclude that monetary policy in the 1970s
was based on a temporary belief about the Phillips curve which was later
revised.
17
Sims and Zha (2006), on the other hand, formally estimate a
regime-switching model and nd some evidence of a policy switch, but only
in terms of shock variances and not in terms of the nature of policy. Despite
vigorous research in this area, these important issues have not yet been fully
settled.
4 Other Contributions by Sargent and Sims
Sargent and Sims have continued their research on the interaction between
money and economic activity. Both scholars have preserved in placing special
emphasis on how expectations are formed. These contributions are briey
described below. Sargent and Sims have also made other important contri-
butions to macroeconomics, broadly dened, which are not covered here.
Sargent and robust control Over the last decade, joint research by Lars
P. Hansen and Sargent has explored an approach to decisionmaking and ex-
pectations formation known as robust control theory (see, e.g., their 2008
monograph Robustness). This approach assumes that households and rms
act under strong aversion to uncertainty about the true model of the econ-
omy and adds a new dimension to expectations formation. Following the
engineering literature, the central idea is that decisionmakers are not only
averse to risk, but also do not know the true stochastic process which gen-
erates uncertainty. Moreover, they are very cautious and act as if they are
playing a game with nature which systematically tries to harm them.
Formally, the decisionmaker solves a maxmin problem, i.e., maximizes his
objective (the max part) under the assumption that nature selects the out-
come that is worst for him (the min part). Such behavior reects caution
in that it guarantees a high lower bound on the solution. Expectations about
the future are no longer those dictated by objective uncertainty rational
expectations but rather embody a degree of pessimism.
An interpretation of robust control theory in economics is that households
17
A similar approach is followed in Primiceri (2006).
32
and rms may not know the model, i.e., they do not fully understand what
processes generate the uncertainty in their economic environment. In this
sense, Sargent and Hansens new research program exploits a form of bounded
rationality, albeit with agents who exhibit a high degree of sophistication,
since they are still able to solve complex problems.
Sims and rational inattention An alternative approach to rational ex-
pectations formation is to directly model agents capacity constraints on in-
formation processing. This approach, labeled rational inattention, was initi-
ated in Sims (2003, 2006). It relates to earlier literature in nancial economics
where all market participants may not have the same information. In the
nance literature, this is most often modeled as dierent agents receiving
dierent information signals. Owing to signal-extraction conditions and cer-
tain incompleteness of markets, all information is not revealed and spread
through market prices (see Grossman and Stiglitz, 1980).
Rational inattention can potentially explain how dierent agents might
act on dierent information sets. It is not that they cannot access the infor-
mation, but that it is costly to interpret it. Sims imports the formal concept
of Shannon capacity from information theory and shows how agents opti-
mally choose the nature of the signals they will receive and upon which they
subsequently will act. More informative signals allow better decisions but are
also more costly, and agents have a limited amount of processing capacity
available for dierent tasks. Thus, even if information is abundant, agents
are restricted by their ability to interpret and act on information.
This research program suggests that rational inattention can produce
sluggish movements in both prices and quantities, something which is ob-
served in most contexts, as in the VARs discussed above. Earlier models,
based on various kinds of adjustment costs, could generate sluggishness but
then either for prices alone or for quantities alone. Another interesting in-
sight in this context is that optimizing price-setting sellers may not choose
prices on a continuous scale. They may instead select prices from a discrete
set, because this allows ecient information processing for buyers, or because
the sellers themselves face information-processing constraints. The literature
on rational inattention oers a novel perspective on expectations formation
and is attracting increasing interest.
33
5 Conclusion
Thomas J. Sargent and Christopher A. Sims developed the methods that now
predominate in empirical studies of the two-way relations between money or
monetary policy and the broader macroeconomy.
Sargent is the father of modern structural macroeconometrics. He has
shown how to characterize and estimate modern macroeconomic models re-
lying on full microeconomic foundations and he has demonstrated the power
of this approach in numerous applications. More generally, Sargent has pi-
oneered the empirical study of expectations formation. He has not only
demonstrated how active expectations can be incorporated into empirical
models, but has also been at the forefront of further work on expectations
formation itself, by considering many alternatives, including learning. Sar-
gent has made substantive contributions to the study of monetary policy in
his analyses of historical ination episodes throughout the world.
Sims is the father of vector autoregressions (VARs) as an empirical tool
in macroeconomics. This has become an indispensable tool for applied re-
searchers, alongside structural econometrics. Sims spearheaded the use of
VARs as an overall approach for studying macroeconomic aggregates: for
interpreting time series, for forecasting, and for policy analysis. Since his
early papers in the 1970s, Sims has contributed in many ways to the further
development of VAR methodology and to its successful application. VAR
analysis has provided a prolic means of identifying macroeconomic shocks
to variables like technology and monetary policy, and of examining the causal
eects of such shocks.
In their entirety, the research contributions of Sargent and Sims are not
merely always and everywhere central in empirical macroeconomic research
it would be nearly impossible to imagine the eld without them. Thomas
J. Sargent and Christopher A. Sims are awarded the 2011 Sveriges Riksbank
Prize in Economic Sciences in Memory of Alfred Nobel
for their empirical research on cause and eect in the macroeconomy.
34
6 References
Bernanke, Ben S., (1986), "Alternative Explanations of the Money-Income
Correlation", Carnegie-Rochester Conference Series on Public Policy 25, pp.
49-99.
Blanchard, Olivier J. and Danny Quah, (1989), "The Dynamic Eects of
Aggregate Demand and Supply Disturbances", American Economic Review
79, pp. 655-73.
Blanchard, Olivier J. and Roberto Perotti, (2002), "An Empirical Char-
acterization of the Dynamic Eects of Changes in Government Spending and
Taxes on Output", Quarterly Journal of Economics 117, pp. 13291368.
Calvo, Guillermo A., (1983), "Staggered Prices in a Utility-Maximizing
Framework", Journal of Monetary Economics 12, pp. 383-98
Canova, Fabio and Gianni De Nicol, (2002), "Monetary Disturbances
Matter for Business Fluctuations in the G-7", Journal of Monetary Eco-
nomics 49, pp. 1131-59
Christiano, Lawrence J., Martin Eichenbaumand Charles L. Evans, (1999),
"Monetary Policy Shocks: What Have We Learned and to What End?",
Handbook of Macroeconomics, Volume 1A, pp. 65-148, Amsterdam, New
York and Oxford: Elsevier Science, North-Holland.
Del Negro, Marco and Frank Schorfheide, (2004), "Priors from General
Equilibrium Models for VARs", International Economic Review 45, pp. 643-
73.
Doan, Thomas, Robert B. Litterman and Christopher A. Sims, (1986),
"Forecasting and Conditional Projection Using Realistic Prior Distribution",
Federal Reserve Bank of Minneapolis, Sta Report: 93.
Engle, Robert F. and Clive W.J. Granger, (1987), "Co-integration and
Error Correction: Representation, Estimation and Testing", Econometrica
55, pp. 251-276.
Evans, George W. and Seppo Honkapohja, (2001), Learning and Expec-
tations in Macroeconomics, Princeton, NJ and Oxford: Princeton University
Press.
Faust, Jon, (1998), "The Robustness of Identied VAR Conclusions about
Money", Carnegie-Rochester Conference Series on Public Policy 49, pp. 207-
44.
Gal, Jordi, (1999), "Technology, Employment, and the Business Cycle:
Do Technology Shocks Explain Aggregate Fluctuations?", American Eco-
nomic Review 89, pp. 249-71.
35
Gal, Jordi and Mark Gertler, (1999), "Ination Dynamics: A Structural
Econometric Analysis", Journal of Monetary Economics 44, pp. 195-222.
Granger, Clive W.J., (1969), "Investigating Causal Relations by Econo-
metric Models and Cross-spectral Methods", Econometrica 37, pp. 424-438
Grossman, Sanford J. and Joseph E. Stiglitz, (1980), "On the Impossi-
bility of Informationally Ecient Markets", American Economic Review 70,
pp. 393-408.
Hansen, Lars P. and Thomas J. Sargent, (1980), "Formulating and Es-
timating Dynamic Linear Rational Expectations Models", Journal of Eco-
nomic Dynamics and Control 2, pp. 7-46.
Hansen, Lars P. and Thomas J. Sargent, (2008), Robustness, Princeton,
NJ and Oxford: Princeton University Press.
Litterman, Robert B., (1986), "Forecasting with Bayesian Vector Autore-
gressions - Five Years of Experience", Journal of Business and Economic
Statistics 4, pp. 25-38.
Lucas, Robert E. Jr., (1972), "Expectations and the Neutrality of Money",
Journal of Economic Theory 4, pp. 103-24
Lucas, Robert E. Jr., (1973), "Some International Evidence on Output-
Ination Tradeos", American Economic Review 63, pp. 326-34
Lucas, Robert E. Jr., (1976), "Econometric Policy Evaluation: A Cri-
tique", Carnegie-Rochester Conference Series on Public Policy 1, pp. 19-46.
Lucas, Robert E. Jr. and Thomas J. Sargent, (1981), Rational Expecta-
tions and Econometric Practice, Minneapolis MN, University of Minnesota
Press.
Marcet, Albert and Thomas J. Sargent, (1989a), "Convergence of Least-
Squares Learning in Environments with Hidden State Variables and Private
Information", Journal of Political Economy 97, pp. 1306-22.
Marcet, Albert and Thomas J. Sargent, (1989b), "Least-Squares Learning
and the Dynamics of Hyperination", in William Barnett, John Geweke and
Karl Shell (eds.) Sunspots, Complexity and Chaos, Cambridge; New York
and Melbourne: Cambridge University Press.
Primiceri, Giorgio E., (2006), "Why Ination Rose and Fell: Policy-
Makers Beliefs and U.S. Postwar Stabilization Policy", Quarterly Journal
of Economics 121, pp. 867-901.
Quah, Danny and Thomas J. Sargent, (1993), "A Dynamic Index Model
for Large Cross Sections", NBER Studies in Business Cycles, vol. 28. Chicago,
IL and London: University of Chicago Press.
36
Romer, Christina D. and David Romer, (1989), "Does Monetary Pol-
icy Matter? A New Test in the Spirit of Friedman and Schwartz", NBER
Macroeconomics Annual, Cambridge MA: MIT Press.
Sargent, Thomas J., (1971), "A Note on the Accelerationist Contro-
versy", Journal of Money Credit and Banking 3, pp. 721-25.
Sargent, Thomas J., (1973), "Rational Expectations, the Real Rate of
Interest, and the Natural Rate of Unemployment", Brookings Papers on Eco-
nomic Activity 2, pp. 429-72.
Sargent, Thomas J., (1976), "A Classical Macroeconometric Model for
the United States", Journal of Political Economy 84, pp. 207-37.
Sargent, Thomas J., (1977), "The Demand for Money During Hyperina-
tions under Rational Expectations: I", International Economic Review 18,
pp. 59-82.
Sargent, Thomas J., (1978a), "Rational Expectations, Econometric Exo-
geneity, and Consumption", Journal of Political Economy 86, pp. 673-700.
Sargent, Thomas J., (1978b), "Estimation of Dynamic Labor Demand
Schedules under Rational Expectations", Journal of Political Economy 86,
pp. 1009-44.
Sargent, Thomas J., (1979), Macroeconomic Theory, New York: Acad-
emic Press.
Sargent, Thomas J., (1983), "The Ends of Four Big Inations," in Roberth
E. Hall (ed.), Ination: Causes and Eects, University of Chicago Press, pp.
4197.
Sargent, Thomas J., (1987), Dynamic Macroeconomic Theory, Cambridge,
MA: Harvard University Press.
Sargent, Thomas J., (1989), "Two Models of Measurements and the In-
vestment Accelerator", Journal of Political Economy 97, pp. 251-87.
Sargent, Thomas J., (2001), The Conquest of American Ination, Price-
ton, NJ and Oxford: Princeton University Press.
Sargent, Thomas J. and Christopher Sims, (1977), "Business Cycle Mod-
eling without Pretending to Have Too Much A Priori Economic Theory",
Federal Reserve Bank of Minneapolis, Working Paper 55.
Sargent, Thomas J. and Paolo Surico, (2011), "Two Illustrations of the
Quantity Theory of Money: Breakdowns and Revivals", American Economic
Review 101, pp. 109-28.
Sargent, Thomas J. and Neil Wallace, (1973), "The Stability of Models of
Money and Growth with Perfect Foresight", Econometrica 41, pp. 1043-48.
37
Sargent, Thomas J. and Neil Wallace, (1975), "Rational Expectations,
the Optimal Monetary Instrument, and the Optimal Money Supply Rule",
Journal of Political Economy 83, pp. 241-54.
Sargent, Thomas J. and Neil Wallace, (1976), "Rational Expectations
and the Theory of Economic Policy", Journal of Monetary Economics 2, pp.
169-83.
Sargent, Thomas J. and Neil Wallace, (1981), "Some Unpleasant Mon-
etarist Arithmetic", Federal Reserve Bank of Minneapolis Quarterly Review
5, pp. 117.
Sargent, Thomas, Noah Williams and Tao Zha, (2006), "Shocks and Gov-
ernment Beliefs: The Rise and Fall of American Ination", American Eco-
nomic Review 96, pp. 1193-1224.
Sims, Christopher A., (1972), "Money, Income, and Causality", American
Economic Review 62, pp. 540-52.
Sims, Christopher A., (1980), "Macroeconomics and Reality", Economet-
rica 48, pp. 1-48.
Sims, Christopher A., (1986), "Are Forecasting Models Usable for Policy
Analysis?", Minneapolis Federal Reserve Bank Quarterly Review 10, pp. 2-
16.
Sims, Christopher A., (1987), "A Rational Expectations Framework for
Short-run Policy Analysis", in Kenneth J. Singleton (ed.), New Approaches
to Monetary Economics, Proceedings of the Second International Symposium
in Economic Theory and Econometrics, pp. 293-308, Cambridge; New York
and Melbourne: Cambridge University Press.
Sims, Christopher A., (1989), "Models and Their Uses", American Jour-
nal of Agricultural Economics 71, pp. 489-94.
Sims, Christopher A., (1992), "Interpreting the Macroeconomic Time Se-
ries Facts: The Eects of Monetary Policy", European Economic Review 36,
pp. 975-1011.
Sims, Christopher A., (1993), "Business Cycles, Indicators, and Forecast-
ing", NBER Studies in Business Cycles 28, pp. 179-204.
Sims, Christopher A., (1994), "A Simple Model for the Study of the
Determination of the Price Level and the Interaction of Monetary and Fiscal
Policy," Economic Theory 4, pp. 381-399.
Sims, Christopher A., (2003), "Implications of Rational Inattention",
Journal of Monetary Economics 50, pp. 665-90.
Sims, Christopher A., (2006), "Rational Inattention: Beyond the Linear-
Quadratic Case", American Economic Review 96, pp. 158-163.
38
Sims, Christopher A, James H. Stock and Mark W. Watson, (1990), "In-
ference in Linear Time Series Models with Some Unit Roots", Econometrica
58, pp. 113-144.
Sims, Christopher A. and Tao Zha, (2006), "Were There Regime Switches
in U.S. Monetary Policy?", American Economic Review 96, pp. 54-81.
Stock, James H. and Mark W. Watson, (2001), "Vector Autoregressions",
Journal of Economic Perspectives 15, pp. 101-15.
Uhlig, Harald, (2005), "What Are the Eects of Monetary Policy on
Output? Results from an Agnostic Identication Procedure", Journal of
Monetary Economics 52, pp. 381-419.
Woodford, Michael, (1994), Monetary Policy and Price Level Determi-
nacy in a Cash-in-Advance Economy", Economic Theory 4, pp. 345-380.
39
11 OCTOBER 2010
MARKETS WITH SEARCH FRICTIONS
Markets with Search Frictions
1 Introduction
Most real-world transactions involve various forms of impediments to trade,
or frictions. Buyers may have trouble nding the goods they are looking
for and sellers may not be able to nd buyers for the goods they have to
oer. These frictions can take many forms and may have many sources,
including worker and rm heterogeneity, imperfect information, and costs
of transportation. How are market outcomes inuenced by such frictions?
That is, how should we expect prices to form andgiven that markets will
not clear at all points in timehow are quantities determined? Do these
frictions motivate government intervention? These questions are perhaps
particularly pertinent in the labor market where costly and time-consuming
transactions are pervasive and where the quantity determination may result
in unemployment: some workers will not nd job openings or their applica-
tions will be turned down in favor of other workers.
This years Prize is awarded for fundamental contributions to search
and matching theory. This theory oers a framework for studying frictions
in real-world transactions and has led to new insights into the workings
of markets. The development of equilibrium models featuring search and
matching started in the early 1970s and has subsequently developed into a
very large literature. The Prize is granted for the closely related contribu-
tions made by Peter Diamond, Dale Mortensen, and Christopher Pissarides.
These contributions include the analysis of price dispersion and eciency in
economies with search and matching frictions as well as the development of
what has come to be known as the modern search and matching theory of
unemployment.
The research of Diamond, Mortensen and Pissarides focuses on specic
frictions due to costly search and pairwise matching, i.e., the explicit di-
culties buyers and sellers have in locating each other, thereby resulting in
failure of markets to clear at all points in time. Buyers and sellers face
costs in their attempts to locate each other (search) and meet pairwise
1
when they come into contact (matching). In contrast, standard market
descriptions involve a large number of participants who trade at the same
time. Access to this marketplace and all the relevant information about it
is costlessly available to all economic agents; in particular, all traders would
trade at the same market price. One of the main issues, therefore, is how
price formation works in a market with search frictions. In particular, how
much price dispersion will be observed, and how large are the deviations
from competitive pricing?
Peter Diamond addressed these questions in an important paper from
1971, where he showed, rst, that the mere presence of costly search and
matching frictions does not suce to generate equilibrium price dispersion.
Second, and more strikingly, Diamond found that even a minute search cost
moves the equilibrium price very far from the competitive price: he showed
that the only equilibrium outcome is the monopoly price. This surprising
nding has been labeled the Diamond paradox and generated much follow-
up research.
Another important issue in search markets is whether there is too much
or too little search, i.e., whether or not the markets deliver ecient outcomes.
Since there will be unexecuted trade and unemployed resourcesbuyers who
have not managed to locate sellers, and vice versathe outcome might be
regarded as necessarily inecient. However, the appropriate comparison is
not with an economy without frictions. Given that the friction is a funda-
mental one that the economy cannot avoid, the relevant issue is whether the
economy is constrained ecient, i.e., delivers the best outcome given this
restriction. It should also be noted that aggregate welfare is not necessarily
higher with more search since search is costly. Diamond, Mortensen, and
Pissarides all contributed important insights into the eciency question,
with the rst results appearing in the late 1970s and early 1980s (Diamond
and Maskin, 1979, 1981; Diamond 1982a; Mortensen, 1982a,b; Pissarides
1984a,b). A generic result is that eciency cannot be expected and policy
interventions may therefore become desirable.
Along similar lines, Diamond argued that a search and matching envi-
ronment can lead to macroeconomic unemployment problems as a result of
the diculties in coordinating trade. This argument was introduced in a
highly inuential paper, Diamond (1982b), where a model featuring multi-
ple steady-state equilibria is developed. The analysis provides a rationale
for aggregate demand management so as to steer the economy towards
the best equilibrium. The key underlying this result is a search externality,
whereby a searching worker does not internalize all the benets and costs
to other searchers. The model Diamond developed in this context has also
2
become a starting point for strands of literature in applied areas such as
monetary economics and housing, which feature specic kinds of exchange
that are usefully studied with Diamonds search and matching model.
The research on search and matching theory thus raises general and im-
portant questions relevant for many applied contexts. However, the theory
has by far had its deepest impacts within labor economics. The question
of why unemployment exists and what can and should be done about it is
one of the most central issues in economics. Labor markets do not appear
to clear: there are jobless workers who search for work (unemployment)
and rms that look for workers (vacancies). It has proven to be a di-
cult challenge to formulate a fully specied equilibrium model that gener-
ates both unemployment and vacancies. The research by Peter Diamond,
Dale Mortensen and Christopher Pissarides has fundamentally inuenced
our views on the determinants of unemployment and, more generally, on the
workings of labor markets. A key contribution is the development of a new
framework for analyzing labor markets for both positive and normative pur-
poses in a dynamic general equilibrium setting. The resulting class of models
has become known as the Diamond-Mortensen-Pissarides model (or DMP
model). This canonical model originated in the rst search-matching in-
sights from the 1970s although the crucial developments occurred later on.
Especially important contributions were Mortensen and Pissarides (1994)
and Pissarides (1985). The DMP model allows us to consider simultane-
ously (i) how workers and rms jointly decide whether to match or to keep
searching; (ii) in case of a continued match, how the benets from the match
are split into a wage for the worker and a prot for the rm; (iii) rm entry,
i.e., rms decisions to create jobs; and (iv) how the match of a worker and
a rm might develop over time, possibly leading to agreed-upon separation.
The resulting models and their further developments were quite rich and
the applied research on labor markets, both theoretical and empirical, has
ourished. Theoretical work has included policy analysis, both positive and
normative. It was now straightforward to examine the eects of policies con-
cerning hiring costs, ring costs, minimum wage laws, taxes, unemployment
benets, etc. on unemployment and economic welfare. Empirical work has
consisted of systematic ways of evaluating the search and matching model
using aggregate data on vacancies and unemployment, including the de-
velopment of data bases and analyses of labor market ows, i.e., ows of
workers between dierent labor-market activities as well as job creation and
job destruction ows.
The DMP model has also been used to analyze how aggregate shocks
are transmitted to the labor market and lead to cyclical uctuations in
3
unemployment, vacancies, and employment ows. The rst step towards
a coherent search-theoretical analysis of the dynamics of unemployment,
vacancies and real wages was taken by Pissarides (1985).
Applications of search and matching theory extend well beyond labor
markets. The theory has been used to study issues in consumer theory,
monetary theory, industrial organization, public economics, nancial eco-
nomics, housing economics, urban economics and family economics.
2 General aspects of search and matching markets
The broad theoretical work on search and matching addresses three fun-
damental questions. The rst is price dispersion; i.e., whether the law of
one price should be expected to hold in markets with frictions. A central
result here is the Diamond paradox and the subsequent attempts to resolve
it. The second issue concerns eciency, which began to be addressed in
the late 1970s and into the next decade. The third question focuses on the
possibility of coordination failures based on the stylized model in Diamond
(1982b).
2.1 Price formation
The rst strand of models with explicit search activity had an entirely mi-
croeconomic focus and examined workers or consumers optimal search be-
havior under imperfect information about wages or prices. Important early
contributions to this microeconomic literature include McCall (1970) and
Mortensen (1970a,b). These models generated new results regarding the de-
terminants of search activity and, in particular, the duration of unemploy-
ment. The problem addressed in the prototype microeconomic job search
model concerns the optimal rule for accepting job oers. The unemployed
worker searching for employment is portrayed as unaware of wage oers
available at single rms but aware of the distribution of wage oers across
rms. The worker is then envisioned as sampling wage oers sequentially
and attempting to maximize the expected present value of future income.
Optimal search behavior involves a reservation wage, at which the worker is
indierent between accepting a job and remaining unemployed. The reser-
vation wage is thus set so as to equate the value of unemployment, whose
immediate return is any unemployment benet the worker receives, to the
present discounted value of future wage incomes from this job, which in-
volves the likelihood of keeping the job, the interest rate by which the fu-
ture earnings are discounted, and any expected wage movements on the job.
4
A fundamental question left unanswered in the early micro literature was
whether the postulated distribution of prices, or wages, could be rationalized
as an equilibrium outcome.
Diamonds (1971) article A Model of Price Adjustment established
what came to be known as the Diamond paradox. He demonstrated, surpris-
ingly, that under rather general conditions in an environment where buyers
and sellers search for each other, and where the sellers set, i.e., commit to,
prices in advance of meeting customers, the single monopoly price would
prevail. Diamond argued that, even with very minor search costs and with
a large number of sellers, a search and matching environment would deliver
a rather large departure from the outcome under perfect competition (which
would prevail if the search costs were zero). Thus, a small search friction
can have a large eect on price outcomes, and it would not lead to any price
dispersion at all.
A heuristic explanation of Diamonds argument is as follows. Suppose
there are many identical buyers, each in search of one unit of a good, and
that each consumer is willing to buy the good provided it costs no more than
. Suppose also that there are many identical sellers, who each commit to a
price at the beginning of the game. The buyers are perfectly informed about
the price distribution, but at each point in time a buyer only knows the price
asked by a particular seller. Each buyer must then decide whether to be
satised with this price or search more to learn the price of one additional
seller (sequential search). This search, however, only occurs at a cost, which
is assumed to be xed. It is easy to see that optimal search policy in this
context involves a cuto price : the buyer buys the good as soon as she
encounters a price at or below . The precise level of this cuto price
depends on the parameters of the model, such as the xed search cost, and on
the endogenous price distribution. Given that all consumers have identical
search costs and face the same price distribution, it must then follow that
they have the same cuto price. This immediately implies that all the
sellers will charge . However, if there is no dispersion in prices, it cannot
be optimal to learn more than one price. Thus, the unique equilibrium is one
in which all sellers charge the highest price buyers are willing to pay, i.e.,
,
the monopoly price. Put dierently, no below
can be an equilibrium,
since any given rm would deviate and choose a price ever so slightly higher
than , by an amount small enough that it would not be worthwhile for any
consumer to search for another rm. This logic works no matter how small
the search cost is, as long as it is positive.
Diamonds surprising result inspired subsequent research on the exis-
tence of price and wage dispersion in search equilibrium where rms set
5
prices (wages) optimally. Some authors, for example Albrecht and Axell
(1984), developed models where some heterogeneity across workers and/or
rms prevailed ex ante and were able to show how wage dispersion emerged
as an equilibrium outcome. Other authors maintained the assumption of ex
ante identical agents but considered alternatives to sequential search. An
important contribution in this genre is Burdett and Judd (1983), who re-
laxed the assumption of sequential search and were able to prove that price
dispersion may exist in equilibrium.
A dierent resolution of the Diamond paradox was oered in a paper by
Burdett and Mortensen (1998). They developed a model with monopsonistic
wage competition in an economy with search frictions and were able to solve
explicitly for the equilibrium wage distribution. Workers are identical ex
ante but individual heterogeneity arises ex post as workers become employed
or unemployed. A key innovation was to allow for on-the-job search and
recognize that reservation wages among employed and unemployed searchers
generally dier. Reservation wage heterogeneity creates a tradeo for rms
between volume and margin: high-wage rms are able to attract and
retain more workers than low-wage rms are, but the rent per worker that
high-wage rms can extract is relatively low. As in traditional models of
monopsony, an appropriately set minimum wage can increase employment
and welfare.
The literature on wage dispersion is nicely summarized in the recent
book by Mortensen (2005). One strand of arguments in the literature on
wage dispersion suggests that models with quantitatively large wage disper-
sion require that workers can search for other jobs while employed; see, e.g.,
Burdett (1978) for a partial-equilibrium analysis, Postel-Vinay and Robin
(2002) for an equilibrium model, and Hornstein et al. (2007) for a quantita-
tive comparison of models with and without on-the-job search.
2.2 Eciency
Frictional markets involve search externalities that may not be internalized
by agents. Consider a model where the unemployed worker determines how
intensely to search for jobs. An increase in search eort implies a higher
individual probability of becoming employed. However, there are two ex-
ternalities which are not taken into account by the individual worker. On
the one hand, by searching harder, the individual worker makes other un-
employed workers worse o by reducing their job nding rates (congestion
externality). On the other hand, by searching harder, the worker makes em-
ployers better o by increasing the rate at which they can ll their vacancies
6
(thick market externality). Congestion and thick market externalities are
common in search and matching models and it is a priori unclear whether
decentralized decisions on search and wage setting will internalize them.
In a series of contributions, Diamond examined the eciency proper-
ties of markets with frictions (Diamond and Maskin, 1979, 1981; Diamond,
1982a). Building on an earlier paper by Mortensen (1978) on ecient labor
turnover, Diamond and Maskin (economics laureate in 2007) developed a
model where individuals meet pairwise and negotiate contracts to carry out
projects (Diamond and Maskin, 1979). The quality of the match is stochas-
tic and matched individuals have the option to keep searching (at a cost) for
better matches. A unilateral separation (breach of contract) occurs when
a partner has found a better match. The authors studied alternative com-
pensation rules for such breaches of contract and examined how eciency
is related to the properties of the meeting technology, i.e., the matching
function. In general, the compensation rules under study do not result in
ecient outcomes.
Diamond (1982a) considers a labor market with search on both sides
of the market albeit with a xed number of traders. Contacts between
traders unemployed workers and rms with vacancies are governed by a
matching function and wages are determined through Nash bargaining. The
paper identies search externalities and is a precursor to more recent work
on congestion and thick market externalities.
Other important contributions in this area include Mortensen (1982a,b)
and Pissarides (1984a,b). Mortensen (1982a) species an explicit matching
technology and treats the agents search eorts as endogenous. An ecient
outcome is shown to require that the match surplus should be completely
allocated to the match maker, i.e., the agent who initiated the contact.
However, there is no mechanism to achieve that optimum; the equilibrium
is thus generically inecient. Mortensen (1982b) studies dynamic games,
including a patent race and a matching problem, where actions taken by
a single agent aect future outcomes for other agents. The main result is
similar to Mortensen (1982a): eciency requires that the agent who initi-
ated an event should obtain the whole surplus, less a compensation paid to
agents who are adversely aected. The result is sometimes referred to as
the Mortensen principle.
Pissarides (1984a) considers an economy with endogenous search inten-
sities on both sides of the market and shows that search intensities are gen-
erally too low and equilibrium unemployment too high. Pissarides (1984b)
analyzes the eciency properties of a search economy with stochastic match
productivity and nds that there can be too little or too much job rejection.
7
Pissarides argues that too little job rejection is the most plausible outcome, a
result that may suggest a role for unemployment benets so as to encourage
more rejections of low-productivity matches.
These studies on eciency by Diamond, Mortensen and Pissarides are
forerunners to the comprehensive treatment of search externalities in match-
ing models provided by Hosios (1990). The so-called Hosios condition states
that the equilibrium outcome is constrained ecient if the elasticity of
matching with respect to unemployment is equal to the workers relative
bargaining power.
1
With Nash bargaining over wages, there is no reason
why the Hosios condition should apply. Recent work on eciency proper-
ties of search equilibria has considered alternatives to Nash bargaining. One
strand of literaturecompetitive search equilibrium theoryhas shown how
the Hosios condition can arise endogenously; see, e.g., Moen (1997). In one
version of these models, rms post wages so as to attract more applicants.
Job seekers allocate themselves across rms, while recognizing that a higher
oered wage is associated with a lower probability of getting hired since a
higher wage leads to a longer queue of seekers. In equilibrium, workers are
indierent about which rm to consider.
A related strand of the search literature begins with Lucas and Prescott
(1974), who develop an island model of search. On each island, markets
are competitive (with many rms competing for many workers) and there
are no search costs, but workers may search among islands and be imper-
fectly informed about conditions on specic islands. As formulated, these
models do not feature any externalities either and decentralized equilibria
are ecient.
2.3 Coordination failures
In Diamond (1982b), it is argued that search externalities can even generate
macroeconomic coordination problems. In order to make a comprehensive
logical argument, Diamond constructed an abstract model that allowed for
careful examination of these issues. Variants and further developments of
this model have had a large impact in several areas of economics, not only for
the study of coordination problems but also as a prototype way of studying
equilibria with search and matching.
1
Using a common functional form for the matching function, the relevant elasticity,
denoted , is constant. In terms of the notation of the labor-market model in Section 3
below, the matching function can be written as ( ) =
1
, where and denote
unemployment and vacancies. The Hosios condition says that = , where is a measure
of the workers relative bargaining power.
8
Consider a continuum of risk-neutral agents who derive utility from con-
suming an indivisible good and who discount utility at rate r; time is con-
tinuous, and the ow utility of consuming a good is j. Consumers need to
trade: they each produce a good, but they do not consume the good they
produce and therefore need to nd a trading partner in order to exchange
goods. For simplicity, Diamond assumes that a consumer is willing to con-
sume any good other than his own. Production of goods occurs randomly
and with a random cost structure. The opportunity to produce a good ar-
rives according to a Poisson process at a ow probability rate j. When a
production possibility appears, the cost of production is c, with the cost
drawn from a distribution function G(c). The consumer can then choose
to produce or not depending on (i) how costly it is and (ii) the value of
being endowed with a good that can be used for trade, which depends on
how easy it is to encounter other consumers endowed with goods. Thus,
the production-consumption structure assumed here is an abstract way of
capturing gains from bilateral trade; though expressed a very particular way
in the model, the idea and applicability of the argument seem quite general.
Bilateral meetings, however, do not occur without frictions in Diamonds
model. Let the number of consumers endowed with a good, and thus search-
ing for trading partners, be denoted : (for searchers). We focus on the
case where the economy is in a steady state, so that : is constant. Let the
ow probability of meeting a trading partner be /(:), where /(0) = 0 and
/
0
(:) 0. The greater the number of agents searching for partners, the
higher is the probability of nding one for any given agent.
The pool of searchers is diminished at each point in time by the number
of traders who nd partners and thus can consume, :/(:). The pool is
increased by the number of agents who have a production opportunity and
who decide to produce: (1 :)jG(c
), i.e., the number of non-searchers

times the probability of a production opportunity times the probability that
the production cost is below the cuto cost, c
. Flow equilibrium implies:

:/(:) = (1 :)jG(c
). (1)
The cuto cost is to be determined in equilibrium. Production oppor-
tunities are accepted for c c
and rejected for c c
. To determine the
cuto cost, consider the ow utility of a searching consumer which is given
as
r\
= /(:) [j (\
)] , (2)
where \
is the expected lifetime utility of a searching agent and \
the
expected lifetime utility of an agent who does not search. The searching
9
agent meets a trading partner at the rate /(:), consumes j and switches
from searcher to non-searcher, thereby experiencing a loss in lifetime utility
given by \
. The ow utility of a non-trader reads

r\
= j max
0
(c + \
)dG(c). (3)
The non-searcher nds a production opportunity at the rate j and decides
whether or not to pay the cost c, thereby experiencing a capital gain of
\
.
The steady state of the model is straightforward to analyze. Clearly, it
must be that the cuto cost satises c
= \
. It is thus possible to
subtract (3) from (2) to obtain
rc
= /(:)(j c
) + j
Z

0
cdG(c) c
jG(c
). (4)
Equation (4) and the steady-state condition given by (1) determine c
and :. The equations can be depicted as two positively sloped relationships

in the (c
, :) space. In general, multiple equilibria are possible and equilibria

involving a higher level of economic activity yield higher welfare. Thus, there
is potentially a role for demand management, i.e., for government policy
inducing higher activity so that the economy could move from a bad to a
good steady state.
Literally, a proof that a good steady state is better than a bad steady
state is not a proof of ineciency. The move from a bad steady state to a
good steady state would require a transition period whereby agents would
rst start producing only to be able to trade later. Initially it would be hard
to nd trading partners since there are very few of them due to the low level
of production. In a later paper, Diamond and Fudenberg (1989) analyzed
the model from this perspective and indeed established ineciency as well as
entirely expectations-driven equilibrium multiplicity. They also demon-
strated that this economy could feature business-cycle-like uctuations in
output without uctuations in the fundamental parameters.
A key ingredientsubject to much discussion and empirical evaluation
in Diamonds setting is the assumption that /(:) is increasing: the larger
the number of traders in the market, the higher the meeting rates. That is,
the more traders there are, the lower are the search frictions. This assump-
tion, emphasizing the importance of scale, is usually referred to as one of
increasing returns to scale. It is an open question in any given trading
10
context whether this assumption is appropriate. For labor markets, many
argue that constant returnsin which case multiple steady states cannot
coexistdescribe reality better (see, e.g., Petrongolo and Pissarides, 2001).
The model setup has also been used in other contexts, see e.g., Due et al.
(2005) for an application to nancial markets.
Diamonds (1982b) article is often viewed as dening a new approach,
based on a careful analysis using microeconomic foundations, to analyzing
some of the central themes of Keyness business-cycle theory.
2
Coordina-
tion problems were of central importance in Keyness writings; they can be
viewed as a way of allowing for sentiments to inuence the economy, such
as Keyness well-known parable of the animal spirits of investors. If in-
vestors sense that other investors will be active and produce, they produce
too, thereby leading to high economic activity. But another equilibrium in
the same economy involves low activity.
3 Equilibrium unemployment
Unemployment suggests missing opportunities from a societal perspective
and potential ineciency of market outcomes. Through a long series of
systematic and partly overlapping contributions, Diamond, Mortensen and
Pissarides have built a foundation for the analysis of labor markets based
on search and matching frictions. This work, which began with Mortensen
(1970a,b), has fundamentally inuenced the way economists and policy-
makers approach the subject of unemployment. Their canonical model
the DMP modelhas more broadly become a cornerstone of macroeco-
nomic analysis of the labor market. Key contributions are Diamond (1981,
1982a,b), Mortensen (1982a,b), Pissarides (1979, 1984a,b, 1985), and Mor-
tensen and Pissarides (1994). Pissaridess inuential monograph (1990/2000)
provides synthesis and extensions.
3
The DMP model is a theoretical framework with a common core and a
range of specic models that deal with particular issues and invoke alterna-
tive assumptions. Wages are usually determined via bargaining between the
worker and the rm. Frictions in the market imply that there are rents to
be shared once a worker and a rm have established contact. Rents are typ-
2
Diamonds Wicksell lectures (Diamond, 1984) includes a broad discussion of the search
equilibrium approach to the microfoundations of macroeconomics.
3
Mortensen and Pissarides (1999b,c) review the search and matching model with appli-
cations to labor economics and macroeconomics. A recent comprehensive survey of search
models of the labor market is provided by Rogerson et al. (2005).
11
ically shared through the Nash solution, but the basic model is compatible
with other wage-setting rules.
An important concept in the DMP model is the so-called matching func-
tion that relates the ow of new hires to the two key inputs in the matching
process: the number of unemployed job searchers and the number of job va-
cancies. This concept has allowed researchers to incorporate search frictions
into macro models without having to specify the complex details of those
frictions (such as geographical or informational detail).
3.1 A benchmark model
The benchmark labor-market model that emerged from the work of Dia-
mond, Mortensen and Pissarides can be described in a relatively compact
way. In the following, a simple version of the setup developed in Pissarides
(1985) is described. This setup, which does not address wage dispersion,
can perhaps be viewed as the canonical equilibrium model of search unem-
ployment. Albeit simple, the model is exible enough to be useful for both
confronting data and analyzing policy issues.
3.1.1 Labor-market ows
Consider a labor market in a steady state with a xed number of labor
force participants, 1, who are either employed or unemployed. Time is
continuous and agents have innite time horizons. Jobs are destroyed at
the exogenous rate c; all employed workers thus lose their jobs and enter
unemployment at the same rate. Unemployed workers enter employment
at the rate c which is endogenously determined. Frictions in the labor
market are summarized by a matching function of the form H = /(n1, 1),
where n1 is the number of unemployed workers and 1 the number of job
vacancies. The matching function is taken as increasing in both arguments,
concave and exhibiting constant returns to scale. Unemployed workers nd
jobs at the rate c = /(n1, 1),n1 = /(1, ,n) = c(0), where 0 ,n
is a measure of labor market tightness. Firms ll vacancies at the rate
= /(n1, 1),1 = /(n,, 1) = (0). Obviously, c
0
(0) 0,
0
(0) < 0 and
c(0) = 0(0). The tighter the labor market, the easier it is for workers to
nd a job, and the more dicult for rms to ll a vacancy.
A steady state entails equilibrium in the labor market in the sense that
the unemployment rate is unchanging over time. This occurs when the inow
from employment into unemployment, c(1 n)1, equals the outow from
unemployment to employment, c(0)n1. The steady-state unemployment
12
Figure 1: The Beveridge curve
rate is thus given as:
=

+ ()
(5)
Since , this equation also implies a negative relationship between
unemployment and vacancies known as the Beveridge curve, after the British
economist William Beveridge (18791963). It is depicted in Figure 1.
A deterioration of matching eciency, i.e., a decline in job nding given
a certain level of tightness, involves an outward shift of the Beveridge curve
in the ( ) space. An increase in the job destruction rate, possibly induced
by faster sectoral reallocation of jobs, is also associated with an outward shift
of the Beveridge curve. On the other hand, since other model parameters,
such as the productivity of a match between worker and employer (due to
technology or aggregate-demand factors), do not appear in this relation,
movements in these parameters imply movements along the curve. These
dierences between model parameters allow us to gain insights into which
fundamental factors are the likely determinants of and .
4
4
Equation (5) is a steady-state relation, and thus it is not immediate that it can be
used to analyze time-series data. However, if the adjustments to steady state are rather
13
Figure 2: The U.S Beveridge curve since 2000
U.S. monthly data on unemployment and vacancies since 2000 are de-
picted in Figure 2.
5
The movements in and indicate a strong negative
relationship, with little evidence of strong shifts for most of the period, thus
suggesting that movements in productivity/demand account for most of the
aggregate uctuations in the labor market. During the current crisis, a
marked outward shift has been observed. The reasons for this shift are not
yet well understood.
3.1.2 Workers
The benchmark model features exogenous search eort and workers can only
inuence unemployment through their impact on wage setting. Workers care
about their expected present values of incomes and recognize that these val-
ues depend on labor market transition rates as well as wages while employed
quick, the equation is a good approximation also over shorter time horizons.
5
Source: U.S. Bureau of Labor Statistics, Job Openings and Labor Turnover Survey
Highlights June 2010. August 11, 2010. The job openings rate (vacancy rate) is the
number of openings divided by (employment plus job openings). The unemployment rate
is unemployment divided by the labor force.
14
and unemployment benets while unemployed. Let l denote the expected
present value of income of an unemployed worker and \ the corresponding
present value of an employed worker. With an innite time horizon and
continuous time, these value functions can be written as:
rl = / +c(0)(\ l) (6)
r\ = n +c(l \) (7)
where r is the discount rate, / is unemployment compensation (or the value
of leisure or home production during unemployment), and n is the wage.
Since we consider a steady state here, l and \ are constant. The ow
value of unemployment, rl, involves an instantaneous income / as well as
the prospect of moving from unemployment to employment; this happens
at the rate c(0) and involves a capital gain of \ l. The ow value of
employment, r\, includes instantaneous wage income n and the risk c of
a job loss and the associated capital loss of l \. From (6) and (7) one
can solve for rl and r\ as functions of /, n, r, c(0) and c.
3.1.3 Firms
Jobs are created by rms that decide to open new positions. Job creation
involves some costs and rms care about the expected present value of prof-
its, net of hiring costs. Assume for simplicity that rms are small in the
sense that each rm has only one job that is either vacant or occupied by
a worker. There is a ow cost, /, associated with a vacancy. Let \ denote
that expected present value of having a vacancy and J the corresponding
value of having a job occupied by a worker. A vacancy is lled at the rate
(0), whereas an occupied job is destroyed at the rate c. The value functions
can thus be written as:
r\ = / +(0)(J \ ) (8)
rJ = j n +c(\ J) (9)
where j is output per worker, which is taken as exogenous. The ow value
of a vacancy, r\ , involves an immediate cost / as well as the prospect of
nding a worker and thereby turning the vacancy into an occupied job. The
ow value of a lled job, rJ, involves the instantaneous prot j n but also
a risk of job destruction.
15
Free entry of vacancies implies \ = 0 in equilibrium: rms open vacan-
cies as long as it is protable to do so. By imposing the free-entry condition
on eqs. (8) and (9), one obtains the key demand-side relationship of the
model:
j n =
(r +c)/
(0)
. (10)
This free-entry condition implies a negative relationship between the
wage and labor market tightness. The tighter the labor market, the more
costly it is to recruit new workers. This has to be oset by lower wages so as
to maintain zero prots. Note that j n must hold because of hiring costs,
/ 0. In equilibrium, the excess of the marginal product of labor over the
wage cost is equal to the expected capitalized value of the vacancy cost. The
incentives to create vacancies are reduced by a higher real interest rate, a
higher job destruction rate and a higher vacancy cost. Vacancy creation is
encouraged by improved matching eciency that exogenously increases the
rate at which the rm meets job searchers.
3.1.4 Wage bargaining
Since the labor market is characterized by frictions and bilateral meetings,
the standard wage determination mechanism does not come into play. So
how are wages determined? The main approach that has been used in
the literature assumes that there is bargaining between the employer and
the worker. So suppose that wages are set through individual worker-rm
bargains and that the Nash solution applies, i.e.,
max
= [\(n) l]
[J(n) \ ]
1
,
where , is a measure of the workers relative bargaining power, , (0, 1).
\(n) and J(n) represent present values associated with a particular wage
n in this bilateral bargain (to be distinguished from the wage used in other
matches), i.e.,
r\(n) = n +c[l \(n)] ,
rJ(n) = j +c[\ J(n)] .
The value of unemployment is independent of n and is obtained from eqs.
(6) and (7). Note that the threat points in the Nash bargain are taken to be
16
l and \ , i.e., what the worker and the rm would receive upon separation
from each other.
The outcome of this maximization is a surplus-sharing rule of the form:
\(n) l = , [\(n) l +J(n) \ ] . (11)
The wage is set so as to give the worker a fraction , of the total surplus
from a wage agreement. Eq. (11) can be rewritten in several ways so as to
yield a wage equation, i.e., the bargained wage as a function of labor market
tightness and the parameters of the problem. A useful partial-equilibrium
wage equation expresses the wage as a weighted average of labor productivity
and the ow value of unemployment:
6
n = ,j + (1 ,)rl. (12)
It is possible to go one step further to obtain the following:
7
n = (1 ,)/ +,(j +/0). (13)
This expression has the intuitive property that the bargained wage is an
increasing function of unemployment benets, labor productivity and labor
market tightness.
3.1.5 Equilibrium
The overall steady-state equilibrium is now characterized by eqs. (5), (10)
and (13). Eqs. (10) and (13) determine n and 0 and the unemployment
rate follows from (5). The vacancy rate is obtained by using the fact that
= n0. The equilibrium unemployment rate is determined by /, j, /, ,, r,
c as well as by the parameters of the matching function. It is possible, by
variable substitution, to reduce the set of equations to one equation in one
unknown: labor-market tightness.
3.1.6 Comparative statics, policy analysis, and model evaluation
Given that the model can be analyzed in such a simple way, comparative
static analysis is straightforward. Consider for example an increase in un-
6
Use vW(n) = n +c[l W(n)] and vJ(n) = n +c[\ J(n)], substitute these
expressions into (11) and impose the free-entry condition \ = 0.
7
Impose free entry in (8) and obtain J = Iq(0). Use J = Iq(0) in (11) to obtain a
relationship between W l and Iq(0). Substitute the expression into (6) to eliminate
W l and substitute the resulting expression for vl back into (12).
17
employment benets. This raises the value of unemployment and reduces
the workers gain from a wage agreement; the resulting increase in wage
pressure leads to a decline in job creation, higher unemployment and higher
real wages. A higher real interest rate has an adverse impact on job creation
which leads to fewer vacancies, higher unemployment and lower real wages.
It also easy to verify that unemployment increases if there is an increase
in the vacancy cost, the job destruction rate, or the workers relative bar-
gaining power. The matching function enters via two routes: () in eq.
(5)the Beveridge curveand () in eq. (10), the free-entry condition.
Improvements in the matching technology reduce unemployment directly
(holding the number of vacancies constant) as well as indirectly (by eec-
tively reducing hiring costs and thereby encouraging job creation) and real
wages increase.
The impact of productivity on unemployment is intriguing. In the bench-
mark model as spelled out above, a higher level of productivity leads to lower
unemployment; the positive impact on job creation dominates the osetting
eect arising from higher wage pressure. Arguably, this result is reasonable
for the short run but not for the long run, since the level of productivity is a
positively trended variable whereas unemployment does not appear to have
a trend over a long enough period in time. A model able to replicate the
stylized facts of balanced growth should thus feature increasing real wages
but constant unemployment. Two slight modications of the benchmark
model are sucient for achieving that goal. The specications of vacancy
costs and unemployment benets, possibly including the value of home pro-
duction, are crucial. Suppose that unemployment benets are indexed
to real wages (or productivity) and the hiring cost grows in tandem with
real wages (or productivity). Then real wages will be responsive to general
productivity improvements and the model would, in fact, yield predictions
consistent with stylized balanced growth facts.
8
The model provides a useful framework for analyses of various policy
issues. The eects of hiring and ring costs are two pertinent examples.
The impact of ring costs depends on whether the costs involve transfers to
workers who are laid o or appear as red tape costs perhaps associated
with stringent employment protection rules. Layo costs that take the form
8
The modications of the benchmark model can be rationalized in various ways. Un-
employment benets are in practice typically indexed to wages and recruitment activities
are labor intensive activities. More generally, the workers imputed income during unem-
ployment can be regarded as proportional to his permanent income, i.e., . See Pissarides
(2000), chapter 3, for a discussion of some of the issues involved.
18
of severance pay to laid-o workers do not alter the total surplus of a match
and will not aect job creation and unemployment. Red tape costs reduce
the surplus of a match and lead to lower job creation.
There is also a large literature that assesses the model quantitatively,
using a variety of evaluation methods and dierent data sets. The devel-
opment of search and matching theory has led to a large empirical liter-
ature. The early microeconomic models of job search initiated new data
collection eorts focusing on individual labor market transitions, in par-
ticular transitions from unemployment to employment. The more recent
macroeconomics-oriented search and matching theory has been developed
in parallel with improved data availability on worker ows and job ows
(see Section 3.3 below).
The microeconomic search models have stimulated numerous empirical
studies of the determinants of unemployment duration. The methodological
literature on econometric duration analysis has expanded substantially over
the past couple of decades, a development that is to a large extent driven
by the growth and impact of microeconomic search theory. The eects of
unemployment benets on individual unemployment duration constitute the
most widely researched issue in this strand of literature. The early papers,
dating back to the late 1970s, typically identied the impact by exploiting
cross-sectional benet variation across individuals. More recent studies have
exploited information from policy reforms and quasi-experiments. The em-
pirical studies generally suggest that more generous benets tend to increase
the duration of unemployment. A key theoretical prediction from Mortensen
(1977)that the exit rate from unemployment increases as the worker ap-
proaches benet exhaustionhas been corroborated in a very large number
of studies from many countries.
Although information about how individuals respond to benet changes
is useful, it captures only a partial equilibrium relationship since rm behav-
ior is ignored. The equilibrium outcome will almost certainly dier quan-
titatively and conceivably also qualitatively from the partial equilibrium
relationship. Moreover, there are many policies, such as minimum wages
or employment subsidies, that cannot be analyzed within the partial equi-
librium framework. These concerns have initiated a number of attempts to
estimate models of equilibrium search econometrically using micro data. A
seminal paper is Eckstein and Wolpin (1990), who estimated the Albrecht
and Axell (1984) model. A more recent study is van den Berg and Ridder
(1998), who estimated an extended version of the Burdett and Mortensen
(1998) model. Mortensen (2005) includes a comprehensive discussion of
wage dierences in Denmark from the perspective of search and matching
19
models. Eckstein and van den Berg (2007) provide a recent survey of this
literature.
The aggregate matching functionan important tool kit in search and
matching theoryhas been the subject of considerable empirical research.
Blanchard and Diamond (1990) examine U.S. data and nd support for
a stable matching process. Petrongolo and Pissarides (2001) survey the
empirical literature and conclude that there is strong evidence in favor of
the conventional constant returns assumption.
Empirical research on the determinants of unemployment has often used
panel data for OECD countries in order to study the role of policies and
labor market institutions. These studies are in general theoretically eclec-
tic, drawing informally on search and matching theory as well as on models
associated with Layard, Nickell and Jackman that do not explicitly consider
labor market ows (Layard et al., 1991). The studies provide fairly strong
evidence that policies and institutions matter for long-run unemployment
outcomes. It is typically found that policies that reduce labor market ows,
such as employment protection laws, have little eect on aggregate unem-
ployment although they do increase youth unemployment and the average
duration of unemployment.
Search and matching models have been widely applied in calibration
and simulation exercises, typically to shed light on specic policy issues.
An example is Mortensen (1994b), who examined a variety of labor mar-
ket policies using the Mortensen and Pissarides (1994) model. Another
example is Mortensen and Pissarides (1999a) who extended the Mortensen
and Pissarides (1994) model to account for heterogeneous labor and studied
unemployment responses to skill-biased technology shocks and their inter-
actions with labor market policies. Search-and-matching models have also
been used in comparisons of unemployment outcomes between the U.S. and
Europe; see Rogerson and Shimer (2010) for a discussion of this literature.
The search-and-matching model also suggests exploring data sets for
rms and, in particular, data sets with information about employer-employee
pairings. Reliable data sets of this kind have not been studied until rather
recently; see Lentz and Mortensen (2010) for a survey of this literature,
which also provides links with the eld of industrial organization.
3.2 Extensions of the benchmark model
The benchmark model outlined above can be extended in many ways and
numerous extensions have appeared in the literature. As regards the impact
of productivity, note that a model where unemployment is invariant with
20
respect to the level of productivity does not rule out that unemployment re-
sponds to changes in the rate of growth of productivity. The model has been
extended so as to incorporate technological growth, thus allowing studies of
how an exogenous change in the rate of growth aects unemployment; see
e.g. Pissarides (1990/2000) and Mortensen and Pissarides (1998).
It is straightforward to introduce endogenous search eort, thereby giv-
ing the worker some direct inuence over the duration of unemployment.
Some authors have introduced stochastic job matchings, i.e., the idea that
the productivity of a match is uncertain before a worker and a vacancy
have met but is revealed once a contact is established. When uncertainty is
resolved, some matches will be accepted but others will be rejected. Endoge-
nous work hours and labor force participation are easily incorporated. The
small rm assumption of the benchmark model can be recast as a model
of a large rm that uses labor and capital and produces under constant
returns to scale.
The rst generation of DMP models focused on job creation whereas job
destruction was treated as exogenous. Changes in unemployment were thus
determined by changes in job nding whereas a workers layo risk was taken
as exogenous. Mortensen and Pissarides (1994) initiated a second-generation
version of the framework by allowing for endogenous job destruction and
thereby also endogenous worker separations. This paper introduces stochas-
tic productivity shocks and analyzes how rms and workers respond to these
shocks. At job creation, the rm is free to choose its (irreversible) technology
and prot maximization implies that new jobs are created at maximum pro-
ductivity. When hit by a productivity shock, the rm determines whether
it should remain in business or shut down, while recognizing the possibility
of renegotiating the wage contract. Changes in exogenous variables, such
as unemployment benets or matching eciency, inuence unemployment
through the impact on job creation (job nding) as well as job destruction
(worker separations). Mortensens and Pissaridess theoretical work on these
matters has been nicely matched by a growing empirical literature based on
new data on job creation and job destruction (see, e.g., Davis et al., 1996).
Most versions of the DMP model ignore on-the-job search and job-to-job
mobility. Labor turnover is then identical to job turnover. That is, workers
leave their rms only when jobs are destroyed and they nd new jobs only
via a spell of unemployment. In reality, however, job-to-job quitting with-
out intervening unemployment accounts for a substantial fraction of total
job separations. Contributions by Pissarides (1992, 2000) and Mortensen
(1994a) show how equilibrium search models can be extended so as to in-
21
corporate on-the-job search and job-to-job movements.
9
Recent extensions and generalizations of the DMP core include stud-
ies that examine the microeconomic foundations of the matching function
(Lagos, 2000; Stevens, 2007). Others have considered alternatives to the
random matching assumption of the canonical model; see, e.g., Coles and
Petrongolo (2008) and Ebrahimy and Shimer (2010). Several recent papers
have suggested modications of the standard Nash bargaining assumptions
so as to improve the models ability to account for cyclical uctuations in
unemployment and vacancies (more on this below). Risk aversion in the
context of equilibrium search-and-matching models has been less studied;
in most relevant cases, it necessitates numerical model solution (see, e.g.,
Acemoglu and Shimer, 1999).
3.3 Cyclical uctuations
The rst step towards a coherent search-theoretical analysis of the dynamics
of unemployment, vacancies and real wages was taken by Pissarides (1985).
Pre-existing dynamic general equilibrium models of business cycles had ei-
ther abstracted from unemployment or viewed unemployment as volun-
tary, i.e., as an implication of workers labor supply decisions. Given that
unemployment is a key cyclical indicator, Pissaridess framework of analysis
was a very important step forward in the literature on business cycles.
The core model is a slightly altered version of the benchmark model out-
lined previously. It thus describes an economy where vacancies and unem-
ployed workers meet according to an aggregate matching function. Produc-
tivity is match-specic so that only some of the meetings between vacancies
and workers result in actual matches, but there is also an aggregate compo-
nent to productivity that varies randomly over time. Wages are determined
through Nash bargaining and there is free entry of new vacancies. Vacan-
cies, determined by rms, are fully exible and respond instantaneously to
aggregate shocks. In terms of modeling, Pissaridess treatment of vacan-
cies constituted an innovation in this literature. Unemployment is partly
predetermined since the job creation process is time consuming.
Pissarides employs the model to study the responses to unanticipated
shocks to productivity and derives the cyclical correlation between unem-
ployment and vacancies. The model also predicts that the response of unem-
ployment to an adverse shock will be faster and sharper than the response
9
Burdett (1978) provided the seminal paper on employee search and quit rates in a
partial equilibrium setting.
22
to a positive shock. The reason for this asymmetry is that an adverse shock
results in an immediate increase in job separations and thereby to an up-
ward jump in the unemployment rate. A positive shock leads to a gradual
fall in unemployment driven by the time-consuming hiring process.
Does the model with search and matching frictions generate aggregate
uctuations, in labor markets and in other aggregates, that are also quantita-
tively consistent with data? Two early studies (Merz, 1995 and Andolfatto,
1996) propose unied models essentially combining the real-business-cycle
model due to Kydland and Prescott (1982) and the DMP model, with some
success. Following Shimer (2005), a stream of recent papers has looked at
this issue and found that the baseline model generates too little variability
compared with the data. The search and matching approach to analyzing
uctuations per se is not necessarily at issue in this debate, but some specic
details of the model are. For example, the Nash bargaining assumption for
wage determination has been scrutinized and it seems as if it leads to excess
real wage exibility, and too little unemployment volatility, in response to
shocks (Hall, 2005). Following Binmore et al. (1986), Hall and Milgrom
(2008) suggest that the relevant threat points in the wage bargain should
be payos during delays rather than payos available if the parties separate
from each other. Gertler and Trigari (2009) replace continuous Nash bar-
gaining by staggered multi-period Nash bargaining. These contributions and
others indicate that the models cyclical performance can be improved by
introducing elements of wage stickiness. However, Pissarides (2009) surveys
the empirical literature on wage exibility and argues the wage stickiness
is not the answer since wages in new matches are highly exible. Another
ingredient that matters for the models cyclical properties is the value as-
signed to leisure (or home production) during unemployment (Hagedorn and
Manovskii, 2008; Mortensen and Nagypal, 2007).
The DMP model has also been very important in allowing researchers
to connect to data more focused on how jobs appear and disappear. Based
on the longitudinal data rst studied by Davis and Haltiwanger (1992),
Cole and Rogerson (1999) specically address the ability of the model in
Mortensen and Pissarides (1994) to match the cyclical movements in job
destruction and job creation rates. They nd that so long as unemployment
duration is long enough, the model ts the facts quite well. The research on
cyclical uctuations in labor markets is very much ongoing and, although
the DMP model remains the main workhorse, alternatives will undoubtedly
be explored and compared to it in the years to come.
23
4 Other applications of search and matching the-
ory
Search and matching theory has also been applied to studies of issues in
monetary theory, with Nobuhiro Kiyotaki and Randall Wright as the main
contributors; see for example Kiyotaki and Wright (1989, 1993). This lit-
erature has examined the role of money in economies with search frictions.
The central role of money as a medium of exchange is formalized in models
with costly transactions. Fiat money (or commodity money) emerges en-
dogenously in these models. The models can be used to study the welfare
role of money and the possibility of equilibria with multiple currencies.
Search theory is a useful framework for studies of the housing market. In
an early paper, Wheaton (1990) developed a model of the housing market
with search frictions and bargained prices. In addition to a set of positive
predictions, the model has the normative implication that private search
decisions are suboptimal. Recent contributions to a growing literature on
search in the housing market include Albrecht et al. (2007).
Search theory has been invoked to study various issues in public nance,
especially concerning labor taxation and social insurance. With search fric-
tions and decentralized wage bargaining, labor taxation will generally aect
search eort as well as bargained outcomes. Standard results from competi-
tive models may no longer hold. Bovenberg (2006) includes a comprehensive
treatment of models of labor taxation in economies with search frictions or
other imperfections. The search framework is also the natural framework for
studies of positive and normative aspects of unemployment insurance and
has been widely used for those purposes. An early contribution by Diamond
(1981) shows that the presence of search externalities could motivate the
introduction of unemployment compensation even though all agents are risk
neutral. The model provides an eciency argument for public intervention
that makes job searchers more selective in their job acceptance decisions.
The idea has been revived in recent empirical and theoretical work on un-
employment insurance; see for example Acemoglu and Shimer (2000).
Due et al. (2005) make use of a search and matching model to study
issues in nancial economics. They develop a variant of the model in Dia-
mond (1982b) and study the interaction between agents in over-the-counter
markets. Search frictions aect prices and allocations, and the equilibrium
outcomes are not necessarily socially ecient. A similar market microstruc-
ture approach to the study of nancial markets can be found in Weill (2007).
Search theory has turned out to be a valuable tool for theoretical and
24
empirical research in urban economics; see Zenou (2009) for a recent mono-
graph on urban labor economics. This research introduces spatial frictions
(commuting costs) and location choice into a search and matching frame-
work. This helps to explain the determinants of urban spatial structure, the
choice of transport mode and segregation.
Search theory has been applied to analyses of the marriage market by,
among others, Mortensen (1988), Burdett and Coles (1997, 1999) and Shimer
and Smith (2000). The marriage market has features that can be well cap-
tured by the search and matching framework. Agents (singles) need to
spend time and incur costs to meet; agents typically strive for long-term
relationships; and there is competition among agents. The literature has
been concerned with the derivation of equilibrium outcomes with dierent
sorting characteristics, such as assortative mating where the traits of the
spouses are positively correlated.
5 Other contributions
Both Mortensens and Pissaridess research careers have centered around
search and matching theory and, in particular, building foundations for la-
bor market models featuring unemployment. Diamonds work, however, also
contains important contributions not directly related to search and match-
ing. These contributions deserve a brief discussion here partly because they
deal with frictions in markets in a broader sense.
Peter Diamonds work on frictions spans a very broad set of issues. An
extreme form of friction is the case of missing markets. This type of friction
naturally arises in the context of uncertainty. Here, Diamond has made
seminal contributions. When economic agents face uncertainty in the form
of a large number of possible states of nature, a complete market structure
would require a very large number of nancial assets. With such a market
structure, it would then be possible to fully insure against all kinds of risk,
including risks that are specic to individual rms or consumers. With
access to a perfect market to insure against all kinds of risk, life would be
unrealistically simple. Diamond argued that a better description of how
agents have to deal with uncertainty is one where insurance possibilities are
limited or even absent.
Diamond (1967) pioneered the analysis of economies where some insur-
ance markets are missing and asked how the absence of such markets would
inuence allocations, economic eciency and perhaps provide a reason for
government intervention. He argued that potentially useful government pol-
25
icy would not be able to directly overcome the lack of insurance markets by
simply introducing them. Instead, he insisted that in order to provide a fair
comparison between what markets and governments can do, the scope of
government policy should also be limited by the incomplete market struc-
ture. The idea that a government should be restricted by the same frictions
as those faced by markets is fundamental and has profoundly inuenced the
subsequent literature. The treatment of eciency in Diamonds early work
is paralleled in the search and matching literature, where governments also
have to play by the rules of the frictions; i.e., they have no way of directly
overcoming the frictions. Much of that literature, including Diamonds own
later work on health insurance (Diamond and Mirrlees, 1978), extends be-
yond his initial constrained eciency concept by more carefully analyzing
why markets are missing (such as private information).
Since Diamonds initial 1967 paper, an incomplete-markets literature
has developed. This literature has examined the assumptions under which
equilibria exist, are unique, and are constrained ecient. Important contri-
butions to this literature include Stiglitz (1972), Jensen and Long (1972),
Hart (1975), Grossman and Hart (1979), and Geanakoplos and Polemar-
chakis (1986) who demonstrated that incomplete asset markets may be inef-
cient in a such a way that government policy could lead to Pareto improve-
ments. Another branch of the incomplete asset market literature develops
important macroeconomic applications. Beginning with Bewley (undated),
Huggett (1993), and Aiyagari (1994), general-equilibrium models have been
developed where markets for idiosyncratic consumer risks are not present
(but where consumers can buer-save, as in Schechtman, 1976, and Bewley,
1977).
Diamond has also analyzed intergenerational market ineciencies. Cur-
rent and future, not-yet-born, generations are not directly connected to each
other in the marketplace, and market outcomes may not be ecient under
these circumstances. In perhaps his best-known work, Diamond (1965) de-
velops an overlapping generations (OLG) model with capital accumulation
using a one-sector neoclassical production technology in order to analyze
how government debt policy inuences market outcomes and consumer wel-
fare. Allais (1947) had already demonstrated that debt policy in a similar
OLG model may be used to increase steady state consumption and welfare.
A similar result is derived in Diamond (1965). His analysis of debt and
its potentially welfare-improving consequences has had profound impact on
the profession. His formulation of the overlapping-generations model, which
combines the Solow-Swan growth model with an overlapping-generations
population structure, still constitutes the benchmark model of government
26
debt, social security, and intergenerational redistribution.
10
In particular, it
has lent itself to policy-oriented modeling with a high degree of realism; see
for instance Auerbach and Kotliko (1987).
Finally, Diamond has also made important contributions to the more
traditional theory of public nance. Here, the best-known contribution
is the production eciency result of Diamond and Mirrlees (1971a, 1971b).
The argument here is that production eciency should be maintained in the
second-best allocation where by necessitysince non-distortionary taxes
are not availablemany other private decisions are distorted. As a result,
taxes on intermediate goods should all be zero and all of the tax burden
should be borne by other tax bases. Thus, production distortions cannot
help mitigate other distortions and can therefore never be benecial. The
production eciency result has direct and easily communicated implications
for optimal tax systems as well as for trade policy. For example, corporate
income tax rates and taxes on goods that are only used as intermediate goods
(such as machine parts and construction material) should be zero. Observed
real-world deviations from this result therefore cannot be motivated from an
eciency point of view.
6 Conclusions
Search and matching theory has evolved from microeconomic decision theory
to the leading paradigm in macroeconomic analyses of the labor market.
The theory has also proven to be eminently fruitful in many other areas. It
has shed light on a host of policy issues and has initiated a vast empirical
literature. The three leading contributors to search and matching theory
are Peter Diamond, Dale Mortensen, and Christopher Pissarides.
Diamond, Mortensen, and Pissarides have developed search and match-
ing theory in a number of important ways. These include (i) the litera-
ture on wage and price dispersion in search equilibrium; (ii) the literature
on macroeconomic coordination problems; (iii) foundational work on con-
strained eciency in search and matching markets; and (iv) and the devel-
opment of the canonical Diamond-Mortensen-Pissarides model of unemploy-
ment, which has become a workhorse in macroeconomic analysis and is an
important tool for policymaking.
10
Allais (1947) used a two-sector model with particular dierences in production func-
tions across consumption and investment sectors. His analysis was less general but reached
the same, key insights about the role and desirability of debt as in Diamonds 1965 paper.
27
References
Acemoglu, D and R Shimer (1999), Ecient Unemployment Insurance,
Journal of Political Economy 107, 893928.
Acemoglu, D and R Shimer (2000), Productivity Gains from Unemploy-
ment Insurance, European Economic Review 44, 11951224.
Aiyagari, S R (1994), Uninsured Idiosyncratic Risk and Aggregate Saving,
Quarterly Journal of Economics 109, 659684.
Albrecht, J and B Axell (1984), An Equilibrium Model of Search Unem-
ployment, Journal of Political Economy 92, 824840.
Albrecht, J, A Anderson, E Smith and S Vroman (2007), Opportunistic
Matching in the Housing Market, International Economic Review 48,
641664.
Allais, M (1947), Economie et Intrt, Imprimiere Nationale.
Andolfatto, D (1996), Business Cycles and Labor-Market Search, American
Economic Review 86, 112132.
Auerbach, A and L Kotliko (1987), Dynamic Fiscal Policy, Cambridge
University Press.
Bewley, T (undated), Interest Bearing Money and the Equilibrium Stock
of Capital, manuscript, Yale University.
Bewley, T (1977), The Permanent Income Hypothesis: A Theoretical For-
mulation, Journal of Economic Theory 16, 252292.
Binmore, K, A Rubinstein and A Wolinsky (1986), The Nash Bargaining
Solution in Economic Modeling, Rand Journal of Economics 17, 176
188.
Blanchard, O and P Diamond (1990), The Aggregate Matching Function,
in P Diamond (ed.), Growth, Productivity, Unemployment, MIT Press.
Bovenberg, L (2006), Tax Policy and Labor Market Performance, in J
Agell and P Birch Sorensen (eds.), Tax Policy and Labor Market Per-
formance, MIT Press.
Burdett, K (1978), A Theory of Employee Search and Quit Rates, American
28
Burdett, K and M Coles (1997), Marriage and Class, Quarterly Journal of
Economics 112, 141168.
Burdett, K and M Coles (1999), Long-Term Partnership Formation: Mar-
riage and Employment, Economic Journal 109, F307-334.
Burdett, K and K Judd (1983), Equilibrium Price Dispersion, Economet-
rica 51, 955969.
Burdett, K and D Mortensen (1998), Wage Dierentials, Employer Size
and Unemployment, International Economic Review 39, 257273.
Cole, H and R Rogerson (1999), Can the Mortensen-Pissarides Matching
Model Match the Business-Cycle Facts?, International Economic Re-
view 40, 933959.
Coles, M and B Petrongolo (2008), A Test Between StockFlow Matching
and the Random Matching Approach, International Economic Review
49, 11131539.
Davis, S and J Haltiwanger (1992), Gross Job Creation, Gross Job Destruc-
tion, and Employment Reallocation, Quarterly Journal of Economics
107, 819863.
Davis, S, J Haltiwanger and S Schuh (1996), Job Creation and Job De-
struction, MIT Press.
Diamond, P (1965), National Debt in a Neoclassical Growth Model, Amer-
ican Economic Review 55, 11261150.
Diamond, P (1967), The Role of a Stock Market in a General Equilibrium
Model with Technological Uncertainty, American Economic Review
57, 759776.
Diamond, P (1971), A Model of Price Adjustment, Journal of Economic
Theory 3, 156168.
Diamond, P (1981), Mobility Costs, Frictional Unemployment, and E-
ciency, Journal of Political Economy 89, 798813.
Diamond, P (1982a), Wage Determination and Eciency in Search Equi-
librium, Review of Economic Studies 49, 217227.
Diamond, P (1982b), Aggregate Demand Management in Search Equilib-
rium, Journal of Political Economy 90, 881894.
29
Diamond, P (1984), A Search-Equilibrium Approach to the Micro Founda-
tions of Macroeconomics, MIT Press.
Diamond, P and D Fudenberg (1989), Rational Expectations Business Cy-
cles in Search Equilibrium, Journal of Political Economy 97, 606619.
Diamond, P and E Maskin (1979), An Equilibrium Analysis of Search and
Breach of Contract, I: Steady States, Bell Journal of Economics 10,
282316.
Diamond, P and E Maskin (1981), An Equilibrium Analysis of Search
and Breach of Contract, II: A Non-Steady State Example, Journal of
Economic Theory 25, 165195.
Diamond, P and J Mirrlees (1971a), Optimal Taxation and Public Produc-
tion I: Production Eciency, American Economic Review 61, 827
Diamond, P and J Mirrlees (1971b), Optimal Taxation and Public Produc-
tion II: Tax Rules, American Economic Review 61, 261278.
Diamond, P and J Mirrlees (1978), A Model of Social Insurance with Vari-
able Retirement, Journal of Public Economics 10, 295336.
Due, D, N Grleanu and L H Pedersen (2005), Over-the-Counter Markets,
Econometrica 73, 18151847.
Eckstein, Z and K Wolpin (1990), Estimating a Market Equilibrium Search
Model from Panel Data on Individuals, Econometrica 58, 783808.
Eckstein, Z and G van den Berg (2007), Empirical Labor Search: A Survey,
Journal of Econometrics 136, 531564.
Ebrahimy, E and R Shimer (2010), Stock-Flow Matching, Journal of Eco-
nomic Theory 145, 13251353.
Geanakoplos, J and H Polemarchakis, (1986), Existence, Regularity, and
Constrained Suboptimality of Competitive Allocations when the As-
set Market is Incomplete, in W Heller, R Starr and D Starret (eds.),
Uncertainty, Information and Communication: Essays in Honor of
Kenneth J. Arrow, Vol. 3, Cambridge University Press.
Gertler, M amd A Trigari (2009), Unemployment Fluctuations with Stag-
gered Nash Wage Bargaining, Journal of Political Economy 117, 38
86.
30
Grossman, S and O Hart (1979), A Theory of Competitive Equilibrium in
Stock Market Economies, Econometrica 47, 293329.
Hagedorn, M and I Manovskii (2008), The Cyclical Behavior of Equilibrium
Unemployment and Vacancies Revisited, American Economic Review
98, 16921706.
Hall, R (2005), Employment Fluctuations with Equilibrium Wage Sticki-
ness, American Economic Review, 95, 5065.
Hall, R and P Milgrom (2008), The Limited Inuence of Unemployment
on the Wage Bargain, American Economic Review 98, 16531674.
Hart, O (1975), On the Optimality of Equilibrium when the Market Struc-
ture is Incomplete, Journal of Economic Theory 11, 418443.
Hornstein, A, P Krusell and G Violante (2007), Frictional Wage Dispersion
in Search Models: A Quantitative Assessment, NBER Working Paper
No. 13674.
Hosios, A (1990), On the Eciency of Matching and Related Models of
Search and Unemployment, Review of Economic Studies 57, 279298.
Huggett, M (1993), The Risk-Free Rate in Heterogeneous-Agent Incomplete-
Insurance Economies, Journal of Economic Dynamics and Control 17,
953969.
Jensen M and J Long (1972), Corporate Investment Under Uncertainty and
Pareto Optimality in the Capital Markets, Bell Journal of Economics
and Management Science 3, 151174.
Kiyotaki, N and R Wright (1989), On Money as a Medium of Exchange,
Kiyotaki, N and R Wright (1993), A Search-Theoretic Approach to Mone-
tary Economics, American Economic Review 83, 6377.
Kydland, F and E Prescott (1982), Time to Build and Aggregate Fluctu-
ations, Econometrica 50, 13451370
Lagos, R (2000), An Alternative Approach to Search Frictions, Journal of
Political Economy 108, 851873.
Layard, R, S Nickell and R Jackman (1991), Unemployment: Macroeco-
nomic Performance and the Labour Market, Oxford University Press.
31
Lentz, R and D Mortensen (2010), Labor Market Models of Worker and
Firm Heterogeneity, Annual Review of Economics 2, 577602.
Lucas, R and E Prescott (1974), Equilibrium Search and Unemployment,
Journal of Economic Theory 7, 188209.
McCall, J (1970), The Economics of Information and Job Search, Quarterly
Journal of Economics 84, 113126.
Merz, M (1995), Search in the Labor Market and the Real Business Cycle,
Journal of Monetary Economics, 36, 269300.
Moen, E (1997), Competitive Search Equilibrium, Journal of Political
Economy 105, 385411.
Mortensen, D (1970a), A Theory of Wage and Employment Dynamics,
in E S Phelps et al., Microeconomic Foundations of Employment and
Ination Theory, Macmillan.
Mortensen, D (1970b), Job Search, the Duration of Unemployment and the
Phillips Curve, American Economic Review 60, 847862.
Mortensen, D (1977), Unemployment Insurance and Job Search Decisions,
Industrial and Labor Relations Review 30, 505517.
Mortensen, D (1978), Specic Capital and Labor Turnover, Bell Journal
of Economics 9, 572586.
Mortensen, D (1982a), The Matching Process as a Noncooperative Bar-
gaining Game, in J McCall (ed.), The Economics of Information and
Uncertainty, University of Chicago Press.
Mortensen, D (1982b), Property Rights and Eciency in Mating, Racing
and Related Games, American Economic Review 72, 968979.
Mortensen, D (1988), Matching: Finding a Partner for Life or Otherwise,
American Journal of Sociology 94 (Supplement), S215S240.
Mortensen, D (1994a), The Cyclical Behavior of Job and Worker Flows,
Journal of Economic Dynamics and Control 18, 11211142.
Mortensen, D (1994b), Reducing Supply-side Disincentives to Job Creation,
in Reducing Unemployment: Current Issues and Policy Options, Fed-
eral Reserve Bank of Kansas City.
32
Mortensen, D (2005), Wage Dispersion: Why Are Similar Workers Paid
Dierently?, MIT Press.
Mortensen, D and C Pissarides (1994), Job Creation and Job Destruction
in the Theory of Unemployment, Review of Economic Studies 61, 397
415.
Mortensen, D and C Pissarides (1998), Technologial Progress, Job Creation
and Job Destruction, Review of Economic Dynamics 1, 733753
Mortensen, D and C Pissarides (1999a), Unemployment Responses to Skill-
biased Technology Shocks: The Role of Labour Market Policy. Eco-
nomic Journal 109, 242265.
Mortensen, D and C Pissarides (1999b), New Developments in Models of
Search in the Labor Market, in O Ashenfelter and D Card (eds.),
Handbook of Labor Economics, vol 3, part 2, Elsevier.
Mortensen, D and C Pissarides (1999c), Job Reallocation, Employment
Fluctuations and Unemployment, in J Taylor and M Woodford (eds.),
Handbook of Macroeconomics vol 1, part 2, Elsevier.
Mortensen, D and E Nagypal (2007), More on Unemployment and Vacancy
Fluctuations, Review of Economic Dynamics 10, 327347.
Petrongolo, B and C Pissarides (2001), Looking into the Black Box: A
Survey of the Matching Function, Journal of Economic Literature 39,
390431.
Pissarides, C (1979), Job Matching with State Employment Agencies and
Random Search, Economic Journal 89, 818833.
Pissarides, C (1984a), Search Intensity, Job Advertising, and Eciency,
Journal of Labor Economics 2, 128143.
Pissarides, C (1984b), Ecient Job Rejection, Economic Journal 94 (Sup-
plement), 97108.
Pissarides, C (1985), Short-Run Equilibrium Dynamics of Unemployment,
Vacancies, and Real Wages, American Economic Reivew 75, 676690.
Pissarides, C (1990/2000), Equilibrium Unemployment Theory, Basil Black-
well; second edition, MIT Press.
33
Pissarides, C (1992), Loss of Skill during Unemployment and the Persis-
tence of Employment Shocks, Quarterly Journal of Economics 107,
1371392
Pissarides, C (1994), Search Unemployment with On-the-job Search, Re-
view of Economic Studies 61, 457475.
Pissarides, C (2009), The Unemployment Volatility Puzzle: Is Wage Stick-
iness the Answer?, Econometrica 77, 13391369.
Postel-Vinay and J M Robin (2002), Equilibrium Wage Dispersion with
Worker and Employer Heterogeneity, Econometrica 70, 22952350.
Rogerson, R and R Shimer (2010), Search in Macroeconomic Models of
the Labor Market, forthcoming in O Ashenfelter and D Card (eds.),
Handbook of Labor Economics, vol 4, Elsevier.
Rogerson, R, R Shimer and R Wright (2005), Search-Theoretic Models
of the Labor Market: A Survey, Journal of Economic Literature 43,
959988.
Schechtman, J (1976), An Income Fluctuation Problem, Journal of Eco-
nomic Theory 12, 218241.
Shimer, R (2005), The Cyclical Behavior of Equilibrium Unemployment
and Vacancies, American Economic Review 95, 25-49.
Shimer, R and L Smith (2000), Assortative Matching and Search, Econo-
metrica 68, 343369.
Stevens, M (2007), New Microfoundations for the Aggregate Matching
Function, International Economic Review 48, 847868.
Stiglitz, J (1972), On the Optimality of the Stock Market Allocation of
Investment, Quarterly Journal of Economics 86, 2560.
van den Berg, G and G Ridder (1998), An Empirical Equilibrium Search
Model of the Labor Market, Econometrica 66, 11831221.
Weill, P-O (2007), Leaning Against the Wind, Review of Economic Studies
74, 13291354
Wheaton, W (1990), Vacancy, Search, and Prices in a Housing Market
Matching Model, Journal of Political Economy 98, 12701292.
Zenou, Y (2009), Urban Labor Economics, Cambridge University Press.
34
12 OCTOBER 2009
ECONOMIC GOVERNANCE

1 (19)

Economic Governance
Introduction
Institutions are sets of rules that govern human interaction. The main purpose of many institu-
tions is to facilitate production and exchange. Examples of institutions that affect human pros-
perity by enabling production and exchange include laws, business organizations and political
government. Economic governance research seeks to understand the nature of such institutions
in light of the underlying economic problems they handle.

One important class of institutions is the legal rules and enforcement mechanisms that protect
property rights and enable the trade of property, that is, the rules of the market. Another class of
institutions supports production and exchange outside markets. For example, many transactions
take place inside business firms. Likewise, governments frequently play a major role in funding
pure public goods, such as national defense and maintenance of public spaces. Key questions
are therefore: which mode of governance is best suited for what type of transaction, and to what
extent can the modes of governance that we observe be explained by their relative efficiency?

This years prize is awarded to two scholars who have made major contributions to our under-
standing of economic governance, Elinor Ostrom and Oliver Williamson.

In a series of papers and books from 1971 onwards, Oliver Williamson (1971, 1975, 1985) has
argued that markets and firms should be seen as alternative governance structures, which differ
in how they resolve conflicts of interest. The drawback of markets is that negotiations invite
haggling and disagreement; in firms, these problems are smaller because conflicts can be re-
solved through the use of authority. The drawback of firms is that authority can be abused.

In markets with many similar sellers and buyers, conflicts are usually tolerable since both sellers
and buyers can find other trading partners in case of disagreement. One prediction of
Williamsons theory is therefore that the greater their mutual dependence, the more likely
people are to conduct their transactions inside the boundary of a firm.

The degree of mutual dependence in turn is largely determined by the extent to which assets can
be redeployed outside the relationship. For example, a coal mine and a nearby electric generating
plant are more likely to be jointly incorporated the greater the distance to other mines and plants.

Elinor Ostrom (1990) has challenged the conventional wisdom that common property is poorly
managed and should be completely privatized or regulated by central authorities. Based on
numerous studies of user-managed fish stocks, pastures, woods, lakes, and groundwater basins,
Ostrom concluded that the outcomes are often better than predicted by standard theories. The

2 (19)

perspective of these theories was too static to capture the sophisticated institutions for decision-
making and rule enforcement that have emerged to handle conflicts of interest in user-managed
common pools around the world. By turning to more recent theories that take dynamics into
account, Ostrom found that some of the observed institutions could be well understood as equi-
librium outcomes of repeated games. However, other rules and types of behavior are difficult to
reconcile with this theory, at least under the common assumption that players are selfish mate-
rialists who only punish others when it is their own interest. In field studies and laboratory expe-
riments individuals willingness to punish defectors appears greater than predicted by such a
model. These observations are important not only to the study of natural resource management,
but also to the study of human cooperation more generally.

The two contributions are complementary. Williamson focuses on the problem of regulating
transactions that are not covered by detailed contracts or legal rules; Ostrom focuses on the
separate problem of rule enforcement.

Both Ostroms and Williamsons contributions address head-on the challenges posed by the
1991 Laureate in Economic Sciences, Ronald Coase (1937, 1960). Coase argued that no satis-
factory theory of the firm could rely entirely on properties of production technologies, because
economies of scale and scope might be utilized either within or across legal boundaries. Instead,
the natural hypothesis is that firms tend to form when administrative decision-making yields
better outcomes than the alternative market transaction. While Coases argument eventually
convinced economists about the need to look inside the boundaries of firms, it offered only the
preliminaries of an actual theory of the firm. Without specifying the determinants of the costs
associated with individual bargains or the costs of administrative decision-making, Coases
statement has little empirical content. The challenge remained to find ways of sharpening the
theory enough to yield refutable predictions. What exactly do organizations such as firms and
associations accomplish that cannot be better accomplished in markets?
Oliver Williamson
In a seminal paper in 1971, and an ensuing book, Markets and Hierarchies, four years later,
Oliver Williamson developed a detailed theory of the firm in the Coasean spirit. For reasons to
be explained below, Williamson claimed that organizing transactions within firms is more
desirable when transactions are complex and when physical and human assets are strongly
relationship-specific. Since both complexity and specificity can be usefully measured,
Williamsons theory had the required empirical content.

3 (19)

Theory
Williamsons theoretical argument is fourfold. First, the market is likely to work well unless
there are obstacles to writing or enforcing detailed contracts.
1

For example, at the beginning of a
buyer-seller relationship, there is usually competition on at least one side of the market. With
competition, there is little room for agents on the long side of the market to behave strategically,
so nothing prevents agreement on an efficient contract. Second, once an agent on the long side
of the market has undertaken relationship-specific investments in physical or human capital,
what started out as a transaction in a thick market, is transformed into a thin market re-
lationship in which the parties are mutually dependent. Absent a complete long-term contract,
there are then substantial surpluses (quasi-rents) to bargain over ex post. Third, the losses
associated with ex-post bargaining are positively related to the quasi-rents. Fourth, by
integrating transactions within the boundaries of a firm, losses can be reduced.
The first two points are relatively uncontroversial, but the third may require an explanation.
Why do bargaining costs tend to be higher when it is harder to switch trading partners?
Williamson offers two inter-related arguments. First, parties have stronger incentives to haggle,
i.e., to spend resources in order to improve their bargaining position and thereby increase their
share of the available quasi-rents (gross surplus from trade). Second, when it is difficult to
switch trading partners, a larger surplus is lost whenever negotiations fail or only partially
succeed due to intense haggling.

The final point says that these costs of haggling and maladaptation can be reduced by
incorporating all complementary assets within the same firm. Due to the firms legal status,
including right-to-manage laws, many conflicts can be avoided through the decision-making
authority of the chief executive.
2

Williamsons initial contributions emphasized the benefits of vertical integration, but a
complete theory of the boundaries of firms also has to specify the costs. Such an argument,
based on the notion that authority can be abused, is set forth in a second major monograph from
1985, The Economic Institutions of Capitalism (especially Chapter 6). The very incompleteness
of contracting, that invites vertical integration in the first place, is also the reason why vertical
integration is not a uniformly satisfactory solution. Executives may pursue redistribution even
when it is inefficient.
3

1
One obstacle to contracting is that parties have private information at the contracting stage. Here, we
disregard pre-contractual private information problems aside here. These issues form the core of the
mechanism design literature, which offers a complementary perspective on economic governance; cf. the
2007 Prize in Economic Sciences.

2
See Masten (1988) for a discussion of relevant legislation in the United States. As regards his own
understanding of what exactly goes on inside firms, Williamson gives substantial credit to Barnard (1938).
3
A commonly held view has been that hierarchical organization is costly because it entails administrative
costs. However, as noted by Williamson (1985), this view is unsatisfactory, because it is eminently pos-
sible to move the boundaries of firms without changing administrative routines at all.

4 (19)

To summarize Williamsons main argument, suppose that the same set of people can attempt to
conduct the same set of transactions either in the market or within the boundaries of a firm.
Organizing the transaction within a firm centralizes decision rights, thereby saving on
bargaining costs and reducing the risk of bargaining impasse, but at the same time allows
executives more scope to extract rents in inefficient ways. The net effect of this trade-off
depends on both the difficulty of writing useful contracts ex ante and the extent to which assets
are relationship-specific ex post. Williamsons hypothesis is that governance will be aligned to
the underlying technology and tastes depending on this trade-off. Transactions will be
conducted inside firms if they involve assets which are only valuable to particular sellers or
buyers, especially if uncertainty or complexity raise the cost of writing complete and
enforceable contracts. Otherwise, they will take place in the market.
Evidence
By now, there is a wealth of evidence showing that vertical integration is affected by both com-
plexity and asset specificity. Shelanski and Klein (1995) provide a survey of empirical work
specifically directed toward testing Williamsons hypotheses, and Masten (1996) presents a
compilation of some of the best articles in this genre. More recent studies include Novak and
Eppinger (2001) and Simester and Knez (2002). Lafontaine and Slade (2007) provide a broad
overview of the evidence concerning the determinants of vertical integration. They summarize
their survey of the empirical literature as follows:
The weight of the evidence is overwhelming. Indeed, virtually all predictions from
transaction cost analysis appear to be borne out by the data. In particular, when the
relationship that is assessed involves backward integration between a manufacturer
and her suppliers, there are almost no statistically significant results that contradict
TC [transaction cost] predictions. (p. 658)
Consider, for example, Joskows (1985, 1987) studies of transactions between coal mines and
electric generators. The mining of coal and the burning of coal to generate electricity are two
quite unrelated processes. However, it is quite costly to transport coal, so if there is only one
mine nearby that produces coal of adequate quality, there is a high degree of mutual dependence
between the owner of the mine and the owner of the electric plant. Roughly speaking,
Williamsons theory says that the further away a mine/generator pair is located from other
mines and generators, the greater is the likelihood that the pair is jointly owned.

The natural variation in asset specificity that arises due to the difference in distance between
adjacent coal repositories implies that Joskow could credibly identify a causal relationship be-
tween asset specificity and contractual relations. As Williamsons theory predicts, the contracts
are relatively rudimentary and have shorter duration when asset specificity is low, and become
more detailed and long-lasting as asset specificity increases. In cases of extreme specificity,
contracts either last very long (up to fifty years), or the mine and the generator are both owned

5 (19)

by the same firm. Thus, as asset specificity goes from low to high, the relationship between
mine operators and electric generators is gradually transformed from a pure market relationship
to a pure non-market relationship.
Normative implications
Williamsons major contribution is to provide an explanation for the location of firms bounda-
ries. However, the theory also has normative implications for firms as well as for competition
legislation. Let us briefly address these normative implications.

The evidence cited above suggests that vertical integration of production is affected by the trade-
off that Williamson identified. This does not imply that the owners of these firms have understood
the underlying economic logic. More plausibly, the empirical regularity instead emerges because
firms that have inappropriate boundaries tend to be less profitable and are hence less likely to
survive. If so, Williamsons theory has normative content that is of value to managers.

In fact, Williamsons books have frequently been compulsory reading in courses on corporate
strategy at business schools throughout the world, with the explicit aim of training managers to
improve their decision-making. To the extent that this teaching is successful, Williamsons
research not only helps to explain observed regularities but also entails better utilization of the
worlds scarce resources.

Williamsons theory of vertical integration clarifies why firms are essentially different from
markets. As a consequence, it challenges the position held by many economists and legal scho-
lars in the 1960s that vertical integration is best understood as a means of acquiring market
power. Williamsons analysis provides a coherent rationale for, and has probably contributed to,
the reduction of antitrust concerns associated with vertical mergers in the 1970s and 80s. By
1984, merger guidelines in the United States explicitly accepted that most mergers occur for
reasons of improved efficiency, and that such efficiencies are particularly likely in the context
of vertical mergers.
4

Subsequent work: Broadening the arguments
By now there is a huge literature on the boundaries of the firm, and we shall not attempt to de-
scribe it all here; see Gibbons (2005) for an overview of both closely and more distantly related
work. We offer only a brief look, starting with some of the complementary research that
emerged soon after 1975.

4
Prior to his pioneering work on vertical integration, Williamson had already begun to have an impact on
U.S. competition policy. While working for the Assistant Attorney General for Antitrust at the U.S.
Department of Justice, he wrote a paper, subsequently published as Williamson (1968), suggesting that
horizontal mergers should sometimes be allowed on efficiency grounds. According to Kolasky and Dick
(2003), Williamsons suggestion noticeably affected the U.S. Justice Departments very first Merger
Guidelines, which were issued in 1968.

6 (19)

Whereas Williamson focused on the problem of efficient conflict resolution, much subsequent
work instead emphasized that incomplete contracts in combination with asset specificity can
create inefficiency even if conflicts are resolved efficiently ex post. When parties are obliged to
make large relationship-specific investments, they do not care primarily about the efficiency of
the division of future surplus, but about their own private return. For example, if a supplier must
invest in highly specific machinery in order to serve a particular customer, and the state-
contingent terms of trade cannot be easily detailed in advance, the supplier might worry that the
customer could extract a significant portion of the surplus when the price is finally negotiated.
This problem is known as the hold-up problem.
5

An important early statement of the hold-up problem is due to Klein, Crawford, and Alchian
(1978), and the first formal studies of hold-up problems with explicitly non-contractible invest-
ments are Grossman and Hart (1986) and Hart and Moore (1990) (henceforth GHM). GHM
focus on ex-ante investment distortions instead of ex-post conflict costs. Their key argument is
that asset ownership entails a negotiation advantage. Thus, instead of asking which assets
should have the same owner, GHM asks who should own which assets. Put simply, while neg-
lecting several important caveats, GHM conclude that ownership should be given to the party
who makes the most important non-contractible relationship-specific investment. In relation to
Williamsons theory, GHMs theory is complementary. For reasons explained by Whinston
(2003), the additional predictions have proven harder to test, but the limited evidence that exists
is supportive (Lafontaine and Slade, 2007).
Subsequent work: Deepening the analysis
In comparison with other modern research in economics, Williamsons theory of the firm re-
mains relatively informal. A likely reason is that the economics profession has not yet perfected
the formal apparatus required to do justice to the theory (Williamson, 2000). Two major
challenges have been to model contractual incompleteness and inefficient bargaining.

Contractual incompleteness is presumably related to bounded rationality, and useful models of
bounded rationality have taken a long time to emerge despite the pioneering efforts of the
1978 Laureate in Economic Sciences Herbert Simon (1951, 1955). Nowadays, however, there
are several detailed formal models of the relationship between bounded rationality and
contractual incompleteness, including for example Anderlini and Felli (1994), Segal (1999), and
Tirole (2009).

5
Implicit in the above statement is the assumption that the ex-ante investments cannot be contracted
upon. As shown by Crawford (1988), the hold-up problem vanishes if investments are contractible
(see also Fudenberg, Holmstrm and Milgrom, 1990 and Milgrom and Roberts, 1990).

7 (19)

As regards inefficient bargaining, the most common approach is to ascribe disagreement to
asymmetric information.
6

Consistently with this view, Williamson (1975, p. 26) argued that
conflict may arise due to opportunistic bargaining strategies such as selective or distorted
information disclosure and self-disbelieved threats and promises. Abreu and Gul (2000) and
Compte and Jehiel (2002) develop state-of-the-art bargaining models in which there can be
substantial losses due to the latter form of strategic posturing. Thus, the relevant question is not
whether Williamsons theory can be formalized, but when we will see fully fledged
formalizations of it.
Williamsons work has also inspired a wealth of research that seeks to articulate how conflicts
are resolved within firms. One line of research, initiated by Kreps (1990), studies the crucial
problem of how conflicts are resolved in the absence of a contract that will be enforced exter-
nally. Kreps uses the theory of repeated games to explain how reputational mechanisms can
substitute for contracts, and sets forth a game theoretic model of the firm or its owner/
manager as a bearer of reputations. (Repeated game logic is also an important aspect of
Ostroms contributions; see below.) Baker, Gibbons, and Murphy (2002) study this issue in a
model that is more explicitly geared to analyze internal governance; see also the related work by
Garvey (1995) and Halonen (2002).
Wider implications
Although Williamsons main contribution was to formulate a theory of vertical integration, the
broader message is that different kinds of transactions call for different governance structures.
More specifically, the optimal choice of governance mechanism is affected by asset specificity.
Among the many other applications of this general idea, ranging from theories of marriage
(Pollak, 1985) to theories of regulation (Goldberg, 1976), one has turned out to be particularly
important, namely corporate finance.

Williamson (1988) notes that the choice between equity and debt contracts closely resembles
the choice between vertical integration and separation. Shareholders and creditors not only re-
ceive different cash flows, but have completely different bundles of rights. For example, con-
sider the relationship between an entrepreneur and different outside investors. One class of
investors, creditors, usually do not acquire control rights unless the entrepreneur defaults, whe-
reas another class of investors, stockholders, typically have considerable control rights when the
entrepreneur is not in default. Williamson suggests that non-specific assets, which can be re-
deployed at low cost, are well suited for debt finance. After a default, creditors can simply seize

6
Other theories of inefficient bargaining outcomes rely on irreversible strategic commitments, as
suggested by Schelling (1956), a recipient of the 2005 Prize in Economic Sciences (see also Crawford,
1982) or cognitive biases, as suggested by Babcock and Loewenstein (1997). In recent work on
incomplete contracts, Hart and Moore (2008) have made the assumption that ex-post bargaining
is inefficient because of such psychological mechanisms.

8 (19)

these assets from the entrepreneur. Specific assets on the other hand are less well suited for debt
finance, because control rights lose their value if they are redeployed outside the relationship.

Subsequent formal modeling, by Aghion and Bolton (1992), Hart and Moore (1989), Hart
(1995) and many others, confirms the usefulness of the incomplete contracts approach for ana-
lyzing corporate finance decisions. More generally, this line of work has been instrumental in
promoting the merger between the fields of corporate finance and corporate governance a
merger process that was initiated by Jensen and Meckling (1976).

Another far-reaching lesson from Williamsons governance research is that core questions
concerning actual and desirable social organization span several disciplines. Both through his
writings and his founding editorship of the Journal of Law, Economics and Organization,
Oliver Williamson has contributed to eliminating many of the barriers to intellectual exchange
among different disciplines of the social sciences.
Elinor Ostrom
Common-pool resources (CPRs) are resources to which more than one individual has access,
but where each persons consumption reduces availability of the resource to others. Important
examples include fish stocks, pastures, and woods, as well as water for drinking or irrigation.
On a grander scale, air and the oceans are common pools.

Some common pools exist primarily due to technological properties of the resource. For exam-
ple, difficulties in controlling peoples resource usage prevent the transformation of a common
pool resource into a private resource.

However, not all costs of precluding access are strictly technological. There are also cases in
which common pools could be profitably privatized, whereupon access could easily be con-
trolled, but where privatization attempts fail because the users cannot agree on the terms. For
example, water basins and oil pools are frequently located underneath land that has many
different owners. Although these owners as a group would benefit from consolidating explora-
tion under the umbrella of a single firm, it can be remarkably difficult to reach private agree-
ment about the division of the surplus (see e.g. Libecap and Wiggins, 1984, 1985). In general, a
combination of technological and institutional factors determines whether resources are
managed as common property.

The overexploitation of common-pool resources is a well-known problem that has occupied
social thinkers for at least two millennia and probably even longer. Individual users of a
resource may have strong private incentives to act in ways that are detrimental to the group as a
whole. Early formal analyses of this problem are due to Warming (1911) and Gordon (1954),
who studied the special case of open access, i.e., when there is entry of users until the marginal
benefit equals the marginal cost to the last entrant. The case of a fixed number of users was later

9 (19)

studied by Clark (1976) and Dasgupta and Heal (1979).
7

The models provide plausible condi-
tions, at least under the simple but restrictive assumption that users interact in a single period
only, under which excess utilization is the unique equilibrium outcome.
More than forty years ago, the biologist Garrett Hardin (1968) observed that overexploitation of
common pools was rapidly increasing worldwide and provided the problem with a catchy and
relevant title: The Tragedy of the Commons.

In economics, two primary solutions to the common-pool problem have been suggested. The
first is privatization. The feasible forms of privatization depend on technologies available for
measurement and control. For example, if detailed monitoring of appropriation is prohibitively
expensive, effective privatization may require concentration of ownership in the hands of one or
a few agents.

An alternative solution, often associated with Pigou (1920), is to let the central government own
the resource and levy a tax extraction. This solution initially entails coercion, in the sense that
original users are disenfranchised. But under ideal circumstances especially zero monitoring
costs and full knowledge of appropriators preferences the taxes will be the same as the prices
of an efficient market. Under such ideal circumstances, there is also an equivalent solution to
the problem based on quotas instead (Dasgupta and Heal, 1979).

Coase (1960) argued that the Pigovian solution works so well in theory only because the real
problems are assumed away. Taxation is a perfect solution in the absence of transaction costs,
but governmental regulation itself is unnecessary in this case. Absent transaction costs, private
agreements between the parties concerned suffice to achieve efficiency. Thus, if it is possible to
determine fully efficient taxes or quotas, it might also be possible for the users to negotiate the
optimal outcome.

Coase insisted that the case of zero transaction costs is a purely theoretical construction. In practice,
all forms of governance have costs. The real challenge is to compare various private and public
orderings while taking all the relevant transaction costs into account. Depending on the transaction
costs, the market, the firm or the government may constitute the best governance mechanism.

A third solution previously discarded by most economists is to retain the resource as com-
mon property and let the users create their own system of governance. In her book Governing
the Commons: The Evolution of Institutions for Collective Action (1990), Elinor Ostrom objects
to the presumption that common property governance necessarily implies a tragedy. After

7
The case of a fixed number of users echoes two other classic problems of cooperation, namely the
problem of voluntary provision of public goods and the problem of oligopoly behavior. In each case, the
individuals incentive is in conflict with the group interest. (Typically, the corresponding one-shot game
has a unique and inefficient equilibrium.) See Olson (1965) for a seminal study that addresses the
problem of cooperation more generally.

10 (19)

summarizing much of the available evidence on the management of common pools, she finds
that users themselves envisage rules and enforcement mechanisms that enable them to sustain
tolerable outcomes. By contrast, governmentally imposed restrictions are often counterproduc-
tive because central authorities lack knowledge about local conditions and have insufficient
legitimacy. Indeed, Ostrom points out many cases in which central government intervention has
created more chaos than order.
8
Ostroms contributions

Ostrom bases her conclusions primarily on case studies. Over the years, Ostroms own field
work gave rise to some of the cases, starting with her doctoral dissertation in 1965. Here, she
studied the institutional entrepreneurship involved in an effort to halt the intrusion of saltwater
into a groundwater basin under parts of the Los Angeles metropolitan area.

However, a few case studies rarely permit broad generalizations. The key to Ostroms break-
through insights was instead the realization, about twenty years later, that there exist thousands
of detailed case studies of the management of CPRs, and that most of them were written by
authors interested in only one or a small set of cases.
9

By collecting and comparing these iso-
lated studies, it should be possible to make substantially stronger inferences.
In most of the cases, local communities had successfully managed CPRs, sometimes for
centuries, but Ostrom also pays close attention to unsuccessful cases. Ostrom empirically
studies both the rules that emerge when local communities organize to deal with common pool
problems and the processes associated with the evolution and enforcement of these rules. She
documents that local organization can be remarkably efficient, but also identifies cases in which
resources collapse. Such case studies help to clarify the conditions under which local gover-
nance is feasible. They also highlight circumstances under which neither privatization nor state
ownership work quite as well as standard economic analysis suggests.

In order to interpret her material, Ostrom makes extensive use of concepts from non-cooperative
game theory, especially the theory of repeated games, associated with Robert Aumann, a reci-
pient of the 2005 Prize in Economic Sciences. As early as 1959, Aumann proved remarkably
powerful results concerning the extent to which patient people are in principle able to cooperate.
But it took a long time before anyone made the connection between these abstract mathematical
results and the feasibility of CPR management. Moreover, even as theorists developed such
relationships (e.g., Benhabib and Radner, 1992), their results were frequently ignored.

8
In addition to the examples provided in Ostrom (1990), see for instance the cases mentioned in Ostrom
et al. (1999). Both the case of overgrazing in Inner Asia, as documented by Sneath (1998), and the case of
inadequate modern irrigation in Nepal, as documented by Lam (1998), provide striking examples of how
common property management sometimes outperform seemingly attractive alternatives.
9
In a study of common pool management in Indian villages, Wade (1988) is another notable early effort
to generalize from a set of cases. See also the edited volumes by Berkes (1989) and Pinkerton (1989).
Among subsequent studies, Baland and Platteau (1996) is particularly noteworthy.

11 (19)

Over the years, game theorists have provided increasingly exact conditions under which full
cooperation is feasible, in highly structured settings, among individuals with both unbounded
cognitive capacity (e.g., Mailath and Samuelson, 2006) as well as small cognitive capacity (e.g.,
Nowak, 2006). Around 1990, and to some extent even today, theory had much less to say about
the level of cooperation that would be most likely among individuals with plausible cognitive
capacities in settings structured to some extent by the participants themselves. Thus, Ostroms
data could not be used to test any particular game-theoretic model. However, as we shall see,
the data provide valuable inspiration for the development of such models.
Main findings
Under plausible assumptions about the actions available to resource users, repeated game rea-
soning indicates that cooperation becomes more difficult as the size of the group of users in-
creases, or as the users time horizon decreases due, for example, to migration. These predic-
tions are largely borne out by Ostroms empirical studies. However, a more interesting question
is whether when these factors are held constant some groups of users are better able to
cooperate than others. That is, are there any design principles that can be elucidated from the
case material?

Ostrom proposes several principles for successful CPR management. Some of them are quite
obvious, at least with the benefit of hindsight. For example, (i) rules should clearly define who
has what entitlement, (ii) adequate conflict resolution mechanisms should be in place, and (iii)
an individuals duty to maintain the resource should stand in reasonable proportion to the benefits.

Other principles are more surprising. For instance, Ostrom proposes that (iv) monitoring and
sanctioning should be carried out either by the users themselves or by someone who is account-
able to the users. This principle not only challenges conventional notions whereby enforcement
should be left to impartial outsiders, but also raises a host of questions as to exactly why indi-
viduals are willing to undertake costly monitoring and sanctioning. The costs are usually
private, but the benefits are distributed across the entire group, so a selfish materialist might
hesitate to engage in monitoring and sanctioning unless the costs are low or there are direct
benefits from sanctioning. Ostrom (1990, pp. 9498) documents instances of low costs as well
as extrinsic rewards for punishing. However, from Ostrom, Walker and Gardner (1992)
onwards, she came to reject the idea that punishment is always carried out for extrinsic benefit;
intrinsic reciprocity motives also play an important role.

Another nontrivial design principle is that (v) sanctions should be graduated, mild for a first
violation and stricter as violations are repeated.

Ostrom also finds that (vi) governance is more successful when decision processes are demo-
cratic, in the sense that a majority of users are allowed to participate in the modification of the
rules and when (vii) the right of users to self-organize is clearly recognized by outside authorities.

12 (19)

In Governing the Commons, as well as in later publications, Ostrom documents and discusses
such principles and why they contribute to desirable outcomes. Even though these design prin-
ciples do not provide an easy solution to the often complex policy problems involved, in cases
where they are all heeded, collective action and monitoring problems tend to be solved in a
reinforcing manner (Ostrom, 2005, p. 267).

Ostrom furthermore identifies some design principles that are applicable even under privatiza-
tion or state governance. For example, positive outcomes always seem easier to reach when
monitoring is straightforward, and Ostrom carefully sets forth how monitoring can be simplified
in common pools. For example, calendar restrictions (hunting seasons, etc.) are often much
easier to monitor than quantity restrictions.

A final lesson from the many case studies is that large-scale cooperation can be amassed gradu-
ally from below. Appropriation, provision, monitoring, enforcement, conflict resolution and
governance activities can all be organized in multiple layers of nested enterprises. Once a group
has a well-functioning set of rules, it is in a position to collaborate with other groups, eventually
fostering cooperation between a large number of people. Formation of a large group at the out-
set, without forming smaller groups first, is more difficult.

Needless to say, Ostroms research also prompts a number of new questions. It is important to
investigate whether cooperation must be built from below, or whether other approaches are
feasible when dealing with large-scale problems. In recent years, Ostrom has accordingly taken
on the extensive question of whether the lessons from small local commons can be exploited to
solve the problems of larger and even global commons (e.g., Dietz, Ostrom, and Stern, 2003).

A related question, which echoes Williamsons attempt to link governance to asset cha-
racteristics, is to explore in more detail the relationship between the underlying technology
and/or tastes and the mode of governance (e.g., Copeland and Taylor, 2009).
Experiments
Since Ostroms initial research was based on field studies, her theorizing was inductive. While
the ensuing propositions were put to the test as new field studies emerged, the multidimensio-
nality of relevant factors precludes a rejection of the theory. In order to test individual propo-
sitions more directly, Ostrom and colleagues have therefore conducted a series of laboratory
experiments on behavior in social dilemmas; see Ostrom, Gardner, Walker (1994). Building on
the seminal experimental work of Dawes, McTavish, and Shaklee (1977) and Marwell and
Ames (1979, 1980) as well as ensuing work by economists and psychologists in the 1980s
Ostrom examined whether findings from the field could be recreated in the more artificial, but
controlled, laboratory environment.

13 (19)

In a typical experiment, a number of subjects interact during several periods, without knowing
exactly which period is the last. In each period, each subject can contribute towards a public
good. Across the interesting decision range, an individuals marginal cost of contribution is
larger than his marginal benefit, but smaller than the total benefit. Thus, a rational and selfish
individual would contribute nothing if the game were played in a single period only.

An important feature of the experiments was the introduction of sanctioning possibilities. In one
experimental treatment, subjects would be informed about the contributions of all the other
subjects in the previous round and be allowed to selectively punish each of the opponents.
A punishment would be costly to both the punished opponent and the punisher. Thus, a rational
and selfish individual would not punish if the game were played for one period only.

With the notable exception of Yamagishi (1986), previous experimental work did not allow
subjects to punish each other selectively. Since punishment appears to be crucial in the field, it
is of considerable interest to see whether it matters in the laboratory and, if so, why. Ostrom,
Walker and Gardner (1992) find that many subjects engage in directed punishment in the labor-
atory, but that such punishment works much better if subjects are allowed to communicate than
when they are not (Yamagishi, 1986, had confined attention to the no-communication condition).

These laboratory findings have triggered a large volume of subsequent experimental work. For
example, Fehr and Gchter (2000) show that punishment occurs and disciplines behavior in
social dilemmas even if the experimental game has a known horizon and subjects are unable to
gain individual reputations for punishing, thereby suggesting that people get intrinsic pleasure
from punishing defectors. Masclet et al. (2003) demonstrate that purely symbolic sanctions can
be almost as effective as monetary sanctions, suggesting that induviduals fear of explicit disap-
proval is a major reason why sanctions matter. Kosfeld, Okada, and Riedl (2009) show that
subjects voluntarily establish large groups that impose internal sanctions on cheating members,
but that small groups tend to dissemble even if dissembling harms members as well as outsiders
(indeed, the threat that small groups will collapse is presumably what keeps the group large).

These experiments in turn reinforce Ostroms argument that a proper understanding of human
cooperation requires a more nuanced analysis of individuals motives than has been usual in
economic science, especially regarding the nature and origin of reciprocity (Ostrom, 1998,
2000). Such models have been developed at a daunting pace during the last two decades, partly
inspired by Ostroms call. An introduction to the relevant social preference (proximate cause)
literature is given by, e.g., Fehr and Schmidt (2006); for introductions to the evolutionary (ulti-
mate cause) literature, see, e.g., Sethi and Somanathan (2003) and Nowak (2006).

Ostroms evidence from the field and from the laboratory also affects what set of games
theorists should study in order to grasp the logic of the collective action observed in the field.
The conventional parable of a repeated n-person prisoners dilemma has produced a wealth of
conceptual insights, but this parable is too sparse to adequately capture the directed punishments

14 (19)

and rewards that are used in the observed common pools. Sethi and Somanathan (1996) is the
seminal study of cooperation in a CPR game (which has the essential characteristics of an
n-person Prisoners Dilemma) that allows each player to punish any other player after each
round of CPR interaction.
Final remarks
Over the last few decades, economic governance research has emerged as an important area
of inquiry. The works of Elinor Ostrom and Oliver Williamson have greatly contributed to
its advancement.

Oliver Williamson has formulated a theory of the firm as a conflict resolution mechanism and
Elinor Ostrom has subsequently demonstrated how self-governance is possible in common
pools. At first glance, these contributions may seem somewhat disparate. However, in stark
contrast to areas of economic analysis which presume that contracts are complete and automati-
cally enforced by a smoothly functioning legal system, both Ostrom and Williamson address
head on the problems of drawing up and enforcing contracts.

Let us also note that Ostroms and Williamsons endeavors are vital parts of a broader attempt
to understand the problems of conflict resolution and contract enforcement (Dixit, 2004, 2009).
Some of this work relies on verbal theorizing and historical examples (e.g., that of the Laureate
in Economic Sciences Douglass North, 1990, 2005). Other contributions have used repeated
game models to study associations such as merchant guilds (Greif, Milgrom, and Weingast,
1994), as well as the emergence of third parties, such as law merchants and private judges
(Milgrom, North, and Weingast, 1990), and even the Mafia (Dixit, 2003). For a broad
perspective on the emergence of institutions that support market exchange, see Greif (2006a,b).

15 (19)

References
Abreu, D. and F. Gul (2000): Bargaining and Reputation, Econometrica 68, 85117.
Aghion, P. and P. Bolton (1992): An Incomplete Contracts Approach to Financial Contracting, Review
of Economic Studies 59, 473494.
Anderlini, L., and L. Felli (1994): Incomplete Written Contracts, Quarterly Journal of Economics 109,
10851124.
Aumann, R. (1959): Acceptable Points in General Cooperative n-Person Games, in R. D. Luce and A. W.
Tucker (eds.), Contributions to the Theory of Games IV, Annals of Mathematics Study 40, 287324,
Princeton NJ: Princeton University Press.
Babcock, L. and G. Loewenstein (1997): Explaining Bargaining Impasse: The Role of Self-Serving
Biases, Journal of Economic Perspectives 11 (Winter), 109126.
Baker, G., R. Gibbons and K.J. Murphy (2002): Relational Contracts and the Theory of the Firm,
Quarterly Journal of Economics 117, 3984.
Baland, J.-M. and J.-P. Platteau (1996): Halting Degradation of Natural Resources: Is There a Role for
Rural Communities?, Oxford: Clarendon Press.
Barnard, C.I. (1938): Functions of the Executive, Cambridge MA: Harvard University Press.
Benhabib, J. and R. Radner (1992): Joint Exploitation of a Productive Asset, Economic Theory 2, 155190.
Berkes, F. (ed.) (1989): Common Property Resources. Ecology and Community-Based Sustainable
Development, London: Belhaven Press.
Clark, C.W. (1976): Mathematical Bioeconomics: The Optimal Management of Renewable Resources,
New York: Wiley.
Coase, R. (1937): The Nature of the Firm, Economica 4, 386405.
Coase, R. (1960): The Problem of Social Cost, Journal of Law and Economics 3, 144.
Compte, O. and P. Jehiel (2002): On the Role of Outside Options in Bargaining with Obstinate Parties,
Copeland, B.R. and M.S. Taylor (2009): Trade, Tragedy, and the Commons, American Economic Review
99, 725749.
Crawford, V.P. (1982): A Theory of Disagreement in Bargaining, Econometrica 68, 85117.
Crawford, V.P. (1988): Long-Term Relationships Governed by Short-Term Contracts, American
Dasgupta, P.S. and G.M. Heal (1979): Economic Theory and Exhaustible Resources, Cambridge:
Cambridge University Press.
Dawes, R., J. McTavish and H. Shaklee (1977): Behavior, Communication, and Assumptions about Other
Peoples Behavior in a Commons Dilemma Situation, Journal of Personality and Social Psychology 35,
111.
Dietz, T., E. Ostrom and P.C. Stern (2003): The Struggle to Govern the Commons, Science 302, 19071912.
Dixit, A. (2003): On Modes of Economic Governance, Econometrica 71, 449481.
Dixit, A. (2004): Lawlessness and Economics: Alternative Modes of Governance, Princeton NJ: Princeton
University Press.
Dixit, A. (2009): Governance Institutions and Economic Activity, American Economic Review 99, 524.

16 (19)

Fehr, E. and S. Gchter (2000): Cooperation and Punishment in Public Goods Experiments, American
Fehr, E. and K. Schmidt (2006): The Economics of Fairness, Reciprocity and Altruism Experimental
Evidence and New Theories, in S.-C. Kolm and J.M. Ythier (eds.): Handbook of the Economics of Giving,
Altruism and Reciprocity, Volume 1, 615691, Amsterdam: Elsevier.
Fudenberg, D., B. Holmstrm and P. Milgrom (1990): Short-Term Contracts and Long-Term Agency
Relationships, Journal of Economic Theory 51, 131.
Garvey, G. (1995): Why Reputation Favors Joint Ventures over Vertical and Horizontal Integration: A
Simple Model, Journal of Economic Behavior and Organization 28, 387397.
Gibbons, R. (2005): Four Formal(izable) Theories of the Firm, Journal of Economic Behavior and
Organization 58, 20247.
Goldberg, V.P. (1976): Regulation and Administered Contracts, Bell Journal of Economics 7, 426448.
Gordon, H. S. (1954): The Economic Theory of a Common Property Resource: The Fishery, Journal of
Political Economy 62, 124142.
Greif, A. (2006a): Institutions and the Path to the Modern Economy: Lessons from Medieval Trade,
Cambridge: Cambridge University Press.
Greif, A. (2006b): The Birth of Impersonal Exchange: The Community Responsibility System and
Impartial Justice, Journal of Economic Perspectives 20 (Spring), 221236.
Greif, A., P. Milgrom, and B. Weingast (1994): Coordination, Commitment, and Enforcement: The Case
of the Merchant Guild, Journal of Political Economy 102, 745776.
Grossman, S. and O. Hart (1986): The Costs and Benefits of Ownership: A Theory of Vertical and Lateral
Integration, Journal of Political Economy 94, 691719.
Halonen, M. (2002): Reputation and the Allocation of Ownership, Economic Journal 112, 539558.
Hardin, G. (1968): The Tragedy of the Commons, Science 162, 12431248.
Hart, O. (1995): Firms, Contracts, and Financial Structure, Oxford: Clarendon Press.
Hart, O. and J. Moore (1989): Default and Renegotiation: A Dynamic Model of Debt, MIT Working
Paper No 520 (published in revised form in Quarterly Journal of Economics 113, 141, 1998).
Hart, O. and J. Moore (1990): Property Rights and the Theory of the Firm, Journal of Political Economy
98, 11191158.
Hart, O. and J. Moore (2008): Contracts as Reference Points, Quarterly Journal of Economics, 123, 148.
Jensen, M.C. and W.H. Meckling (1976): Theory of the Firm: Managerial Behavior, Agency Costs and
Ownership Structure, Journal of Financial Economics 3, 305360.
Joskow, P. (1985): Vertical Integration and Long-term Contracts: The Case of Coal-burning Electric
Generating Plants, Journal of Law, Economics, and Organization 1, 3380.
Joskow, P. (1987): Contract Duration and Relationship-Specific Investments: Empirical Evidence from
Coal Markets, American Economic Review 77, 168185.
Klein, B, R.G. Crawford and A.A. Alchian (1978): Vertical Integration, Appropriable Rents, and the
Competitive Contracting Process, Journal of Law and Economics 21, 297326.
Kolasky, W.J. and A.R. Dick (2003): The Merger Guidelines and the Integration of Efficiencies into
Antitrust Review of Horizontal Mergers, Antitrust Law Journal 71, 207251.
Kosfeld, M., A. Okada and A. Riedl (2009): Institution Formation in Public Goods Games, American

17 (19)

Kreps, D. (1990): Corporate Culture and Economic Theory, in J.E. Alt and K.E. Shepsle (eds.),
Perspectives on Positive Political Economy, Cambridge: Cambridge University Press.
Lafontaine, F. and M. Slade (2007): Vertical Integration and Firm Boundaries: The Evidence, Journal of
Economic Literature 45, 629685.
Lam, W.F. (1998): Governing Irrigation Systems in Nepal: Institutions, Infrastructure, and Collective
Action. Oakland: ICS Press.
Libecap, G.D. and S.N. Wiggins (1984): Contractual Responses to the Common Pool: Prorationing of
Crude Oil Production, American Economic Review 74, 8798.
Libecap, G.D. and S.N. Wiggins (1985): The Influence of Private Contractual Failure on Regulation: The
Case of Oil Field Unitization, Journal of Political Economy 93, 690714.
Mailath, G.J., and L. Samuelson (2006): Repeated Games and Reputations: Long-Run Relationships,
Oxford: Oxford University Press.
Marwell, G. and R. Ames (1979): Experiments on the Provision of Public Goods I: Resources, Interest,
Group Size, and the Free-Rider Problem, American Journal of Sociology 84, 13351360.
Marwell, G. and R. Ames (1980): Experiments on the Provision of Public Goods II: Provision Points,
Stakes, Experience, and the Free-Rider Problem, American Journal of Sociology 85, 926937.
Masclet, D., C. Noussair, S. Tucker and M.-C. Villeval (2003): Monetary and Nonmonetary Punishment
in the Voluntary Contributions Mechanism, American Economic Review 93, 366380.
Masten, S. E. (1988): A Legal Basis for the Firm, Journal of Law, Economics, and Organization 4, 181198.
Masten, S. E. (ed.) (1996): Case Studies in Contracting and Organization, New York: Oxford University Press.
Milgrom, P. and J. Roberts (1990): Bargaining Costs, Influence Costs, and the Organization of Economic
Activity, in J. E. Alt and K. A. Shepsle (eds.), Perspectives on Positive Political Economy, Cambridge:
Milgrom, P., D.C. North and B.R. Weingast (1990): The Role of Institutions in the Revival of Trade: The
Law Merchant, Private Judges, and the Champagne Fairs, Economics and Politics 2, 123.
North, D.C. (1990): Institutions, Institutional Change and Economic Performance, Cambridge:
North, D.C. (2005): Understanding the Process of Economic Change, Princeton NJ: Princeton University Press.
Novak, S. and S. Eppinger. (2001): Sourcing by Design: Product Architecture and the Supply Chain,
Management Science 47, 189204.
Nowak, M. (2006): Five Rules for the Evolution of Cooperation. Science 314, 15601563.
Olson, M. (1965): The Logic of Collective Action: Public Goods and the Theory of Groups, Cambridge
MA: Harvard University Press.
Ostrom, E. (1965): Public Entrepreneurship: A Case Study in Ground Water Management, Ph.D.
Dissertation, University of California at Los Angeles.
Ostrom, E. (1990): Governing the Commons: The Evolution of Institutions for Collective Actions,
Ostrom, E. (1998): A Behavioral Approach to the Rational Choice Theory of Collective Action:
Presidential Address, American Political Science Association, 1997, American Political Science Review
92, 122.
Ostrom, E. (1999): Coping with the Tragedies of the Commons, Annual Review of Political Science 2,
493535.

18 (19)

Ostrom, E. (2000): Collective Action and the Evolution of Social Norms, Journal of Economic
Perspectives 14, Summer, 137158.
Ostrom, E. (2005): Understanding Institutional Diversity, Princeton NJ: Princeton University Press.
Ostrom, E., R. Gardner and J. Walker, (1994): Rules, Games, and Common-Pool Resources, Ann Arbor
MI: University of Michigan Press.
Ostrom, E., J. Walker and R. Gardner (1992): Covenants with and without a Sword: Self-Governance is
Possible, American Political Science Review 86, 404417.
Ostrom, E. et al. (1999): Revisiting the Commons: Local Lessons, Global Challenges, Science 284, 278282.
Pigou, A. C. (1920): The Economics of Welfare, New York: Macmillan.
Pinkerton, E. (ed.) (1989): Co-operative Management of Local Fisheries. New Directions for Improved
Management and Community Development, Vancouver BC: University of British Columbia Press.
Pollak, R.A. (1985): A Transaction Cost Approach to Families and Households, Journal of Economic
Literature 23, 581608.
Schelling, T. (1956): An Essay on Bargaining, American Economic Review 46, 281306.
Segal, I. (1999): Complexity and Renegotiation: A Foundation for Incomplete Contracts, Review of
Economic Studies 66, 5782.
Sethi, R. and E. Somanathan (1996): The Evolution of Social Norms in Common Property Resource Use,
American Economic Review 86, 766788.
Sethi, R. and E. Somanathan (2003): Understanding Reciprocity, Journal of Economic Behavior and
Organization 50, 127.
Shelanski, H. A. and P. G. Klein (1995): Empirical Research in Transaction Cost Economics: A Review
and Assessment, Journal of Law, Economics and Organization 11, 335361.
Simester, D. and M. Knez (2002): Direct and Indirect Bargaining Costs and the Scope of the Firm,
Journal of Business 75, 283304.
Simon, H. (1951): A Formal Theory of the Employment Relation, Econometrica 19, 293305.
Simon, H. (1955): A Behavioral Model of Rational Choice, Quarterly Journal of Economics 69, 99118.
Sneath, D. (1998): State Policy and Pasture Degradation in Inner Asia, Science 281, 11471148.
Lam, W.F. (1998): Governing Irrigation Systems in Nepal: Institutions, Infrastructure, and Collective
Action, Oakland CA: ICS Press.
Tirole, J. (2009): Cognition and Incomplete Contracts, American Economic Review 99, 265294.
Wade, R. (1988): Village Republics: Economic Conditions for Collective Action in South India,
Warming, J. (1911): Om grundrente av fiskegrunde, Nationalkonomisk Tidsskrift 49, 499505. English
translation: On Rent of Fishing Grounds, History of Political Economy 15, 391396.
Whinston, M. D. (2003): On the Transaction Cost Determinants of Vertical Integration, Journal of Law,
Economics and Organization 19, 123.
Williamson, O. E. (1968): Economies as an Antitrust Defense: The Welfare Tradeoffs, American
Williamson, O. E. (1971): Vertical Integration of Production: Market Failure Considerations, American
Economic Review, Papers and Proceedings 61, 112123.
Williamson, O. E. (1975): Markets and Hierarchies, New York: Free Press.

19 (19)

Williamson, O. E. (1985): The Economic Institutions of Capitalism, New York: Free Press.
Williamson, O. E. (1988): Corporate Finance and Corporate Governance, Journal of Finance 43, 567591.
Williamson, O. E. (2000): The New Institutional Economics: Taking Stock, Looking Ahead, Journal of
Economic Literature 38, 595613.
Yamagishi, T. (1986): The Provision of a Sanctioning System as a Public Good, Journal of Personality
and Social Psychology 51, 110116.

13 October 2008

____________________________________________________________________________________________________
Information Department, Box 50005, SE-104 05 Stockholm, Sweden
Phone: +46 8 673 95 00, Fax: +46 8 15 56 70, E-mail: info@kva.se, Website: www.kva.se

Scientific background on
the Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2008

Trade and Geography Economies of Scale,
Differentiated Products and Transport Costs

Compiled by the Prize Committee of the Royal Swedish Academy of Sciences

Introduction

Over the centuries, international trade and the location of economic activity have been at the
forefront of economic thought. Even today, free trade, globalization, and urbanization remain
as commonplace topics in the popular debate as well as in scholarly analyses. Traditionally,
trade theory and economic geography evolved as separate subfields of economics. More
recently, however, they have converged become more and more united through new
theoretical insights, which emphasize that the same basic forces simultaneously determine
specialization across countries for a given international distribution of factors of production
(trade theory) and the long-run location of those factors across countries (economic
geography).

As of the mid-1970s, trade theory was based on the notion of comparative advantage.
Countries were assumed to trade with each other because of differences in some respect
either in terms of technology, as assumed by David Ricardo in the early 19
th
century, or in
terms of factor endowments, according to the Heckscher-Ohlin theory developed in the 1920s.
The latter was exposited by Bertil Ohlin in his 1933 monograph Interregional and
International Trade; Ohlin was awarded the 1977 Economics Prize for his contributions to
trade theory.

These theories provided good explanations of the trade patterns in the first half of the 20
th

century. But as many researchers began to observe, comparative advantage seemed less
relevant in the modern world. Today, most trade takes place between countries with similar
technologies and similar factor proportions; quite similar goods are often both exported and
imported by the same country. At least among the richer countries, intra-industry trade
whereby, for instance, a country both exports and imports textiles came to dominate relative
to inter-industry trade whereby, for instance, a country exports textiles and imports
agricultural products. Under such circumstances, how could intra-industry trade be explained?
The traditional view, that a given country would have a comparative advantage in terms of
technology or factor endowments when producing a particular type of textile, seems far-
fetched as an explanation.

Many trade theorists came up with interpretations of the observed patterns of intra-industry
trade by referring to economies of scale. In an influential book, Grubel and Lloyd (1975)
2

documented the large amount of intra-industry trade and argued that it could be explained by
economies of scale. If the average cost of producing a given good (for instance, a particular
make of car) would decline with total production, then it could be optimal to split up
production so that countries specialize in different makes of cars. Such specialization would
make sense even without differences in factor proportions and technology. This idea was
however not entirely new in 1975; in fact, the role of economies of scale in explaining trade
patterns already been recognized by Ohlin:

"[T]he advantages of producing a large quantity of a single commodity instead of a little
of all commodities must lead to interregional trade ... insofar as the market for some
articles within each region is not large enough to permit the most efficient scale of
production, division of trade and labor will be profitable. Each region will specialize on
some of these articles and exchange them for the rest ... The tendency toward
specialization because of differences in factor endowments is reinforced by the
advantages of large-scale production. The location of an industry in one region and not
in another might simply be due to chance ... Thus, all interregional trade, whether due to
the one cause or the other, might be regarded as a substitute for geographical mobility of
productive factors."
1

It was not until the late 1970s, with the development of what is now known as the new trade
theory, that these insights were integrated into a stringent and unified theoretical framework.
Such a framework is a prerequisite for systematic empirical work, in turn necessary for
studying the relative importance of different determinants of trade, as well as for systematic
evaluation of policy proposals.

In the field of economic geography, the key questions concern migration flows of individuals
and firms across the geographic landscape, how urban agglomerations arise, and how cities
themselves are spatially organized (urban economics). Here as well, it had long been
recognized that economies of scale are decisive role for the location of economic activity. As
of the 1950s, there was a substantial literature on the effects of the trade-off between
increasing returns in production and costs of transportation on agglomeration and the growth
of cities (Harris, 1954; Myrdal, 1957; Hirschman, 1958; Pred, 1966). But as in the field of
trade theory, these insights were not supported by well-articulated models especially models
that allowed a general-equilibrium analysis, where the location of both consumers and firms
was explained in the model.

1
Scattered phrases from Chapter III of Ohlin (1933), quoted from Krugman (1999).
3

By the late 1980s, researchers had begun to integrate economies of scale into general
equilibrium models of location and trade, thereby giving precision to the verbal analyses of
earlier researchers and adding important new insights. In the resulting work, now commonly
known as the new economic geography, economic geographers made use of the new tools,
along with economists who took a renewed interest in the field. Several researchers took part
in these developments, but the most influential contributions were made by Paul Krugman.

Krugman has published a large number of important articles and monographs in the fields of
both trade and geography. In particular, he made the initial key contributions. He wrote the
first article in the trade theory, soon followed by another influential paper that extended his
initial analysis (Krugman, 1979a and 1980). Further, Krugman (1991a) is commonly viewed
as the starting point of new economic geography. In fact, the seeds of the new economic
geography can already be found in his 1979 (a) article which, in its final section, argues that
patterns of migration can be analyzed within the same framework as the new trade theory.
While this article had an immediate impact on the trade literature, it would take more than ten
years for the final section, on migration and agglomeration, to have an influence on the
geography literature kindled by Krugman himself, in the 1991 (a) paper.

In what follows, we discuss Krugmans contributions to trade and geography. We begin by
considering the key common elements economies of scale, monopoly power, and demand
for variety of both theories, using the setting of Krugmans 1979 (a) paper. We then
examine, separately and in more detail, his work on trade and geography, with special
emphasis in each case on how Krugmans research transformed the literature.

The Basic Model: Economies of Scale and Monopoly Power

In the late 1970s, several researchers Krugman (1979a, 1980), Dixit and Norman (1980,
Chapter 9) and Lancaster (1980) independently formalized the idea that economies of scale
and imperfect competition can give rise to trade even in the absence of comparative
advantage.
2
In related contributions, Ethier (1979, 1982) developed models of intra-industry
trade based on economies of scale in intermediate rather than final goods. It was Paul

2
Chapter 9 in Dixit and Normans book builds on unpublished notes by Norman from 1976 and 1978.
4

Krugman who most clearly and forcefully articulated the revolutionary nature of this new
approach for the theory of international trade. His short paper in the Journal of International
Economics, entitled Increasing Returns, Monopolistic Competition and International Trade
(1979a), is twofold. It contains not only a new trade theory that allows us to explain observed
patterns of intra-industry trade, but also the seeds of a new economic geography where the
location of production factors and economic activity can be stringently analyzed within the
framework of a general-equilibrium model. Remarkably, the paper achieves all of this in only
ten pages, and in a very simple and transparent fashion. The model is extremely simple. There
is only one production factor, and returns to scale are represented by a linear cost function
with fixed costs. But due to its simplicity, it illustrates the key mechanisms in a particularly
clear way.

The central feature in Krugmans approach is economies of scale that are internal to the firm,
i.e., the firm itself can reduce its own average cost by expanding production. Under such
conditions, markets cannot be perfectly competitive. Models of imperfect competition had
often been shunned in trade theory because of their analytical complexity. But Krugman made
use of a recent model of monopolistic competition due to Dixit and Stiglitz (1977) that turned
out to be well suited for the analysis of trade.
3
In that spirit, he assumed that there are n
different goods, and that consumers have a taste for variety that can be expressed by the
following utility function:

) (
1
=
=
n
i
i
c v U , (1)

where
i
c is consumption of the ith good and where ) ( v is an increasing, concave function.
Concavity is crucial as it expresses the taste for variety. Absent price differences between
goods, it implies that the consumer would spread his resources evenly across as many goods
as possible rather than only consume one good. In the following, we assume that ) ( v takes a
particular form:

3
Trade models with external economies of scale had earlier been developed by Matthews (1949), Kemp (1964),
Melvin (1969), Negishi (1969) and Chipman (1970).
5

1 0 , ) ( < < =
i
i
c
c v . (2)

Although this functional form was not used in Krugmans original 1979 (a) paper, it has
subsequently been widely used (not least by Krugman himself) owing to the simple and
elegant analytical solutions that it provides.
4
In the following, we refer to preferences
represented by (1) and (2) as Dixit-Stiglitz preferences.

Krugman assumes further that there is only one factor of production, namely labor (thus
giving all countries identical factor proportions by definition) and that all goods are produced
with identical cost functions. He represents increasing returns to scale by assuming that, for
any good i produced, the labor requirement
i
l is given by

0 , , > + =
i i
x l , (3)

where
i
x is the output of good i, and is a fixed cost. When all markets clear and there is
free entry then all monopolistically competitive firms will have zero profits in equilibrium.
The model can be solved for the three unknowns: the price of each good relative to labor,
w p
i
/ (where w is the wage rate), the output of each good,
i
x , and the number of goods
produced, n.

Let us first discuss some aspects of the monopolistic equilibrium of this economy and then
introduce a two-country version of the model from the perspective of trade and geography.
Consumers maximize their utility subject to their budget constraint ( w c p
i
n
i
i
=
=1
). This
leads to the first-order condition = =
j j i i
p c v p c v / ) ( ' / ) ( ' for any two goods i and j, where
denotes the shadow value of income: the consumers marginal rate of substitution between the
goods should equal the relative price. We then obtain the individuals demand function for
good i as
) 1 /( 1
) ( ) (

=

i i
p p c . Indirect utility can easily be computed and is proportional
to

) / ( p w , where

=
i
i
p p
/ ) 1 ( ) 1 /(
) ( is a price index across goods. We can thus think of
p w/ as the real wage in this economy.

4
In his 1979 paper, Krugman assumes instead that v is such that the elasticity of demand decreases with income,
which leads to slightly different conclusions and makes the exposition somewhat more complicated.
6

Turning to the monopolistic competitor, profits are maximized by setting the price so that
marginal revenue equals marginal cost. With L consumers, the firms total revenue equals
) (
i i
p Lc p and its total costs are w p Lc
i
)) ( ( + . Moreover, when the monopolist chooses
price, the shadow value of income, , can be treated as unaffected by the price, since there is a
large number of goods. It is easy to verify that when marginal revenue equals marginal cost,
the monopolist chooses a constant markup, 1/ , over marginal cost: / 1 ) /( = w p
i
. This
symmetry across goods implies that the price index in equilibrium will satisfy

/
/ ) 1 (
w n p

= . We see that the price index is decreasing in product variety, n, and thus
that the real wage and equilibrium utility are increasing in n. In this economy, new firms will
enter the market, adding product varieties, until profits equal zero. Since profits equal
w x x p
i i i
) ( + and all goods are symmetric and produced in equal quantity, this implies
that equilibrium output is given by x x
i
= ) 1 /( ) / ( . With this expression for
determining the quantity produced of each good, the variety range of goods, n, is determined
by n x L ) ( + = , since the L consumer-workers have to allocate their labor supply to n
different goods.

Now that all equilibrium prices and quantities have been determined, we can investigate how
the size of the economy, which is given by L (the number of consumer-workers), influences
the equilibrium. First, the production of each good, x , is not affected given our assumption on
the shape of the utility function, but the number of goods increases (proportionately) with L.
5

Second, per-capita consumption of each good decreases since x cL = . Third, per-consumer
welfare increases with L since, as we have seen above, real wages increase in product variety.
Thus, the larger economy allows more varieties to be produced, and this is the channel
through which increasing returns to scale operate here. More generally, and in Krugmans
own analysis, increasing returns to scale may also increase the production of each good, thus
lowering unit cost on a good-by-good basis. We are now set to examine the possibility of
trade between countries.

5
This is due to the specific utility function (2). Krugmans alternative assumption about the elasticity of the ) ( v
function implies that the production of each good increases as the economy enlarges.
7

Assume that there are two countries which are identical in preferences, technology, and size.
We can then compare two cases: autarchy, where there is no trade due to, for instance,
prohibitive transport costs, and another where the countries can trade freely at no cost. With
trade, we can regard the size of the economy as the sum of the sizes of the two countries, and
we can determine world production and consumption as if there were only one (large)
country. Thus, the number of goods produced is now larger; since the countries are identical,
the number of goods is twice what it would be under autarchy. Each country specializes in the
production of some goods, thereby exploiting the economies of scale inherent in the fixed-
cost production function. Precisely which goods are produced where is not determined,
however. Trade will amount to half of GDP of each country (in general, if the countries were
of different size, trade would be smaller), and consumers will enjoy a larger number of goods
than under autarchy. The opening-up of trade will therefore be welfare-enhancing, despite the
fact that both countries have identical technologies and factor proportions, since it increases
product variety. Under Krugmans assumption on the utility function (see footnote 4) it would
also lower unit costs good by good. The allocation of goods across countries is not determined
in the model; it may well happen that good i is produced in one country while good j is
produced in the other country, even though goods i and j are very similar (for instance, two
makes of cars). In that sense, the opening-up of trade generates intra-industry trade.

Thus, Krugman (1979a) showed that consumer preferences represented by (1), and production
costs represented by (3), can generate trade patterns consistent with real-world data. In the
final section of the paper, he discusses the implications of impediments to trade between the
two countries when migration of labor is possible. This section contains the precursor of the
new economic geography work he would pursue later. Using the simple model above, he
argues that in the absence of trade, consumer welfare will be highest in the region with the
largest labor force. This is so because with Dixit-Stiglitz preferences, as seen above, the real
wage p w/ depends positively on the number of products n, which is greater in the region
with the largest labor force. There will thus be a tendency for labor to migrate to the region
that happens to have the largest labor force, and thereby the greatest variety of products, at the
outset. This mechanism gives rise to a cumulative process, resulting in concentration and
urbanization. The model thus contains an element of (potentially dramatic) demographic
change. Such a change could, however, be mitigated by a number of features that were not
present in the 1979 model.
8

InternationalTrade

Building on Krugmans analysis, a vast literature has developed exploring the implications of
returns to scale and monopolistic competition for trade patterns in richer model settings. We
now consider some of these further research developments, including the policy and empirical
implications of the new theory.

Transport costs and trade: the home-market effect
An important force behind the growth of trade has been the decrease in transport costs. Yet
such costs were conspicuously absent in most trade models for a long time. In a second
seminal contribution, Krugman (1980) extended his 1979 model by introducing transportation
costs. For analytical convenience, these costs were assumed to be proportional to the quantity
of goods shipped to another nation (sometimes referred to as iceberg costs in the sense that
a fraction of the goods melts away before they reach their destination). This allowed him to
accord analytical precision to the home-market effect earlier discussed by Corden (1970)
according to which firms tend to concentrate, i.e., locate more than in proportion to market
size, in large markets. This explains why preferences matter for trade patterns, a point raised
earlier by Linder (1961). Countries tend to export the goods for which they have a large
domestic market.

The intuition for the home-market effect is simple. With both increasing returns and transport
costs, there is an incentive to concentrate production of a good close to its largest market. By
concentrating production in one place, scale economies can be realized, while by locating near
the largest market, transport costs are minimized. The home-market effect provides a demand
explanation as to why a country can have an advantage in the production of a specific good.
Workers will be better off in the larger economy because of a lower price level, as a smaller
fraction of total consumption is burdened by transport costs.

Further developments of the new trade theory
Krugmans 1979 and 1980 papers demonstrated that models based on the assumptions of
increasing returns to scale and monopolistic competition can explain important patterns of
trade observed in the data. The way in which these new elements interacted with the
traditional factor-proportions mechanism remained to be analyzed. Integrated models of inter-
9

industry trade (based on technology gaps and Heckscher-Ohlin differences in factor
proportions) and intra-industry trade in differentiated goods (based on increasing returns to
scale and monopolistic competition) were provided by Lancaster (1980), Dixit and Norman
(1980), Krugman (1981) and, with greater generality, by Helpman (1981) and Helpman and
Krugman (1985). Integration of the new and old trade theory was particularly important as it
led to testable predictions about cross-country differences in trade patterns. It has formed the
basis for extensive empirical research on bilateral trade flows, thereby allowing researchers to
evaluate the relevance of the new theory.

Another fact about trade patterns is that the bulk of intra-industry trade occurs in knowledge-
intensive products between highly developed countries, often in industries dominated by
multinational corporations. Such a pattern was accounted for by Helpman (1984) and
Markusen (1984), who developed theories of international trade in knowledge-intensive
production sectors dominated by multinational firms, where such firms appear as the market
response to fixed R&D costs.
6
These and other developments were incorporated and extended
in Helpman and Krugmans 1985 monograph. This book provides a comprehensive
exposition of the new trade theory and develops its implications for a variety of issues. It
remains the standard reference in the field.

The new trade theory has profoundly affected the analysis of trade policy. The theory yields
predictions about the impact of trade liberalization on trade patterns, the location of output
and factor remunerations. It can also be used for welfare analysis. Realistic models easily
become too complex to be handled analytically and estimated by econometric methods.
Instead, a vast literature of calibrated numerical models has emerged. Early examples include
Harris (1984), Dixit (1988) and Baldwin and Krugman (1988). Such models are now routinely
used, e.g. by the World Bank in assessing the effects of the WTO rounds of trade
liberalization.

Empirical relevance
As we have seen the development of the new trade theory was largely motivated by the
inability of existing models to account for observed empirical patterns, such as the dominance

6
These models are precursors to later studies of endogenous growth by Grossman and Helpman (1989, 1991),
where innovations are endogenously produced by profit-maximizing firms, and innovations typically
originating in the most highly developed economies are spreading to the rest of the world through trade.
10

of intra-industry trade in the trade between developed countries, and the increase in intra-
industry trade resulting from trade liberalization (e.g. the EEC agreement from 1959).
Krugmans 1979 (a) simple one-factor model was able to explain these observations. By
integrating the new model and the traditional factor-proportions theory, it was possible to
formulate more specific hypotheses. The integrated model states that (i) the volume of trade
between any two countries should increase with the difference in relative factor endowments
and decrease with the difference in country size; and (ii) the share of intra-industry trade in
the total trade between two countries should depend negatively on the difference in capital-
labor ratios and positively on size dispersion. Taken literally, the integrated model based on
Dixit-Stiglitz preferences (2) and a specified production function even gives predictions about
the exact functional forms of the relations between bilateral trade flows, factor endowments
and country sizes.

The richness of model predictions inspired a lively empirical literature. Early empirical tests
by Helpman (1987) were supportive of the new trade theory. Some later studies, however,
showed mixed results; see Leamer and Levinsohn (1995) for a review of the empirical
literature as of the early 1990s. For instance, Hummels and Levinsohn (1993, 1995) found
that the data on trade flows fit the theory almost too well, when they analyzed both trade
between developed countries (for which consumer preferences for product variety and trade in
differentiated products were reasonable descriptions of reality) and trade between less
developed countries (where trade in monopolistically produced, differentiated goods can
hardly be of much importance). Their conclusion was that something else, besides the factor
endowments emphasized by the old theory, and the increasing returns and differentiated
products emphasized by the new theory, must lie behind a large part of real-world trade
flows. More recent studies tend to be in line with the theoretical predictions. Antweiler and
Trefler (2002) note that the model exposited in Helpman and Krugman (1985) gives
predictions regarding the factor content of exports and imports, depending on the degree of
scale economies. They found that allowing for scale economies improved the fit of the model,
and that around one third of all industries could be characterized by increasing returns to
scale. Similarly, Evenett and Keller (2002) conclude that trade patterns are best explained by
a combination of increasing returns to scale and factor proportions.

11

Economic Geography

It has long been recognized that factor mobility and trade may act as substitutes for one
another. Impediments to trade could lead to factor-price differences that would induce
migration of labor and capital. This had already been analyzed in a Heckscher-Ohlin world by
Mundell (1957). As mentioned above, the final section of Krugman (1979a) considers factor
mobility in a world of differentiated products and monopolistic competition. In the absence of
trade, the larger region would offer its inhabitants higher welfare due to a greater variety of
products, thereby providing incentives for migration. The incentives for migration would be
stronger, the greater the number of people that moved to the larger region and, absent any
impediments to migration, the whole population would in equilibrium end up in the region
that happened to be largest at the outset. Note that such a process could be triggered solely by
the initial size difference, if there were no inherent differences between regions. If the regions
differed, e.g. in labor productivity, it would still be possible to end up in an inefficient
equilibrium under certain initial conditions. If the region with the lowest exogenous
productivity was given a head start by a larger initial market size, then migration might lead to
an equilibrium with the entire population concentrated in the region with low productivity.

It would take another twelve years until these ideas were developed into the so-called core-
periphery model of Krugman (1991a), the starting point of the new economic geography. To
assess the far-reaching nature of this model, some background on spatial economics is called
for.

Spatial economics some background
Combining space and competitive equilibrium is a major challenge.
7
Trade theory has
traditionally taken the heterogeneity of space as exogenously given and analyzed the trade
patterns resulting from differences in factor proportions and technology. Much of regional
analysis, starting with Marshall, has recognized that agglomeration of economic activity is
driven by economies of scale, while assuming that the scale economies are external to the
individual firm but internal to the industry or the city, and hence consistent with perfect

7
In fact, Starrett (1978) formulated a spatial impossibility theorem, which states that there cannot exist a spatial
equilibrium involving transportation in an Arrow-Debreu economy with homogeneous space and costly
transportation. If all activities were perfectly divisible there would exist an equilibrium in which each region
operated as an autarchy. In practice, however, this is an uninteresting case, since it implies no trade and no
division of labor.
12

competition. An example would be pure technological spillovers between firms in a well-
defined area due to transfers of knowledge useful in production. A shortcoming of this
approach is that the nature of these external scale economies are typically left rather vague
and it is, therefore, hard to measure the externalities in order to test and use the theory
empirically. The new economic geography initiated by Krugman broke with this tradition by
assuming internal economies of scale and imperfect competition. Agglomeration is then
driven by pecuniary externalities mediated through market prices as a large market allows
greater product variety and lower costs. When a household or a firm transacts in a market,
there is generally an effect on other agents through impact of the transaction on the price. In
those cases where there is some form of market imperfection at the outset, this effect can be
viewed as an externality. In Krugmans work, the presence of scale economies and of
monopolistic competition implies a market imperfection, so that the externality can be traced
to these fundamental features of the economy.

The forerunning contribution built on external economies of scale is Henderson (1974). Here
an industry-specific externality in production causes the marginal cost of a firm to be
declining in the level of industry output within an entire city. Equilibrium city size then
depends on the trade-off between this externality, which determines the agglomeration gains,
and the costs of spatial concentration of activity (such as commuting costs). Cities tend to
specialize by industry and those industries in which external scale economies are more
substantial tend to be concentrated in larger cities.

A more elaborate model based on Marshallian externalities was later provided by Fujita and
Ogawa (1982). Whereas Henderson assumed cities to be monocentric around a central
business district (CBD) as in the classical models of von Thnen (1826) and Alonso (1964)
Fujita and Ogawa solve for land prices, wages and the equilibrium allocation of land to
production and housing, while allowing for the possibility of secondary business districts in
coexistence with the CBD. A key finding is that cities may undergo drastic structural changes
when transport costs and other key parameters change. The impact of the literature which
relies on the external-economies-of-scale assumption may have been limited by the lack of
explicit micro foundations. Are the externalities due to knowledge spillovers, thin markets for
specialized inputs, backward and forward linkages in the production chain, local public goods
or to other factors? This is an area of active research; see Duranton and Puga (2004) for an
13

account of different theoretical approaches and Rosenthal and Strange (2004) for an
assessment of the empirical literature.

The new economic geography the core-periphery model
The more recent development of economic geography builds primarily on models of internal,
as opposed to external, economies of scale and monopolistic competition. The seminal
contribution that launched the new economic geography is the core-periphery model
developed in Krugman (1991a). Important precursors to Krugmans analysis were published
by Abdel-Rahman (1988) and Fujita (1988), who developed models of location within an
agglomeration based on Dixit-Stiglitz monopolistic competition and derived equilibrium
patterns of location. In these models, however, there is no agricultural sector and no migration
across regions.
8

More than half of the worlds population lives in cities. In most countries, a majority of the
population lives in a few highly developed urban regions (the core), whereas a minority of the
population remains in a mainly agricultural hinterland (the periphery). The model of Krugman
(1991a) is aimed at explaining the prevalence of this pattern. It does so by introducing
mobility of workers (=consumers) in a model similar to the trade model with transport costs
of Krugman (1980). There are two types of products: food, which is assumed homogeneous
and produced under constant returns to scale, and manufactured goods, which exist in a large
number of varieties, each produced under increasing returns to scale and sold in
monopolistically competitive markets. Consumer preferences are expressed by a Cobb-
Douglas function in food and the subutility of manufactured goods, the latter given by a Dixit-
Stiglitz utility function with a constant elasticity of substitution; cf. eq. (2).

Consider an example of two regions with identical fundamental conditions (preferences and
production functions). Goods can be exported from one region to the other. Agricultural
goods are traded costlessly (this assumption is relaxed in later works), whereas manufactured
goods are subject to (iceberg-type) transport costs. There are two types of individuals:
manufacturing workers, who are free to migrate to the region that offers the highest utility

8
The relation between the Fujita and the Krugman models is discussed in Fujita and Krugman (2004).
Krugmans analysis also bears some resemblance to Murphy, Shleifer and Vishny (1989a.b), who analyze
enlargement of a market through simultaneous expansion of many sectors or the expansion of a leading sector
whose income is distributed widely enough. This, in turn, can create a push towards industrialization in a
developing country by making it profitable to adopt increasing-returns technologies.
14

level, and peasants, who do not migrate. The problem analyzed is how population and
economic activity will be allocated between the two regions. Will there be a concentration of
manufacturing into one region? Will the population be split between an industrialized core
and an agricultural periphery?

The model is driven by the location choices of firms and individuals. Firms have an incentive
to locate in the larger market to exploit economies of scale in production and to save on
transport costs (the home-market effect identified in Krugman, 1980). Individuals have an
incentive to move to the larger region, since it offers higher real wages and a larger variety of
goods. This tends to increase the difference in size between the markets and strengthen the
incentive to migrate both for firms and individuals. Hence, there is an element of circular
causality.

To sketch the working of the model and the determination of equilibrium, consider an initial
situation with half of the population in two regions. If the regions are identical, this is
obviously an equilibrium. But now assume that, by chance, this equilibrium is perturbed by
migration, thereby making one region slightly larger than the other. As a result of the initial
migration away from equal-sized regions, would there now be incentives for further
population divergence? If the home-market effect together with the real-wage effect is strong
enough, the initial population perturbation will stimulate further migration to the larger
region. This would set in motion a cumulative process, where migration increases the
population and the size of the market in the larger region even more, thereby raising the real
wage further and thus leading to even more migration etc. Hence, the new equilibrium may be
quite different from the original one. In this way, Krugman was able to build a strict model of
the process of circular causation discussed much earlier by Myrdal (1957), Hirschman (1958),
and others.

However, there are also counteracting forces. If all firms were located in one large region, a
single firm that moved to the periphery would become almost like a monopolist in the market
for manufacturing goods there. It would sell its product to the local farmers and to its own
employees, being disciplined in the goods market by the competition of imports which, in
turn, depends on the transport costs. Similarly, it would be disciplined in the labor market by
the fact that workers, in order to accept living in the smaller region, must have sufficiently
high real wages. Welfare for these manufacturing workers depends on the wage paid by the
15

firm, the price of the consumption good that this firm produces (which is relatively low, since
it is not burdened by transport) and the price they have to pay for imported goods that are
burdened by transport costs. All this combined may offer an incentive for not only one, but
several firms to locate to the smaller region. There may thus be a move towards a
decentralized equilibrium, where industrial production takes place in both regions. Whether
this happens or not depends on a complex interplay between transport costs, economies of
scale, and preferences.

Other factors, such as congestion costs or rising land prices in the larger region, may also lead
to dispersion instead of concentration (features that were absent from Krugmans original
model). The comparative-statics results in Krugmans analysis allow us to understand why
urbanization, and the move towards a core-periphery structure, would tend to result if
transport costs fell or technologies with increasing returns became more prevalent. Arguably,
such trends were important during the process of industrialization.

Krugmans arguments also explain why locational patterns can change catastrophically.
Assume that the initial equilibrium is symmetric, with half of the population living in one
region and half in the other. If trade costs begin to fall, there may be no immediate effects on
migration and the location of production. But once costs fall below a threshold value, a
cumulative process could be set off. This regional inequality arises endogenously even if all
exogenous conditions are equal. Thus, we see how agglomeration tendencies may suddenly
become stronger at a certain point of development. Illustrative solutions for the model,
showing how some parameter values give rise to concentration of manufacturing activity
while others give rise to decentralization, are studied in Fujita, Krugman and Venables (1999,
Chapter 5).

Further developments
The initial analysis in Krugman (1991a) reached a number of noteworthy and striking results
from a quite simple model, based on a number of specific assumptions. Much of the
subsequent literature has been devoted to investigating the importance of these assumptions.
As an example, Ottaviano, Tabuchi, and Thisse (2002) have developed an agglomeration
model with linear rather than Dixit-Stiglitz preferences. They find that the basic insights from
the original model remain unaffected. A central assumption of Krugmans original model is
that agricultural products may be traded freely at no cost. This assumption is not consistent
16

with the data; real-world transport costs appear to be at least as high for agricultural goods as
for manufactured products. This would neutralize the home-market effect (Davis, 1998). But
Fujita, Krugman, and Venables (1999, Chapter 7) have shown that there are similar
mechanisms in a world of transport costs in both sectors and differentiated agricultural
products. In such cases, a reduction in agricultural transport costs may trigger agglomeration.

The basic core-periphery model has stimulated the economic analysis of spatial issues,
thereby integrating economic geography with mainstream economics. A large literature with
an emphasis on economies of scale has developed over the last fifteen years, with Krugman
himself as a major contributor. One line of development has stressed the importance of input-
output linkages among firms as an alternative explanation for agglomeration tendencies. Such
linkages may be crucial for understanding the location of economic activities in situations of
low labor mobility as is the case, for example, among many European countries. Input-output
linkages were first analyzed by Krugman and Venables (1995) and Venables (1996). The
main idea here is that the entry of new firms in a region increases the market for upstream
suppliers (backward linkages as opposed to the forward linkages that stimulate the migration
of workers). When upstream suppliers can produce at lower costs, the costs of downstream
producers fall as well, due to increasing returns to scale. A cumulative agglomeration process
could then be driven by the interaction between upstream and downstream firms. If labor
mobility is low and the supply of labor is inelastic, then concentration of production must lead
to rising wages. This has two opposite effects: higher household income leads to an increase
in demand, whereas higher wages decrease firm profits and make relocation to the periphery
more attractive. As a result, the set of possible equilibria is quite rich.

In other work, Krugman and several co-authors have bridged the gap between the new
economic geography literature and the more traditional research in urban and regional
economics (Fujita and Krugman, 1995, Fujita, Krugman and Venables, 1999, and Fujita,
Krugman and Mori, 1999). These contributions seek, among other things, to answer the
fundamental question of where and when new cities emerge. They emphasize how the land
requirements of the agricultural sector interact with scale economies of the industrial sector.
Provided that population size is not too large, the equilibrium outcome will be a monocentric
economy, consistent with the classical work of von Thnen.

17

The empirical evidence
Recent years have seen the development of a literature that investigates the empirical validity
of the new economic geography; see Head and Mayer (2004) for a survey. There have been
findings in favour of some propositions of the theory, such as the positive relationship
between market size and wages, the relationship between market size and migration (although
the effects seem likely to remain local and thus unable to generate a core-periphery pattern in,
for example, Europe as whole), and the importance of backward linkages. There is also
evidence of the productivity benefits derived from location in densely populated areas,
although identifying the mechanism that drives this pattern remains a research challenge.
Evidence on the home-market effect is mixed, as is the evidence on the relationship between
agglomeration and the importance of both increasing returns to scale and trade costs.

Other contributions

Apart from his work discussed so far, Paul Krugman has made important contributions in
other areas. In trade theory, he has also analyzed so-called strategic trade policy, i.e., the
incentives for one country to affect its terms of trade by introducing some barriers to trade
(for instance, tariffs). This was already a well-established field within the traditional theory
with perfect competition among firms. When trade is brought about by specialization due to
economies of scale as described in the new trade literature, strategic trade policy is still
relevant, and the policy becomes closely connected to issues that involve the regulation of
industry. This insight was exploited in theoretical work initiated by Brander (1981) and
Spencer and Brander (1983). Krugman also contributed to this literature in Brander and
Krugman (1983). The policy conclusions of this approach are explored in the Helpman and
Krugman (1989) monograph, Trade Policy and Market Structure. This work synthesizes the
burgeoning new literature and analyzes a variety of policy issues that arise in models based on
the new trade theory. It is shown that these models lead to qualitatively new and interesting
effects of standard trade policies: protection could reduce domestic output, import subsidies
could improve the terms of trade, and tariffs could reduce domestic prices.

Krugman has also made important contributions to the analysis of international monetary
economics. A framework of analysis that set a new standard in the study of currency crises
was proposed in Krugman (1979b). Here, he assumed that a government is trying to maintain
18

a fixed exchange rate despite some fundamental imbalance (for instance, the country has a
higher long-run inflation rate as compared to the rest of the world) that makes such a peg
impossible to maintain in the long run. By buying and selling currency in large amounts, the
government can maintain the fixed exchange rate in the short run. Krugman analyzed how
the expected future depletion of the governments currency reserve would be taken into
account by rational investors, so as to ignite a speculative, early attack on the countrys
currency. Krugmans simple model captured the essential mechanism of currency crises in a
way that has inspired considerable later research.

In a related strand of research, Krugman analyzed the movement of exchange rates within
target zones. Such zones are relevant when central banks pursue exchange-rate policies which
dictate that a currencys value be allowed to float within a specified band, such as those
stipulated by the European Exchange Rate Mechanism (ERM) that preceded the adoption of
the Euro as a common currency by many EU countries. In particular, Krugman (1991b)
formulated the canonical model for the analysis of how exchange rates will behave within
such a zone. The basic idea is that in the middle of the band, the exchange rate is equally
likely to move upwards and downwards; the expected change is thus zero. Closer to, say, the
upper end of the band, the exchange rate is more likely to move downwards than upwards,
and the expected change is thus negative. This will be taken into account of by rational
investors, and option-pricing models can be used to analyze the movements of the exchange
rate within the band. In fact, a credible band tends to stabilize exchange rate movements. A
large literature has emerged from the basic analysis in Krugman (1991b).

Conclusion

By having integrated economies of scale into explicit general equilibrium models, Paul
Krugman has deepened our understanding of the determinants of trade and the location of
economic activity. His seminal papers published in 1979 (a) and 1980 were instrumental to
the development of the new trade theory, and his 1991 (a) paper inspired the new approach to
economic geography. His monographs, co-authored with Helpman and with Fujita and
Venables, demonstrate the richness of the new theories.

19

References

Abdel-Rahman, H.M. (1988), Product Differentiation, Monopolistic Competition, and
City Size, Regional Science and Urban Economics 18, 69-86.

Alonso, W. (1964), Location and Land Use, Cambridge, MA: Harvard University Press.

Antweiler, W. and D. Trefler (2002), Increasing Returns and All That: A View from
Trade, American Economic Review 92, 93-119.

Baldwin, R.E. and P. Krugman (1988), Market Access and International Competition:
A Simulation Study of 16K Random Access Memories, in R.C. Feenstra, ed.,
Empirical Models for International Trade, Cambridge, MA: MIT Press.

Brander, J . (1981), Intra-Industry Trade in Identical Commodities, Journal of
International Economics 11, 1-14.

Brander, J . and P. Krugman (1983), A Reciprocal Dumping Model of International
Trade, Journal of International Economics 15, 313-321.

Chipman, J . (1970), External Economies of Scale and Competitive Equilibrium,
Quarterly Journal of Economics 85, 347-385.

Corden, W.M. (1970), A Note on Economies of Scale, the Size of the Domestic
Market and the Pattern of Trade, in I.A. McDougall and R.H. Snape, eds.,
Studies in International Economics, Amsterdam, North-Holland.

Davis, D. (1998), The Home Market, Trade and Industrial Structure, American
Economic Review 88, 1264-1276.

Dixit, A. (1988), Optimal Trade and Industrial Policy for the U.S. Automobile
Industry, in R.C. Feenstra, ed., Empirical Models for International Trade,
Cambridge, MA: MIT Press.

Dixit, A. and V. Norman (1980), Theory of International Trade: A Dual General
Equilibrium Approach, Cambridge, Cambridge University Press.

Dixit, A. and J . Stiglitz (1977), Monopolistic Competition and Optimum Product
Diversity, American Economic Review 67, 297-308.

Duranton, G. and D. Puga (2004), Micro-Foundations of Urban Agglomeration
Economics, in J .V. Henderson and J .-F. Thisse, eds., Handbook of Regional and
Urban Economics, Vol. 4, Amsterdam: Elsevier North-Holland.

Ethier, W.J . (1979), Internationally Decreasing Costs and World Trade, Journal of
International Economics 9, 1-25.

Ethier, W.J . (1982), National and International Ieturns to Scale in the Modern Theory
of International Trade, American Economic Review 72, 225-238.

20

Evenett, S.J . and W. Keller (2002), On Theories Explaining the Success of the Gravity
Equation, Journal of Political Economy 110, 281-316.

Fujita, M. (1988), A Monopolistic Competition Model of Spatial Agglomeration:
Differentiated Product Approach, Regional Science and Urban Economics 18,
87-124.

Fujita, M. and H. Ogawa (1982), Multiple Equilibria and Structural Transition of Non-
monocentric Urban Configurations, Regional Science and Urban Economics 12,
161-196.

Fujita, M. and P. Krugman (1995), When is the Economy Monocentric? Von Thnen
and Chamberlin Unified, Regional Science and Urban Economics 25, 505-528.

Fujita, M. and P. Krugman (2004), The New Economic Geography: Past, Present and
the Future, Papers in Regional Science 83, 139-164.

Fujita, M., P. Krugman and T. Mori (1999), On the Evolution of Hierarchical Urban
Systems, European Economic Review 43, 209-251.

Fujita, M., P. Krugman and T. Venables, (1999), The Spatial Economy: Cities, Regions
and International Trade, Cambridge, MA: MIT Press.

Grossman, G. and E. Helpman (1989), Product Development and International Trade,

Grossman, G. and E. Helpman (1991), Innovation and Growth in the Global Economy,

Grubel, H. and P. Lloyd (1975), Intra-industry Trade, London, McMillan.

Harris, C. (1954), The Market as a Factor in the Localization of Production, Annals of
the Association of American Geographers 44, 315-348.

Harris, R. (1984), Applied General Equilibrium Analysis of Small Open Economies
with Scale Economies and Imperfect Competition, American Economic Review
74, 1016-1032.

Head, K. and T. Mayer (2004) The Empirics of Agglomeration and Trade, in J . V.
Henderson and J . F. Thisse, eds., Handbook of Regional and Urban Economic,
Vol. 4, Amsterdam: Elzevier North-Holland.

Helpman, E. (1981), International Trade in the Presence of Product Differentiation,
Economies of Scale, and Monopolistic Competition, Journal of International

Helpman, E. (1984), A Simple Theory of International Trade with Multinational
Corporations, Journal of Political Economy 94, 451-471.

21

Helpman, E. (1987), Imperfect Competition and International Trade: Evidence from
Fourteen Industrial Countries, Journal of the Japanese and International
Economies 92, 451-471.

Helpman, E. and P. Krugman (1985), Market Structure and Foreign Trade: Increasing
Returns, Imperfect Competition, and the International Economy, Cambridge, MA:
MIT Press.

Helpman, E. and P. Krugman (1989), Trade Policy and Market Structure, Cambridge,
MA: MIT Press.

Henderson, J .V. (1974), The Sizes and Types of Cities, American Economic Review
64, 640-656.

Hirschman, A.O. (1958), The Strategy of Development, New Haven, CN: Yale
University Press.

Hummels, D. and J . Levinsohn (1993), Product Differentiation as a Source of
Comparative Advantage, American Economic Review Papers and Proceedings
83, 445-449.

Hummels, D. and J . Levinsohn (1995), International Trade and Monopolistic
Competition: Reconsidering the Evidence, Quarterly Journal of Economics 110,
799-836.

Kemp, M. (1964), The Pure Theory of International Trade, Englewood Cliffs, NJ :
Prentice-Hall.

Krugman, P. (1979a), Increasing Returns, Monopolistic Competition and International

Krugman, P. (1979b), A Model of Balance-of-Payment Crises, Journal of Money,
Credit and Banking 11, 311-325.

Krugman, P. (1980), Scale Economies, Product Differentiation, and the Pattern of
Trade, American Economic Review 70, 950-959.

Krugman, P. (1981), Intra-industry Specialization and the Gains from Trade, Journal

Krugman, P. (1991a), Increasing Returns and Economic Geography, Journal of

Krugman, P. (1991b), Target Zones and Exchange Rate Dynamics, Quarterly Journal
of Economics 106, 669-682.

Krugman, P. (1999), Was it All in Ohlin?, unpublished notes.

Krugman, P. and A. Venables (1995), Globalization and the Inequality of Nations,
Quarterly Journal of Economics, 110, 857-880.
22

Lancaster, K. (1980), Intra-industry Trade under Perfect Monopolistic Competition,
Journal of International Economics 10, 151-171.

Leamer, E. and J . Levinsohn (1995), International Trade Theory: The Evidence, in G.
Grossman and K. Rogoff, eds., Handbook of International Economics, Vol. 3,
Amsterdam: Elzevier Science, 1339-1344.

Linder, S.B. (1961), An Essay on Trade and Transformation, New York, J ohn Wiley.

Markusen, J . (1984), Multinationals, Multi-Plant Economies, and the Gains from

Matthews, R.C.O. (1949), Reciprocal Demand and Increasing Returns, Review of Economic
Studies 17, 149-158.

Melvin, J . (1969), Increasing Returns to Scale as a Determinant of Trade, Canadian
Journal of Economics and Political Science 2, 389-402.

Mundell, R. (1957), International Trade and Factor Mobility, American Economic
Review, 47, 321-335.

Murphy, K. M., A. Shleifer, and R. W. Vishny (1989a), Income Distribution, Market
Size, and Industrialization, Quarterly Journal of Economics, 104, 537-564.

Murphy, K. M., A. Shleifer, and R. W. Vishny (1989b) Industrialization and the Big
Push, Journal of Political Economy 97, 1003-1026.

Myrdal, G. (1957), Economic Theory and Underdeveloped Regions, London:
Duckworth.

Negishi, T. (1969), Marshallian External Economies and Gains from Trade between
Similar Countries, Review of Economic Studies 36, 131-135.

Ohlin, B. (1933), Interregional and International Trade, Cambridge, MA: Harvard
University Press.

Ottaviano, G., T. Tabuchi and J .-F. Thisse (2002), Agglomeration and Trade
Revisited, International Economic Review 43, 409-436.

Pred, A. R. (1966), The Spatial Dynamics of U.S. Urban-Industrial Growth, 1800-1914:
Interpretive and Theoretical Essays, Cambridge, MA: MIT Press.

Rosenthal, S. S. and W. C. Strange (2004), Evidence on the Nature and Sources of
Agglomeration Economics, in J . V. Henderson and J .-F. Thisse, eds., Handbook
of Regional and Urban Economics, Vol. 4, Amsterdam: Elzevier North-Holland.

Spencer, B. and J . Brander (1983), International R&D Rivalry and Industrial Strategy,
Review of Economic Studies 50, 707-722.

23

Starrett, D. (1978), Market Allocations of Location Choice in a Model with Free
Mobility, Journal of Economic Theory 17, 21-37.

Venables, A. J . (1996), Equilibrium Locations of Vertically Linked Industries,
International Economic Review, 37, 341-359.

von Thnen, J .H. (1826), Der Isolierte Staat in Beziehung auf Landwirtschaft und
Nationalkonomie, Hamburg: Perthes. (English translation: The Isolated State,
Oxford: Pergamon, 1966).

15 October 2007

____________________________________________________________________________________________________

Scientific background on
the Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2007

Mechanism Design Theory

Compiled by the Prize Committee of the Royal Swedish Academy of Sciences

1 Introduction
Economic transactions take place in markets, within rms and under a host of other
institutional arrangements. Some markets are free of government intervention while
others are regulated. Within rms, some transactions are guided by market prices,
some are negotiated, and yet others are dictated by management. Mechanism design
theory provides a coherent framework for analyzing this great variety of institutions, or
allocation mechanisms, with a focus on the problems associated with incentives and
private information.
Markets, or market-like institutions, often allocate goods and services eciently.
Long ago, economists theoretically proved this eciency under fairly stringent assump-
tions concerning, among other things, the nature of the goods to be produced and
traded, participants information about these, and the degree of competition. Mecha-
nism design theory allows researchers to systematically analyze and compare a broad
variety of institutions under less stringent assumptions. By using game theory, mech-
anism design can go beyond the classical approach, and, for example, explicitly model
how prices are set. In some cases, the game-theoretic approach has led to a new appre-
ciation of the market mechanism. The theory shows, for example, that so-called double
auctions (where buyers and sellers post their bid- and ask-prices) can be ecient trad-
ing institutions when each trader has private information about his or her valuations of
the goods traded. As the number of traders increases, the double-auction mechanism
will more and more eciently aggregate privately held information, and eventually all
information is reected by the equilibrium prices (Wilson, 1985). These results support
Friedrich Hayeks (1945) argument that markets eciently aggregate relevant private
information.
Mechanism design theory shows which mechanisms are optimal for dierent partic-
ipants, say sellers or buyers (e.g. Samuelson, 1984). Such insights have been used to
better understand market mechanisms that we frequently observe. For example, the
1
theory has been used to identify conditions under which commonly observed auction
forms maximize the sellers expected revenue (Harris and Raviv, 1981; Myerson, 1981;
Riley and Samuelson, 1981). The theory also admits detailed characterizations of opti-
mal auction forms when these conditions do not hold (Myerson, 1981; Maskin and Riley,
1984a). Likewise, mechanism design theory has enabled economists to nd solutions
to the monopoly pricing problem, showing, for example, how the price should depend
on quality and quantity so as to maximize the sellers expected revenue (Maskin and
Riley, 1984b). Again, the theoretical solution squares well with observed practice.
In some cases, no market mechanism can ensure a fully ecient allocation of re-
sources. In such cases, mechanism design theory can be used to identify other, more
ecient institutions. A classic example concerns public goods, such as clean air or na-
tional security. Paul Samuelson (1954) conjectured that no resource allocation mecha-
nism can ensure a fully ecient level of public goods, because it is in the selsh interest
of each person to give false signals, to pretend to have less interest in a given collective
activity than he really has... (page 388 op. cit.). Mechanism design theory permits
a precise analysis of Samuelsons conjecture. More generally, the theory can be used
to analyze the economic eciency of alternative institutions for the provision of public
goods, ranging from markets and consensual collective decision-making through majori-
tarian decision rules all the way to dictatorship. An important insight is that consensual
decision-making is frequently incompatible with economic eciency. The theory thus
helps to justify governmental nancing of public goods through taxation. Applications
of mechanism design theory have led to breakthroughs in a number of other areas of
economics as well, including regulation, corporate nance, and the theory of taxation.
The development of mechanism design theory began with the work of Leonid Hur-
wicz (1960). He dened a mechanism as a communication system in which participants
send messages to each other and/or to a message center, and where a pre-specied
rule assigns an outcome (such as an allocation of goods and services) for every collec-
tion of received messages. Within this framework, markets and market-like institutions
could be compared with a vast array of alternative institutions. Initially, much of the
interest focussed on the informational and computational costs of mechanisms, while
abstracting from the problem of incentives. An important contribution was Marshak
and Radners (1972) theory of teams, which inspired much subsequent literature (e.g.
Groves, 1973). However, in many situations, providing incentives to the participating
agents is an important part of the problem. Mechanism design theory became relevant
for a wide variety of applications only after Hurwicz (1972) introduced the key notion
2
of incentive-compatibility, which allows the analysis to incorporate the incentives of
self-interested participants. In particular, it enables a rigorous analysis of economies
where agents are self-interested and have relevant private information.
In the 1970s, the formulation of the so-called revelation principle and the devel-
opment of implementation theory led to great advances in the theory of mechanism
design. The revelation principle is an insight that greatly simplies the analysis of
mechanism design problems. In force of this principle, the researcher, when searching
for the best possible mechanism to solve a given allocation problem, can restrict at-
tention to a small subclass of mechanisms, so-called direct mechanisms. While direct
mechanisms are not intended as descriptions of real-world institutions, their mathemat-
ical structure makes them relatively easy to analyze. Optimization over the set of all
direct mechanisms for a given allocation problem is a well-dened mathematical task,
and once an optimal direct mechanism has been found, the researcher can translate
back that mechanism to a more realistic mechanism. By this seemingly roundabout
method, researchers have been able to solve problems of institutional design that would
otherwise have been eectively intractable. The rst version of the revelation principle
was formulated by Gibbard (1973). Several researchers independently extended it to
the general notion of Bayesian Nash equilibrium (Dasgupta, Hammond and Maskin,
1979, Harris and Townsend, 1981, Holmstrom, 1977, Myerson, 1979, Rosenthal, 1978).
Roger Myerson (1979, 1982, 1986) developed the principle in its greatest generality and
pioneered its application to important areas such as regulation (Baron and Myerson,
1982) and auction theory (Myerson, 1981).
The revelation principle is extremely useful. However, it does not address the issue
of multiple equilibria. That is, although an optimal outcome may be achieved in one
equilibrium, other, sub-optimal, equilibria may also exist. There is, then, the danger
that the participants might end up playing such a sub-optimal equilibrium. Can a
mechanism be designed so that all its equilibria are optimal? The rst general solution
to this problem was given by Eric Maskin (1977). The resulting theory, known as
implementation theory, is a key part of modern mechanism design.
The remainder of this survey is organized as follows. Section 2 presents key concepts
and results, Section 3 discusses applications, and Section 4 concludes.
3
2 Key concepts and insights
We begin by describing incentive compatibility and the revelation principle. We then
discuss some results obtained for two main solution concepts, dominant-strategy equi-
librium and Bayesian Nash equilibrium, respectively. We consider, in particular, the
classic allocation problem of optimal provision of public goods. We also discuss a simple
example of bilateral trade. We conclude by discussing the implementation problem.
2.1 Incentive compatibility and the revelation principle
The seminal work of Leonid Hurwicz (1960,1972) marks the birth of mechanism design
theory. In Hurwiczs formulation, a mechanism is a communication system in which
participants exchange messages with each other, messages that jointly determine the
outcome. These messages may contain private information, such as an individuals (true
or pretended) willingness to pay for a public good. The mechanism is like a machine
that compiles and processes the received messages, thereby aggregating (true or false)
private information provided by many agents. Each agent strives to maximize his or
her expected payo (utility or prot), and may decide to withhold disadvantageous
information or send false information (hoping to pay less for a public good, say). This
leads to the notion of implementing outcomes as equilibria of message games, where
the mechanism denes the rules of the message game. The comparison of alternative
mechanisms is then cast as a comparison of the equilibria of the associated message
games.
To identify an optimal mechanism, for a given goal function (such as prot to a
given seller or social welfare), the researcher must rst delineate the set of feasible
mechanisms, and then specify the equilibrium criterion that will be used to predict the
participants behavior. Suppose we focus on the set of direct mechanisms, where
the agents report their private information (for example, their willingness to pay for
a public good). There is no presumption that the agents will tell the truth; they will
be truthful only if it is in their self-interest. Based on all these individual reports, the
direct mechanism assigns an outcome (for example, the amount provided of the public
good and fees for its nancing). Suppose we use the notion of dominant strategy equi-
librium as our behavioral criterion.
1
Hurwiczs (1972) notion of incentive-compatibility
can now be expressed as follows: the mechanism is incentive-compatible if it is a dom-
1
A strategy is dominant if it is a agents optimal choice, irrespective of what other agents do.
4
inant strategy for each participant to report his private information truthfully. In
addition, we may want to impose a participation constraint: no agent should be made
worse o by participating in the mechanism. Under some weak assumptions on tech-
nology and taste, Hurwicz (1972) proved the following negative result: in a standard
exchange economy, no incentive-compatible mechanism which satises the participation
constraint can produce Pareto-optimal outcomes. In other words, private information
precludes full eciency.
A natural question emanating from Hurwiczs (1972) classic work thus is: Can
Pareto optimality be attained if we consider a wider class of mechanisms and/or a
less demanding equilibrium concept than dominant-strategy equilibrium, such as Nash
equilibrium or Bayesian Nash equilibrium?
2
If not, then we would like to know how
large the unavoidable social welfare losses are, and what the appropriate standard of
eciency should be. More generally, we would like to know what kind of mechanism will
maximize a given goal function, such as prot or social welfare (whether this outcome
is fully ecient or not). In the literature that followed Hurwicz (1972), these questions
have been answered. Much of the success of this research program can be attributed to
the discovery of the revelation principle.
The revelation principle states that any equilibrium outcome of an arbitrary mech-
anism can be replicated by an incentive-compatible direct mechanism. In its most
general version, developed by Myerson (1979, 1982, 1986), the revelation principle is
valid not only when agents have private information but also when they take unob-
served actions (so-called moral hazard), as well as when mechanisms have multiple
stages. Although the set of all possible mechanisms is huge, the revelation principle
implies that an optimal mechanism can always be found within the well-structured sub-
class consisting of direct mechanisms. Accordingly, much of the literature has focussed
on the well-dened mathematical task of nding a direct mechanism that maximizes
the goal function, subject to the incentive-compatibility (IC) constraint (and, where
appropriate, also the participation constraint).
A rough proof of the revelation principle for the case with no moral hazard goes
as follows. First, x an equilibrium of any given mechanism. An agents private infor-
mation is said to be his type. Suppose that an agent of type t sends the message
m(t) in this equilibrium. Now consider the associated direct mechanism in which each
2
In a Nash equilibrium, each agents strategy is a best response to the other agents strategies. A
Bayesian Nash equilibrium is a Nash equilibrium of a game of incomplete information, as dened by
Harsanyi (1967-8).
5
agent simply reports a type t
0
, where t
0
may be his true type t or any other type. The
reported type t
0
is his message in the direct mechanism, and the outcome is dened
to be the same as when the agent sends the message m(t
0
) in the equilibrium of the
original mechanism. By hypothesis, an agent of type t preferred to send message m(t)
in the original mechanism (the agent could not gain by unilaterally deviating to another
message). In particular, the agent preferred sending the message m(t) to sending the
message m(t
0
), for any for t
0
6= t. Therefore, he also prefers reporting his true type t
in the direct mechanism, rather than falsely reporting any other type t
0
. So the direct
mechanism is incentive compatible: no agent has an incentive to misreport his type. By
construction, the direct mechanism produces the same outcome as the original mecha-
nism. Thus, any (arbitrary) equilibrium can be replicated by an incentive-compatible
direct mechanism.
3
As discussed below, the revelation principle can be used to generalize Hurwiczs
(1972) impossibility result to the case of Bayesian Nash equilibrium. Thus, in settings
where participants have private information, Pareto optimality in the classical sense is
in general not attainable, and we need a new standard of eciency which takes incen-
tives into account. A direct mechanism is said to be incentive ecient if it maximizes
some weighted sum of the agents expected payos subject to their IC constraints.
Armed with this denition, researchers have been able to answer many of the questions
that emanated from Hurwiczs (1972) work. One of the key questions is whether market
mechanisms can be incentive ecient. In partial equilibrium settings, Myerson and Sat-
terthwaite (1983) and Wilson (1985) proved that so-called double auctions are incentive
ecient. Prescott and Townsend (1984) characterized the information structures under
which a competitive general equilibrium is incentive ecient.
We now discuss some results pertaining to economies with public goods, both for
dominant-strategy equilibrium and for Bayesian Nash equilibrium.
3
More formally, suppose there are n agents. The original (indirect) mechanism assigns an outcome
x, say an allocation of private and/or public goods, to all message proles (m
1
, ..., m
n
)
n
i=1
M
i
,
x = h(m
1
, ..., m
n
). A pure strategy for a agent i is a rule (function) s
i
that species for each possible
type t
i
T
i
a message m
i
M
i
. Thus m
i
= s
i
(t
i
) for all agents i and types t
i
. Suppose now that
a strategy prole s
is an equilibrium of the original mechanism. Then a direct mechanism can be

dened in which each agent i announces a type t
0
i
T
i
and the outcome is given by x = h
(t
0
1
, ..., t
0
n
),
where h
is dened by h
(t
1
, ..., t
n
) = h(s
1
(t
1
) , ..., s
n
(t
n
)) for all type reports (t
1
, ..., t
n
). No agent
can gain by reporting his or her type falsely, for if this were possible, then it would also have been
possible for that agent to improve his or her payo in the original mechanism by way of a corresponding
unilateral change of strategy.
6
2.2 Dominant-strategy mechanisms for public goods provision
As mentioned above, a classic problem concerns the optimal provision of public goods.
When individuals have private information about their own willingness to pay for the
public good, they may be tempted to pretend to be relatively uninterested, so as to
reduce their own share of the provision cost. This problem is canonical and arises in
virtually all societies: how should a group of farmers, say, share the cost of a common
irrigation or drainage system; how should the countries in the world share the cost of
reducing global warming; how should grown-up siblings share the burden of caring for
their elderly parents?
Before 1970, economists generally believed that public goods could not be provided
at an ecient level, precisely because people would not reveal their true willingness to
pay. It thus came as a surprise when Edward Clarke (1971) and Theodore Groves (1973)
showed that, if there are no income eects on the demand for public goods (technically,
if utility functions are quasi-linear), then there exists a class of mechanisms in which
(a) truthful revelation of ones willingness to pay is a dominant strategy, and (b) the
equilibrium level of the public good maximizes the social surplus.
4
In the context of a
binary decision (whether or not to build a bridge, for example), the simplest version of
the Clarke-Groves mechanism works as follows. Each person is asked to report his or
her willingness to pay for the project, and the project is undertaken if and only if the
aggregate reported willingness to pay exceeds the cost of the project. If the project is
undertaken, then each person pays a tax or fee equal to the dierence between the cost of
the project and everyone elses reported total willingness to pay. With such taxes, each
person internalizes the total social surplus, and truth-telling is a dominant strategy.
The main drawback of this mechanism is that the total tax revenue typically will not
add up to the cost of the project: the mechanism does not in general satisfy budget
balance (see Green and Laont, 1979). Both too much funding and too little funding
is problematic. For example, sharing surplus funds among the participants will destroy
the participants truth-telling incentives, while wasting surplus funds is inecient.
Outside the quasi-linear economic environments studied by Clarke and Groves, not
much can be achieved by way of dominant-strategy mechanisms. A result to this eect
was given by Gibbard (1973) and Satterthwaite (1975). They showed that in quite
general environments, the only dominant-strategy mechanism is dictatorship, whereby
4
The basic intuition behind the Clarke-Groves mechanism was already present in Vickrey (1961),
so this kind of mechanism is often referred to as the Vickrey-Clarke-Groves (VCG) mechanism. See
Tideman and Tullock (1976) for a discussion of this type of mechanism.
7
one pre-selected agent, the dictator, always gets his favorite alternative. Because
of this and other negative results, the focus of the literature shifted from dominant-
strategy solutions to so-called Bayesian mechanism design.
2.3 Bayesian mechanisms for public goods provision
In a Bayesian model, the agents are expected-utility maximizers. The solution con-
cept is typically Bayesian Nash equilibrium. The general Bayesian mechanism design
problem was formulated by Dasgupta, Hammond and Maskin (1979), Myerson (1979)
and Harris and Townsend (1981). After the discovery of the revelation principle (see
Section 2.1), the main development of the theory of Bayesian mechanism design came in
a series of papers by Roger Myerson (Myerson, 1979, 1981, 1983, Baron and Myerson,
1982, and Myerson and Satterthwaite, 1983). In these papers, the set of possible allo-
cations was unidimensional, and the agents had quasi-linear preferences that satised a
single-crossing property, familiar from the work of James Mirrlees and Michael Spence.
Myerson obtained elegant characterizations of the incentive constraints which admitted
a particularly insightful analysis. The same machinery has subsequently been used in
a large number of applications.
As mentioned in Section 2.2, the Clarke-Groves dominant-strategy mechanism for
the provision of public goods violates budget-balance. Claude dAspremont and Louis-
Andr Grard-Varet (1979) showed that this problem can be solved in the Bayesian
version of the model.
5
In a dominant-strategy mechanism, the IC constraints require
that each agents utility is maximized by reporting the truth, regardless of what the
other agents might report. In the Bayesian model, agents are expected utility maxi-
mizers, and the IC constraints only have to hold in expectation. Accordingly, the IC
constraints are easier to satisfy in the Bayesian model, and dAspremont and Grard-
Varet could obtain more positive results than is possible with dominant strategies. In
fact, dAspremonts and Grard-Varets (1979) mechanism can be seen as an extension
of the Clarke-Groves mechanism to the Bayesian context.
The dAspremont and Grard-Varet (1979) mechanism produces outcomes which are
fully Pareto ecient, but their mechanism violates (interim) participation constraints.
Some individuals, having observed their own type but not yet taken their actions,
would prefer not to participate, so this mechanism is feasible only if participation is
mandatory. If participation is voluntary and decisions to start the project must be
5
Arrow (1979) independently constructed a similar mechanism.
8
taken unanimously, then the problem of free-riding becomes severe. Using techniques
developed by Myerson (1981), Mailath and Postlewaite (1990) show that the probability
of funding a public-goods project tends to zero as the number of agents increases. They
give an example where the asymptotic probability of funding the public project is zero
despite everyone knowing that they can be jointly better o if the project is funded.
6
These results provide a rigorous foundation for Samuelsons (1954) negative conjec-
ture about public goods cited above (Section 1). They give a plausible explanation for
observed failures to provide public goods. For example, the fact that English villages
were much earlier than French villages in deciding on public goods such as enclosure of
open elds and drainage of marshlands can arguably be ascribed to the fact that French
villages required unanimity on such issues whereas the English did not. This may at
least partially explain why the productivity growth in English agriculture outstripped
that of French agriculture in the period 1600-1800 (Grantham, 1980; Rosenthal, 1992).
In a large class of models, classical Pareto eciency is incompatible with voluntary
participation, even if there are no public goods.
7
In these models, the classical notion of
Pareto eciency is usually replaced by the more relevant notion of incentive eciency
(see Section 2.1). Two fundamental impossibility results to this eectshowing the
incompatibility of voluntary participation and classical Pareto eciencywere proved
by Laont and Maskin (1979, Section 6) and Myerson and Satterthwaite (1983). In or-
der to illustrate these results and to convey the avor of the formal analysis of Bayesian
mechanism design, let us consider in some detail the case of bilateral trade in private
goods.
2.4 Example: bilateral trade
Suppose one individual, A, owns an indivisible object. A is considering selling this
object to a prospective buyer, B. The object is worth w to A and v to B. Normalize
both valuations v and w to lie between zero and one. If the object is sold at a price p,
6
Similar results were obtained by Roberts (1976) and Rob (1989).
7
The key dierence between private and public goods is that in the former case, there are plausible
assumptions under which problems of incentives and private information vanish as the economy be-
comes large. For example, Wilson (1977) and Milgrom (1979) found that if private goods are allocated
via auctions, and there is a large numbers of potential buyers, then the equilibrium outcomes satisfy
key properties of classic competitive equilibria (i.e. the price aggregates all privately held information
and reects the "true value" of the good, and the agents nd it optimal to treat prices parametrically).
In sharp contrast, if public goods are present, then incentive problems often become more severe as
the economy gets larger (Mailath and Postlewaite, 1990).
9
then As utility is p w; she has to give up the object worth w to her but receives p in
return. Similarly, Bs utility from such a transaction is v p. If no trade occurs, then
each party obtains utility zero. Suppose that this is a truly bilateral situation; none of
the parties can trade with a third party.
In such a situation, the classical notion of Pareto eciency requires that the object
be sold if w < v and not if w > v. That is, all gains of trade should be realized.
Geometrically, this means that trade occurs if and only if the valuation pair (w, v) lies
above the diagonal of the unit square in the diagram below. Now suppose that B does
not know As valuation, so w is As private information. Similarly, suppose that v is
Bs private information. To be more precise, suppose that the two individuals have
been randomly drawn from a population of individuals with dierent valuations, in
such a way that their types, w and v, are statistically independent and identically
distributed random variables with positive density on the whole unit square. What
kind of mechanism could they use to trade with each other?
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
w
v
One possibility is that A makes a take-it-or-leave-it oer to B. Another possibility
is that B makes such an oer to A. A third possibility would be a double auction, a
mechanism in which both parties (simultaneously) announce a price and, if Bs an-
nouncement exceeds As, they trade at a price between the two announcement (for
example, at the mid-point between the two announcements). It turns out that none of
these mechanisms has the property that trade occurs in equilibrium whenever w v.
For example, if A makes a take-it-or-leave it oer p, then she will surely propose p > w,
10
and B will accept only if v p. So, trade does not occur if w < v < p, an event with
positive probability. The argument is symmetric when instead B makes an oer. The
mechanism would realize all gains of trade if the agents priced at their own valuations.
However, this is not incentive-compatible. Agent A will benet from pricing above
her valuation (in order to obtain a higher selling price), and agent B will benet from
pricing below his valuation (in order to obtain the object at a lower price). Thus, each
party will to try to improve his or her terms of trade by not pricing at their own true
valuations. However, by doing this they will not realize all gains of trade.
If the population value-distribution is uniform, then the double auction has a linear
Bayes Nash equilibrium, that is, one in which each partys (ask- and bid-) price is
linearly increasing in the partys true valuation. More specically, in this equilibrium,
As ask price is p
A
= 2w/3 +1/4, unless her valuation w exceeds 3/4, in which case she
asks her true valuation, p
A
= w. Likewise, B bids the price p
B
= 2v/3 + 1/12, unless
his valuation v falls short of 1/4, in which case he bids his true valuation, p
B
= v.
Notice that if w < 3/4, then p
A
> w, that is, A asks for more than her own valuation.
Similarly, if v > 1/4 then B bids below his valuation. Consequently, if v and w are too
close to each other, no trade occurs even if v > w. Indeed, trade occurs if and only if
Bs valuation, v, exceeds As valuation, w, by at least 1/4. This is the triangular area
above the higher (and thinner) 45
o
-line in the diagram. By contrast, no trade occurs in
the central band between the two straight lines, that is, where the valuations are too
close. Hence, for valuation-pairs falling in this area, no gains of trade are realized.
This situation is quite general. The impossibility results established by Laont and
Maskin (1979, Section 6) and Myerson and Satterthwaite (1983) imply that for bilateral
trade no incentive compatible direct mechanism which satises (interim) participation
constraints can have the property that trade occurs if and only if w v. By the
revelation principle, we can infer that no mechanism whatsoever can realize all gains
from trade.
8
Classical Pareto eciency, in other words, is incompatible with voluntary
participation and free trade in this example. Indeed, although the above equilibrium
outcome apparently violates classical Pareto optimality, further extraction of potential
gains from trade cannot be achieved by any incentive compatible mechanism; it can
be shown that the double auction is incentive ecient and that the linear equilibrium
achieves this upper eciency bound.
8
Note that the double auction is mathematically equivalent to the direct mechanism in which each
party announces its valuation and the object changes hands if and only if the owners valueation is
lower than the propspecitve buyers, at a price between the announced valuations.
11
More exactly, using the revelation principle, Myerson and Satterthwaite (1983) es-
tablished an upper bound for the gains from trade, v w, that are realizable in any
trade mechanism in situations like this. Their approach can be explained as follows.
Consider a Bayesian Nash equilibrium of an arbitrary mechanism for bilateral trade.
Suppose that when As type is w and Bs type is v, the object is sold with probability
q(w, v) at price p(w, v), so As payo is q(w, v) [p(w, v) w], where w and v are statisti-
cally independent random variables. Since A knows her own type but not Bs type, she
calculates that her expected payo is E
v
[q(w, v) [p(w, v) w]], where the expectation
is with respect to Bs type v. Now it must be the case that
E
v
[q(w, v) [p(w, v) w]] E
v
[q(w
0
, v) [p(w
0
, v) w]]
for all w and w
0
6= w. To see this, notice that the left-hand side is what As type w
obtains in equilibrium, while the right-hand side is what her type w would get if she
behaved just like type w
0
(in which case the object would be sold with probability
q(w
0
, v) at the price p(w
0
, v)). The denition of Bayesian Nash equilibrium requires
that the inequality holds, that is, type w should not be able to improve her payo by
mimicking type w
0
. By an analogous reasoning applied to B instead of A, we must have
E
w
[q(w, v) [v p(w, v)]] E
w
[q(w, v
0
) [v p(w, v
0
)]]
for all v and v
0
6= v. The two inequalities we have just derived are nothing but the IC
constraints for the direct mechanism in which A announces w and B announces v, and
where trade occurs with probability q(w, v) at price p(w, v). Thus, we may as well focus
on such direct mechanism this is the revelation principle. Moreover, if a traders
expected payo is negative, he or she would refuse to participate. Therefore, the (ex
ante) participation constraints are
E
v
[q(w, v) [p(w, v) w]] 0 and E
w
[q(w, v) [v p(w, v)]] 0.
Now, to establish an upper bound for the gains from trade that can be achieved
in any mechanism, we need only consider the well-dened mathematical problem of
maximizing the expected gains from trade, E(v w), subject to the above IC and
participation constraints. A mechanism which achieves this upper bound (in some
equilibrium) is incentive ecient. The upper-bound result in Myerson and Satterth-
waite (1983) implies that the double auction, rst studied in detail by Chatterjee and
12
Samuelson (1983), is incentive ecient.
2.5 Implementation theory
Incentive compatibility guarantees that truth-telling is an equilibrium, but not that it is
the only equilibrium. Many mechanisms have multiple equilibria that produce dierent
outcomes. For instance, Leininger, Linhart, and Radner (1989) found that the double
auction (see Section 2.4) has innitely many (indeed uncountably many) non-linear
equilibria, the welfare of which ranges from incentive eciency to zero. Clearly, this
multiplicity of equilibria reduces the appeal of the double auction.
Wilson (1979) analyzed uniform-price auctions for divisible goods and uncovered
equilibria where the bidders divide up the good in question at a very low price. In
these collusive equilibria, each bidder bids aggressively for anything less than his
anticipated equilibrium share, which deters other bidders from trying to acquire more
than their (implicitly agreed upon) shares. Such implicit collusion is highly detrimental
to the seller. According to Klemperer (2004, Chapter 3), precisely this kind of im-
plicit collusion has plagued many real-world auctions, including the U.K. market for
electricity.
Multiple-equilibrium problems are also endemic in social-choice theory. Voters who
are to select one out of many candidates face, in eect, a coordination problem. To
vote for a candidate who has little chance of winning means wasting ones vote.
Accordingly, if there is a commonly held belief in the electorate that a certain candidate
has no chance of winning, then this expectation can be self-fullling. Such phenomena
easily generate multiple equilibria, some of which lead to suboptimal outcomes (see
Section 3.3 for further discussion of voting mechanisms).
In view of these diculties, it is desirable to design mechanisms in which all equi-
librium outcomes are optimal for the given goal function. The quest for this property is
known as the implementation problem.
9
Groves and Ledyard (1977) and Hurwicz and
Schmeidler (1978) showed that, in certain situations, it is possible to construct mecha-
nisms in which all Nash equilibria are Pareto optimal, while Eric Maskin (1977) gave
a general characterization of Nash implementable social-choice functions. He showed
that Nash implementation requires a condition now known as Maskin monotonicity
(see Section 3.3 for an illustration of this property). Maskin (1977) also showed that
9
Formally, weak implementation requires that every equilibrium is optimal, while full imple-
mentation in addition requires that every optimum be an equilibrium.
13
if Maskin monotonicity and a condition called no-veto-power are both satised, and if
there are at least three agents, then implementation in Nash equilibrium is possible.
10
Maskin considered Nash equilibria in games of complete information, but his results
have been generalized to Bayesian Nash equilibria in games of incomplete information
(see Postlewaite and Schmeidler, 1986, Palfrey and Srivastava, 1989, Mookherjee and
Reichelstein, 1990, and Jackson, 1991). For example, Palfrey and Srivastava (1991)
show how the double auction can be modied so as to render all equilibria incentive
ecient.
Maskins results have also been extended in many other directions, such as virtual
(or approximate) implementation (Matsushima, 1988, Abreu and Sen, 1991), imple-
mentation in renegotiation-proof equilibria (Maskin and Moore, 1999) and by way of
sequential mechanisms (Moore and Repullo, 1988). Implementation theory has played,
and continues to play, an important role in several areas of economic theory, such as
social choice theory (Moulin, 1994) and the theory of incomplete contracts (Maskin and
Tirole, 1999).
3 Applications
In many cases, mechanism design has modernized and unied existing lines of research.
For example, while the revenue equivalence of well-known auction formats was known
already to Vickrey (1961), the mechanism-design approach entailed a more general
revenue-equivalence theorem. By contrast, the optimality of the most common auction
formats (within the class of all possible selling mechanisms) could only be established by
mechanism-design techniques. In still other cases, mechanism-design theorists have de-
veloped entirely new research avenues. We are in no position to discuss all applications
of mechanism design but will try to give a avor of a few important ones.
11
3.1 Optimal selling and procurement mechanisms
Auctions and auction-like mechanisms are an important part of modern economic life.
Myersons (1981) seminal analysis of optimal auctions and the large subsequent litera-
10
In Maskins original manuscript, the proof of this result was incomplete. Complete proofs of
Maskins theorem were later provided by Williams (1986), Repullo (1987) and Saijo (1988).
11
An interesting research area not discussed here is the analysis of competing institutions, which
studies the equilibrium allocation of buyers across competing trading mechanisms and price formation
in these (see McAfee, 1993, Peters, 1997, and Ellison, Fudenberg and Mbius, 2004).
14
ture (see Krishna, 2002) have helped economists understand these important real-world
institutions.
12
In a typical scenario, an economic agent has an object to sell, but does
not know how much the prospective buyers (bidders) are willing to pay for it. Which
mechanism will be optimal in the sense of maximizing the sellers expected revenue?
This problem was analyzed by Myerson (1981). Appealing to the revelation princi-
ple, Myerson studied incentive-compatible direct mechanisms, where the bidders report
their willingness-to-pay. The mechanism species who will get the object and at what
price, as a function of these reports. Incentive-compatibility guarantees that truth-
telling is a Bayesian Nash equilibrium. Since participation is voluntary, the equilibrium
must also satisfy an (interim) participation constraint: each bidder who participates
in the auction must be at least as well o as if he or she had abstained. For this
scenario, Myerson proved a general revenue-equivalence theorem. This theorem estab-
lishes conditions, such as risk neutrality and uncorrelated types, under which the seller
achieves the same expected revenue from any auction in which the object goes to the
bidder with the highest valuation (in equilibrium). In particular, four well-known auc-
tion forms (the so-called English and Dutch auctions, and rst-price and second-price
sealed bid auctions, respectively) generate the same expected revenue. Myerson (1981)
showed that if the bidders are symmetric (drawn from one and the same type pool)
and if the seller sets an appropriate reserve price (a lowest price below which the object
will not be sold), then all of the four well-known auction formats are in fact optimal.
13
For example, if the bidders types are independently drawn from a uniform distribution
on the interval from zero to one hundred, then the optimal reserve price is 50, indepen-
dently of the number of bidders. This reserve price induces bidders whose valuations
exceed 50 to bid higher than they would otherwise have done, which raises the expected
revenue. On the other hand, if it so happens that no bidder thinks the object is worth
50, then the object is not sold even if it has a positive value to some bidder and no
value at all to the seller. This outcome is clearly not Pareto ecient in the classical
sense. Nevertheless, the above-mentioned auction forms are incentive ecient in the
sense dened above.
Myerson (1981) assumed the sellers objective was to maximize the expected revenue.
But when the government is privatizing an asset, such as the radio spectrum or a
12
Harris and Raviv (1981) and Riley and Samuelson (1981) independently analysed the problem
of optimal auctions and reached some of the same conclusions, but it was Myersons methods that
provided the foundation for much future work.
13
If the bidders are asymmetric, then the optimal auction format enhances competition by dis-
criminating in favor of the weaker bidders (those drawn from pools with lower willingness-to-pay).
15
publicly-owned production facility, revenue maximization may not be the only (or even
the most important) motive. A bigger concern may be social-welfare maximization: the
asset should go to the individual or rm that values it the most.
Maskin (1992) found that, under certain conditions, an English auction maximizes
social welfare even if each bidders valuation depends on other bidders private infor-
mation. One might be tempted to discount the need for the governments auction to
maximize social welfare, for the following reason. Suppose there are two potential bid-
ders, A and B, and B values the asset more than A. Then, even if the government
allocates the asset to the wrong person, A, would not then B simply buy the asset
from A (assuming it can be traded)? If so, then B (who values the asset the most)
would always get the asset in the end - so the government should not worry too much
about getting the initial allocation right. However, this argument is incorrect, be-
cause it does not take informational constraints into account. The Laont-Maskin and
Myerson-Satterthwaite impossibility results (see Section 2.4) imply that B may not buy
the asset from A even if B values it the most. Therefore, getting the initial allocation of
ownership right may be of the utmost importance. Indeed, there is no presumption that
the initial allocation should be to one individual, since joint ownership may generate
higher social welfare (see Cramton, Gibbons and Klemperer, 1987).
If the seller can produce additional objects at constant cost, it is not important
for buyers to compete with each other directly. Mussa and Rosen (1978) and Maskin
and Riley (1984b) derived the optimal selling mechanism for a monopolist who does
not know its customers types (i.e. their taste parameters). The optimal mechanism
involves quantity discounts (rather than a xed price per unit). Stole (1995) extends
the theory to the case of oligopoly. Other important extensions concern multi-product
monopolies and multi-dimensional types (see Armstrong, 1996). Turning from revenue
to social-welfare maximization, Maskins (1992) eciency result does not generalize to
the case of multi-dimensional types. For this case, Jehiel and Moldovanu (2001) found
that there is in general no incentive-compatible mechanism which always allocates the
object to the person who values it most.
3.2 Regulation and auditing
The regulation of monopolies and oligopolies is an old and important topic in economics.
As discussed by Laont (1994), the older literature made rather arbitrary assumptions
about the regulatory process. The regulator was assumed to face certain constraints,
16
such as a requirement that the monopolist must earn a return above the market rate.
This rate was not derived from an underlying optimization process but was simply
imposed ad hoc. Within such loosely grounded frameworks, it is hard to make sound
normative judgements about regulatory processes. The situation changed dramatically
with the pioneering contributions of Baron and Myerson (1982) and Sappington (1982,
1983), building on work by Weitzman (1978) and Loeb and Magat (1979). In these
papers, the regulatory process was modelled as a game of incomplete information. The
regulator did not have direct access to information about the monopolists true produc-
tion costs. Using the revelation principle, Baron and Myerson (1982) and Sappington
(1982,1983) derived the optimal regulatory scheme, without resorting to ad hoc assump-
tions. In the optimal mechanism, the regulator (usually a government agency) trades o
its objective to extract rents from the monopolist (revenue to the government) against
its objective to encourage an ecient output level. In addition, the monopolist must
be given sucient incentive to participate (i.e. to stay in the market).
A surge in the literature on regulatory economics followed the Baron-Myerson and
Sappington contributions. This literature has provided a solid theoretical foundation
for evaluations of alternative regulatory mechanisms, such as price caps versus cost- and
prot-sharing schemes. Economists have used the Baron-Myerson model to empirically
estimate the eect of regulation on rms behavior (see Wolak, 1994). The original
static model was extended in many directions. The problem of optimal time-consistent
mechanisms, in particular the ratchet eect, when information is gradually revealed
over time, was analyzed by Freixas, Guesnerie, and Tirole (1985) and Laont and Tirole
(1988), among others. Papers by Laont and Tirole (1987), McAfee and McMillan
(1986) and Riordan and Sappington (1987) have produced a synthesis of theories of
optimal auctions and of optimal regulation. Baron and Besanko (1984) and Laont
and Tirole (1986) introduce the possibility of ex post audits of rms costs. Many other
topics, such as collusion between the regulated rm, its auditor and even the regulatory
agency, have been extensively analyzed in the literature on optimal regulation. For a
comprehensive and unied treatment, see Laont and Tirole (1993).
3.3 Social Choice Theory
In the axiomatic social-choice theory pioneered by Kenneth Arrow (1951), there is a
set X of feasible alternatives and n individuals who have preferences over these. A
social choice rule is a rule that selects one or several alternatives from X on the basis of
17
the individuals preferences, for any given such preference prole. Arrows seminal work
was mainly concerned with the normative issue of how a social choice rule can represent
the general will of the people. In the 1970s, attention shifted to the positive question
of strategic behavior of voters under alternative voting procedures. Is it possible to
design a mechanism, that is, a voting procedure, such that voters are induced to reveal
their true preferences over the set X? The impossibility results of Gibbard (1973)
and Satterthwaite (1975) gave a negative answer. They showed that if X contains at
least three alternatives, then there does not exist any non-dictatorial social choice rule
that can be implemented in a mechanism in which revealing ones true preferences is
a dominant strategy. The proof of the Gibbard-Satterthwaite theorem can be directly
translated into a proof of Arrows (1951) celebrated impossibility theorem for normative
social choice (see Muller and Satterthwaite, 1985). This conrms Arrows conjecture
that his axiom of Independence of Irrelevant Alternatives is closely related to the notion
of dominant-strategy mechanisms. Thus, the Gibbard-Satterthwaite theorem provides
a bridge between normative and strategic analyses. The next step was to relax the
requirement of dominant strategies. The resulting literature was greatly inuenced by
Maskins (1977) work on Nash implementation. For a survey of the strategic aspects of
social choice theory, see Moulin (1994).
An early insight of this literature was that if social choice rules are required to
be singleton-valued, that is, if a unique alternative must always be selected, then the
Gibbard-Satterthwaite impossibility result also holds for Nash implementation. To un-
derstand this negative result, recall that a necessary condition for Nash implementation
of a social choice rule is a condition called Maskin monotonicity. This condition says
that, if initially an alternative a X is selected by the social choice rule and a does
not fall in rank in any voters preference ordering, then a must still be selected.
To illustrate the strength of this condition, consider a specic social choice rule,
namely, the plurality rule. An alternative in X is said to be the plurality alternative if
it is top-ranked by the greatest number of voters. The plurality rule simply states that
the plurality alternative should always be selected. Now suppose there are 7 voters, and
X contains three alternatives, a, b and c. Suppose that the voters preference orderings
are as follows. Voters 1, 2 and 3 think a is the best alternative, b the second best, and
c is worst: a > b > c. Voters 4 and 5 think b is the best alternative, a the second
best, and c is worst: b > a > c. Voters 6 and 7, nally, think c is the best alternative,
b the second best, and a is worst: c > b > a. Clearly a is the plurality alternative,
since it is top-ranked by three voters, while b and c are each top-ranked by only two
18
voters. Now suppose voters 6 and 7 change their minds: they decide that alternative
c is, after all, worse than a and b. In their new preference orderings, b has risen to
the rst place and a to the second: b > a > c. If the other voters rankings remain
as before, then alternative a did not fall in anyones preference ordering, but it is no
longer the plurality alternative since b is now top-ranked by four voters while a is still
top-ranked by only three. Hence, the plurality rule is not Maskin monotonic.
By Maskins (1977) theorem, there is no decision mechanism that Nash-implements
the plurality rule.
14
In a similar fashion, other social choice rules that have been
proposed in the literature, such as the well-known Borda rule (proposed by J.C. de
Borda in 1781), also fail Maskin monotonicity. More generally, Muller and Satterthwaite
(1977) showed that no single-valued social choice rule can be Maskin monotonic. This
means that voters strategic behavior will lead any conceivable voting mechanism to
produce Nash equilibria that are suboptimal according to the given social choice rule.
One way out of this dilemma is to drop the requirement that the social choice
rule be single-valued. Many interesting multi-valued social choice rules (such as the
one that always selects all Pareto-ecient alternatives) are Maskin monotonic and can
be Nash-implemented. The drawback is that we are forced to accept a fundamen-
tal indeterminacy: with some preference-proles, more than one alternative should be
acceptable to society. Therefore, more than one Nash equilibrium exists. This indeter-
minacy might be an unavoidable aspect of non-dictatorial systems. The nal outcome
can then depend on negotiations and bargaining among the voters. In the terminology
of Thomas Schelling, voters may coordinate on a focal point equilibrium, an equilib-
rium that appears natural given their cultural background, history, or other social and
psychological factors. An alternative path toward more positive results is to assume
the voters behavior can be captured by way of renement of Nash equilibrium, such
as trembling-hand perfect Nash equilibrium (Selten, 1975); see Maskin and Sjstrm
(2002).
4 Conclusion
Mechanism design theory denes institutions as non-cooperative games, and compares
dierent institutions in terms of the equilibrium outcomes of these games. It allows
14
More precisely, the plurality rule is not Maskin monotonic as long as X contains at least three
alternatives. By contrast, it is easily veried that if X has only two alternatives, then the plurality
rule is Maskin-monotonic.
19
economists and other social scientists to analyze the performance of institutions rela-
tive to the theoretical optimum. Mechanism design has produced a large number of
important insights in a wide range of applied contexts, inuencing economic policy as
well as market institutions. We have discussed here some of the most important results
and applications.
For introductory surveys of mechanism design theory, see Baliga and Maskin (2003)
and Serrano (2004). The revelation principle is discussed in chapter 2 of Salani (1997).
For more on Bayesian mechanism design, see chapter 7 in Fudenberg and Tirole (1993),
chapter 5 in Krishna (2002), chapter 23 in Mas-Colell, Whinston and Green (1995), and
Myerson (1989). For the implementation problem, see Corchn (1996), Jackson (2001),
Maskin and Sjstrm (2002), Moore (1992), chapter 10 in Osborne and Rubinstein
(1994), and Palfrey (2001).
References
[1] Abreu, D. and A. Sen (1991): Virtual implementation in Nash equilibria,
[2] Armstrong, M. (1996): Multiproduct nonlinear pricing, Econometrica 64, 51-
75.
[3] Arrow, K. (1951): Social Choice and Individual Values. Wiley, New York.
[4] Arrow, K.J. (1979): The property rights doctrine and demand revelation under
incomplete information, in M. Boskin (ed.), Economics and Human Welfare.
Academic Press, New York.
[5] Baliga, S. and E. Maskin (2003) Mechanism design for the environment, in
K. Mler and J. Vincent (eds.), Handbook of Environmental Economics. Elsevier
Science, Amsterdam.
[6] Baron, D. and D. Besanko (1984): Regulation, asymmetric information, and
auditing, Rand Journal of Economics 15, 447-470.
[7] Baron, D. and R. Myerson (1982): Regulating a monopolist with unknown
costs, Econometrica 50, 911-930.
20
[8] Bulow, J. and P. Klemperer (1996): Auctions versus negotiations, American
[9] Chatterjee, K. and W. Samuelson (1983): Bargaining under incomplete infor-
mation, Operations Research 31, 835-851.
[10] Clarke, E.H. (1971): Multipart pricing of public goods, Public Choice 11:1733.
[11] Corchn, L. (1996): The Theory of Implementation of Socially Optimal Decisions
in Economics. Palgrave Macmillan.
[12] Cramton, P., R. Gibbons and P. Klemperer (1987): Dissolving a. partnership
eciently, Econometrica 55: 615632
[13] Dasgupta, P., P. Hammond and E. Maskin (1979): The implementation of social
choice rules: some general results on incentive compatibility, Review of Economic
Studies 46, 181-216.
[14] dAspremont, C. and L.A. Grard-Varet (1979): Incentives and incomplete in-
formation, Journal of Public Economics 11, 25-45.
[15] Ellison, G., Fudenberg, D. and M. Mbius (2004): Competing auctions, Journal
of the European Economic Association 2: 30-66
[16] Freixas, X., R. Guesnerie, and J. Tirole (1985): Planning under incomplete
information and the ratchet eect, Review of Economic Studies 52, 173-192.
[17] Fudenberg, D. and J. Tirole (1993): Game Theory. MIT Press
[18] Gibbard, A. (1973): Manipulation of voting schemes: a general result, Econo-
metrica 41, 587-602.
[19] Green, J. and J.J. Laont (1979): Incentives in Public Decision Making. North-
Holland, Amsterdam.
[20] Groves, T. (1973): Incentives in teams, Econometrica 41, 617-663.
[21] Grantham, G. (1980): The persistence of open-eld farming in nineteenth-
century France, Journal of Economic History 40, 515-531.
[22] Groves, T. and J. Ledyard (1977): Optimal allocation of public goods: A solution
to the free rider dilemma, Econometrica 45, 783-811
21
[23] Guesnerie, R. (1995): A Contribution to the Pure Theory of Taxation. Cambridge
University Press.
[24] Harris, M. and A. Raviv (1981): Allocation mechanisms and the design of
auctions, Econometrica 49, 1477-1499.
[25] Harris, M. and R. Townsend (1981): Resource allocation under asymmetric
information, Econometrica 49, 33-64.
[26] Harsanyi, J. (1967-8): Games with incomplete information played by Bayesian
players, Management Science 14, 159-189, 320-334, 486-502.
[27] Hayek, F. (1945): The use of knowledge in society, American Economic Review
35, 519-530.
[28] Holmstrom, B. (1977): On incentives and control in organizations, Ph D dis-
sertation, Stanford University.
[29] Hurwicz, L. (1960): Optimality and informational eciency in resource alloca-
tion processes, in Arrow, Karlin and Suppes (eds.), Mathematical Methods in
the Social Sciences. Stanford University Press.
[30] Hurwicz, L. (1972): On informationally decentralized systems, in Radner and
McGuire, Decision and Organization. North-Holland, Amsterdam.
[31] Hurwicz, L. (1973): The design of mechanisms for resource allocation, American
Economic Review 63, Papers and Proceedings, 1-30.
[32] Hurwicz, L. and D. Schmeidler (1978): Construction of outcome functions guar-
anteeing existence and Pareto-optimality of Nash equilibria, Econometrica 46:
1447-1474.
[33] Jackson, M. (1991): Bayesian implementation, Econometrica 59, 461-477.
[34] Jackson, M. (2001): A crash course in implementation theory, Social Choice
and Welfare 18, 655-708.
[35] Jehiel, P. and B. Moldovanu (2001): Ecient design with interdependent valu-
ations, Econometrica 69, 1237-1259.
22
[36] Joskow, P. (2005): Incentive regulation in theory and practice: Electricity dis-
tribution and transmission networks, CEEPR Working Paper 05-014.
[37] Klemperer, P. (2004): Auctions: Theory and Practice. Princeton Unversity Press,
Princeton.
[38] Krishna, V. (2002): Auction Theory. Academic Press, New York.
[39] Laont, J.-J. and D. Martimort (2002): The Theory of Incentives. Princeton
University Press, Princeton.
[40] Laont, J.-J. and E. Maskin (1979): A dierentiable approach to expected
utility-maximizing mechanisms, in Laont (ed.), Aggregation and Revelation of
Preferences. North-Holland, Amsterdam.
[41] Laont, J.-J. and J. Tirole (1986): Using cost observation to regulate rms,
Journal of. Political Economy 94, 614-641.
[42] Laont, J.-J. and J. Tirole (1987): Auctioning incentive contracts, Journal of
[43] Laont, J.-J. and J. Tirole (1988): The dynamics of incentive contracts, Econo-
metrica 56, 1153-1175.
[44] Laont, J.-J. and J. Tirole (1993): A Theory of Incentives in Procurement and
Regulation. MIT Press, Cambridge.
[45] Laont, J.-J. (1994): The new economics of regulation ten years after, Econo-
metrica 62, 507-537.
[46] Leininger, W., Linhart, P. B. and R. Radner (1989).Equilibria of the sealed-bid
mechanism for bargaining with incomplete information, Journal of Economic
Theory 48, 63-106.
[47] Loeb, M. and W. Magat (1979): A decentralized method of utility regulation,
Journal of Law and Economics 22, 399-404
[48] Mailath, G. and A. Postlewaite (1990): Asymmetric bargaining problems with
many agents, Review of Economic Studies 57, 351-367.
23
[49] Marshak, J. and R. Radner (1972): Economic Theory of Teams. Yale University
Press.
[50] Mas-Colell, A., Whinston M., and J. Green (1995): Microeconomic Theory. Ox-
ford University Press
[51] Maskin, E. (1977): Nash equilibrium and welfare optimality. Paper presented at
the summer workshop of the Econometric Society in Paris, June 1977. Published
1999 in the Review of Economic Studies 66, 23-38.
[52] Maskin, E. (1992): Auctions and privatization, in H. Siebert (ed.), Privatiza-
tion: Symposium in honor of Herbert Giersh. Mohr (Siebek), Tubingen.
[53] Maskin, E. and J. Moore (1999): Implementation and renegotiation, Review of
Economic Studies 66, 39-56.
[54] Maskin, E. and J. Riley (1984a): Optimal auctions with risk-averse buyers,
[55] Maskin, E. and J. Riley (1984b): Monopoly with incomplete information,
RAND Journal of Economics 15, 171-196.
[56] Maskin, E. and T. Sjstrm (2002): Implementation theory, in K. Arrow, A.K.
Sen and K. Suzumura (eds.), Handbook of Social Choice and Welfare, Vol. 1.
Elsevier Science, Amsterdam.
[57] Maskin, E. and J. Tirole (1999): Unforeseen contingencies and incomplete con-
tracts, Review of Economic Studies 66, 84-114.
[58] Matsushima, H. (1988): A new approach to the implementation problem, Jour-
nal of Economic Theory 45, 128-144.
[59] McAfee, R.P. and J. McMillan (1986): Bidding for contracts: A principal-agent
analysis, RAND Journal of Economics 17, 326-338.
[60] McAfee, R.P. (1993): Mechanism design with competing sellers, Econometrica
61, 1281-1312.
[61] McAfee, R.P., J. McMillan and M. Whinston (1989): Commodity bundling by
a monopolist, Quarterly Journal of Economics 104, 371-383.
24
[62] McMillan J. (1994): Selling spectrum rights, Journal of Economic Perspectives
8, 145-162.
[63] Mirrlees, J. (1971): An exploration in the theory of optimum income taxes,
[64] Milgrom, P. (1979): A convergence theorem for competitive bidding with dier-
ential information, Econometrica 47, 679-688.
[65] Mookherjee, D. and S. Reichelstein (1990): Implementation via augmented rev-
elation mechanisms, Review of Economic Studies 57, 453-475.
[66] Moore, J. (1992): Implementation, contracts, and renegotiation in environments
with complete information, in J.J. Laont (ed.), Advances in Economic Theory
(Sixth World Congress) Vol. 1. Cambridge University Press, Cambridge.
[67] Moore, J. and R. Repullo (1988): Subgame perfect implementation, Economet-
rica 56, 1191-1220.
[68] Moulin, H. (1994): Social Choice, in R. Aumann and S. Hart (eds.), Handbook
of Game Theory, Vol. 2. Elsevier Science, Amsterdam.
[69] Muller, E. and Satterthwaite (1977): The equivalence of strong positive associ-
ation and strategy-proofness, Journal of Economic Theory 14, 412-418.
[70] Muller, E. and M. Satterthwaite (1985): Strategy-proofness: the existence of
dominant-strategy mechanisms, in L. Hurwicz, D. Schmeidler and H. Sonnen-
schein (eds.), Social Goals and Social Organization. Cambridge University Press,
Cambridge.
[71] Mussa and Rosen (1978): Monopoly and product quality, Journal of Economic
Theory 18, 301-317.
[72] Myerson, R. (1979): Incentive compatibility and the bargaining problem,
[73] Myerson, R. (1981): Optimal auction design, Mathematics of Operations Re-
search 6, 58-73.
[74] Myerson, R. (1982): Optimal coordination mechanisms in generalized principal-
agent problems, Journal of Mathematical Economics 11, 67-81.
25
[75] Myerson, R. (1983): Mechanism design by an informed principal, Econometrica
52, 461-487.
[76] Myerson, R. (1986): Multistage games with communication, Econometrica 54,
323-358.
[77] Myerson, R. (1989): Mechanism design, in J. Eatwell, M. Milgate and P. New-
man (eds.), The New Palgrave: Allocation, Information and Markets. Norton,
New York,
[78] Myerson, R. and M. Satterthwaite (1983): Ecient mechanisms for bilateral
trading, Journal of Economic Theory 28, 265-281.
[79] Osborne, M. and A. Rubinstein (1994): A Course in Game Theory. MIT Press
[80] Palfrey, T. (2001): Implementation Theory, in R. Aumann and S. Hart (eds.),
Handbook of Game Theory Vol. 3. North-Holland, Amsterdam.
[81] Palfrey, T. and S. Srivastava (1989): Implementation with incomplete informa-
tion in exchange economies, Econometrica 57, 115-134.
[82] Palfrey, T. and S. Srivastava (1991): Ecient trading mechanisms with pre-play
communication, Journal of Economic Theory 55, 17-40.
[83] Peters, M. (1997): A competitive distribution of auctions, Review of Economic
Studies 64, 97-123.
[84] Postlewaite, A. and D. Schmeidler (1986): Implementation in dierential infor-
mation economies, Journal of Economic Theory 39, 14-33.
[85] Prescott, E. and R. Townsend (1984): Pareto optima and competitive equilibria
with adverse selection and moral hazard, Econometrica 52, 21-46.
[86] Repullo, R. (1987): A simple proof of Maskins theorem on Nash-
implementation, Social Choice and Welfare 4, 39-41.
[87] Riley, J. and W. Samuelson (1981): Optimal auctions, American Economic
Review 71, 381-392.
[88] Riordan, M. and D. Sappington (1987): Awarding monopoly franchises, Rand
Journal of Economics, 40: 21-58.
26
[89] Rob, R. (1989): Pollution claims settlement with private information, Journal
of Economic Theory 47, 307-333.
[90] Roberts, J. (1976): The incentives for correct revelation of preferences and the
number of consumers, Journal of Public Economics 6, 359-374.
[91] Rosenthal, J.-L. (1992): The Fruits of Revolution: Property Rights, Litigation,
and French Agriculture, 1700-1860. Cambridge University Press.
[92] Rosenthal, R. (1978): Arbitration of two-party disputes under uncertainty,
[93] Saijo, T. (1988): Strategy space reduction in Maskins theorem, Econometrica
56, 693-700.
[94] Salani, B. (1997): The Economics of Contracts. MIT Press.
[95] Samuelson, P. (1954): The pure theory of public expenditure, Review of Eco-
nomics and Statistics 36, 387-389.
[96] Samuelson, W. (1984): Bargaining under asymmetric information, Economet-
rica 52, 995-1005.
[97] Sappington, D. (1982): Optimal regulation of research and development under
imperfect information, Bell Journal of Economics 13, 354368.
[98] Sappington, D. (1983): Optimal regulation of a multiproduct monopoly with
unknown technological capabilities, Bell Journal of Economics 14, 453463.
[99] Satterthwaite, M. (1975): Strategy-proofness and Arrows conditions: Existence
and correspondence theorems for voting procedures and welfare functions, Jour-
nal of Economic Theory 10, 187-217.
[100] Serrano, R. (2004): The theory of implementation of social choice rules, SIAM
Review 46, 377-414.
[101] Selten, R. (1975): A re-examination of the perfectness concept for equilibrium
points in extensive form games, International Journal of Game Theory 2, 141-
201.
27
[102] Stole, L. (1995): Nonlinear pricing and oligopoly, Journal of Economics and
Management Strategy 4, 529-562.
[103] Tideman, T.N. and G. Tullock (1976): A new and superior process for making
social choices, Journal of Political Economy 84, 1145-59.
[104] Tirole, J. (2005): The Theory of Corporate Finance. Princeton University Press,
Princeton.
[105] Vickrey, W. (1961): Counterspeculation, auctions and competitive sealed ten-
ders, Journal of Finance 16, 8-37.
[106] Weitzman, M. (1978): Optimal rewards for economic regulation, American
[107] Williams, S. (1986): Realization and Nash implementation: Two aspects of
mechanism design, Econometrica 54, 139-151.
[108] Wilson, R. (1977): Abidding model of perfect competition, Review of Economic
Studies 44, 511-518.
[109] Wilson, R. (1979): Auctions of shares, Quarterly Journal of Economics 93,
675-689.
[110] Wilson, R. (1985): Incentive eciency of double auctions, Econometrica 53,
1101-1116.
[111] Wolak, F. (1994): An econometric analysis of the asymmetric information,
regulator-utility interaction, Annales dEconomie et de Statistique 34, 13-69.
28
____________________________________________________________________________________________________

9 October 2006

Advanced information on Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel
Edmund Phelpss Contributions to
Macroeconomics

1
Edmund Phelpss Contributions to Macroeconomics

The relationship between inflation and unemployment and the tradeoff between the welfare of
current and future generations are key issues in macroeconomic research. They have a strong
influence on the choice of macroeconomic policy. Both issues concern tradeoffs between
different objectives. How should fiscal and monetary policy resolve the conflict between the
goals of low inflation and low unemployment? How should society trade off consumption today
against consumption in the future, i.e., how much should be saved in order to increase future
consumption? Edmund Phelps has made major contributions to the analysis of both of these
tradeoffs. In particular, he had the insight that also the balance between inflation and
unemployment reflects a fundamentally intertemporal problem.

Phelps pointed out that current inflation depends not only on unemployment but also on inflation
expectations. Such dependence is due to the fact that wages and prices are adjusted only
infrequently. Consequently, when adjustments are made, they are based on inflation forecasts.
Thus, the higher the anticipated rate of inflation, the higher the unemployment required to reach
a certain actual inflation rate. Phelps formulated the so-called expectations-augmented Phillips
curve. The intertemporal perspective implies that current inflation expectations affect the future
tradeoff between inflation and unemployment. A higher current inflation rate typically leads to
higher inflation expectations in the future, so that it then becomes more difficult to achieve the
objectives of stabilization policy.

Phelpss contributions from the late 1960s and early 1970s radically changed our perception of
the interaction between inflation and unemployment. The new theory made it possible to better
understand the underlying causes of the increases in inflation and unemployment that took place
2
during the 1970s. A key result was that the long-run rate of unemployment cannot be influenced
by monetary or fiscal policy affecting aggregate demand. Phelpss analysis thus identified
important limitations on what demand-management policy can achieve. This view has become
predominant among macroeconomic researchers as well as policymakers. As a result,
macroeconomic policy is carried out very differently today from what it was forty years ago.

Modern theory of capital accumulation and economic growth originated with the so-called
Solow-Swan model from the mid-1950s, which shows how capital accumulation and
technological change in combination lead to growth in output and consumption. But the original
analysis of the model did not provide any guidance concerning appropriate rates of accumulation
of physical capital and R&D. Through a number of contributions in the 1960s, Phelps focused on
the intergenerational aspect of the savings problem. He showed that optimal capital
accumulation, subject to a requirement of equal treatment of future generations of consumers,
delivers a simple rule of thumb for long-run saving: the aggregate savings rate should equal the
share of capital in national income. He explored conditions under which all generations would
gain from a change in the rate of savings. He also analyzed the role of human capital for
economic growth by arguing that an educated, well-trained workforce is better able to adopt
available new technologies. As a result, the rate of output growth should be expected to be
positively related to the average level of human capital a proposition that has found empirical
support.

Phelpss work on inflation and unemployment
In the early postwar period, macroeconomics was dominated by Keynesian views on how the
economy operates. According to Keynesian theory, there was no conflict between full
employment and price stability. As long as the economy was below full employment, inflation
3
was assumed to be unaffected by an increase in aggregate demand, which could be achieved
through fiscal or monetary measures. Indeed, from the Keynesian perspective the task of
stabilization policy seemed almost trivial: simply keep aggregate demand high enough to avoid
underemployment but not so high as to lead to excess demand for labor (overemployment) and
inflation.

According to the Phillips curve (Phillips, 1958) there was instead a stable negative relationship
between inflation and unemployment. This led to a revision of the standard Keynesian model of
the economy. The Phillips curve implied a tradeoff between inflation and unemployment. While
it was still conceivable that employment could be increased permanently using aggregate-
demand policy, this would occur at the cost of higher inflation. The rise in inflation, however,
would be a one-shot change from one stable level to another. The Phillips curve thus seemed to
provide a menu of choice for policymakers, who could choose between inflation and
unemployment according to their preferences.

There were several problems with this view. One problem was the absence of convincing
microeconomic foundations for several relationships. The Phillips curve, in particular, was
essentially a statistical correlation with rather tenuous theoretical underpinnings. The common
interpretation, first suggested by Lipsey (1960), was that unemployment could be seen as a proxy
for excess demand (supply) in the labor market, so that the Phillips curve should be regarded as
an equation describing how the price (wage) level responds over time to an imbalance between
demand and supply. A second problem was inherent in the view that higher employment could
be permanently achieved by allowing higher inflation. This idea was in conflict with the
traditional presumption in economic theory that, in the long run, real magnitudes in the economy
are determined by real rather than nominal forces. A third problem was the lack of any theory
4
regarding the determinants of the unemployment prevailing at full employment, so-called
frictional unemployment. Although it was generally accepted that full employment did not
literally mean zero unemployment, there was no theory that specified the determinants of
frictional unemployment.

Phelpss research program from the end of the 1960s aimed at rectifying the theory of inflation
and unemployment by explicitly modeling firms wage and price-setting behavior. Phelps
brought agents expectations to the forefront of the analysis, made the crucial distinction between
expected and unexpected inflation, and examined the macroeconomic implications of this
distinction. His reformulation of the Phillips curve has become known as the expectations-
augmented Phillips curve. Unlike previous studies, such as Lipsey (1960), Phelps emphasized
that it was the difference between actual and expected inflation, and not inflation per se, that is
related to unemployment.

Phelpss analysis was at variance with earlier views that higher employment could permanently
be achieved by inflationary demand policies. Indeed, it implied that there was no long-run
tradeoff between inflation and unemployment, as there could be no permanent discrepancy
between actual and expected inflation. This hypothesis of a vertical long-run Phillips curve at the
equilibrium unemployment rate is one of the most influential ideas in macroeconomics over the
past 50 years.
1
The hypothesis has become crucial for the conduct of monetary policy. Modern
central banks generally make their decisions on interest rates on the basis of estimates of the

1
The Phillips curve is typically illustrated in a figure with inflation on the vertical axis and unemployment on the
horizontal axis.
5
equilibrium unemployment rate.
2
In this context, Phelps also provided the first analysis of the
determinants of the equilibrium unemployment rate.

Phelpss most important contributions to the theory of inflation and unemployment consist of
three papers: Phelps (1967, 1968a, and 1970b). The 1970 paper is an extension of his 1968
one and appeared in the famous monograph Microeconomic Foundations of Employment and
Inflation Theory, usually referred to as the Phelps volume (1970a). In his 1967 paper,
Phelps analyzed optimal demand policy when there is no long-run tradeoff between inflation
and unemployment. The 1968 and 1970 papers studied wage setting and equilibrium
unemployment when markets are characterized by frictions. Combined, these papers contain
the core of the new insights in Phelpss program.

The Phillips curve and optimal inflation policy
How should monetary (or fiscal) policy be conducted when there is a short-run but no long-run
tradeoff between inflation and unemployment? Phelps (1967) took as given the inflation-
unemployment mechanism for which he provided micro foundations in Phelps (1968a). Here, for
the first time the notion of the expected rate of inflation was used in the Phillips-curve
literature. In particular, Phelps introduced an expectations-augmented Phillips curve:

( )
e
f u = + ,

where is the actual rate of inflation,
e
is the expected rate of inflation and f(.) a decreasing
function of unemployment, u.
3
According to the equation, actual inflation depends on both

2
As recognized by Phelps (1968a), the idea that perfectly anticipated inflation has no real effects on the
economy had appeared in earlier studies, such as Lerner (1949). However, this literature had not integrated this
hypothesis within explicit models of unemployment and wage dynamics.
6
unemployment and the expected rate of inflation. A one percentage point increase in expected
inflation leads to a one percentage point increase in actual inflation at a given rate of
unemployment. The equilibrium rate of unemployment is obtained as the outcome of an
expectational equilibrium with
e
= , i.e., a situation with equality between the current and the
expected inflation rate. That is, the equilibrium rate,
*
u , is given by
*
( ) 0 f u = . If expectations
are based on observations of inflation in the recent past so-called adaptive expectations it
follows that inflation will be increasing as long as
*
u u < and decreasing when
*
u u > .
4
This is
often referred to as the accelerationist hypothesis. The term NAIRU (the non-accelerating
inflation rate of unemployment) is often used as a synonym for the equilibrium unemployment
rate.

Phelpss formulation of the Phillips curve implies that the task of demand management policy is
no longer to solve the static optimization problem of achieving the best attainable combination of
inflation and unemployment at a given point in time. Instead, the problem is dynamic and
involves finding the socially optimal paths for inflation and unemployment over time. To that
end, Phelps introduced a social utility function that takes the form of a (possibly) discounted sum
of instantaneous utility flows. The utility flow at each point in time depends in turn on
consumption and leisure. To solve the optimization problem, policymakers have to recognize
that lower unemployment and higher inflation today will raise expected inflation. This implies an
intertemporal cost as it deteriorates the tradeoff between inflation and unemployment in the
future. In Phelpss model, the dynamic path of unemployment converges to a steady state where
actual and expected inflation coincide and unemployment coincides with the equilibrium

3
Phelps actually focused on employment rather than unemployment, but employment is of course just the mirror
image of unemployment if the labor force is fixed.
4
More generally, the adaptive expectations hypothesis can be written as d
e
/dt =(-
e
), where >0.
According to the equation the expected inflation rate at each point in time is adjusted to the difference between
actual and expected inflation. Combining this equation with the one in the text and holding unemployment
constant gives d
e
/dt =d/dt =f(u), resulting in d/dt >0.
7
unemployment rate. The optimization will also determine the preferred inflation rate in the
steady state. The exact path depends, inter alia, on the initial expected inflation rate. If this rate is
higher than the preferred steady-state level, it will be optimal to go through a period of
temporary underemployment in order to bring down inflation.

This intertemporal approach to monetary policy has become standard. Contemporary academic
analyses of monetary policy, as well as policy deliberations by central banks, focus on
intertemporal tradeoffs, where short-run changes in activity are weighed against the effects on
the possibilities to hold down both inflation and unemployment in the future. The theoretical
underpinnings for the policy of inflation targeting, which many central banks have adopted since
the early 1990s, are to a large extent derived from the framework developed in Phelpss 1967
paper.
5

Later work on monetary policymaking has departed from Phelpss assumption that inflation
expectations are adaptive. Based on the original work of Muth (1961) and Lucas (1972, 1973),
the standard theoretical assumption now is that expectations are rational, i.e., they are forward
looking and correct on average. Empirical studies suggest that inflation expectations do have a
backward-looking component, thus pointing to the practical relevance of Phelpss insight that
both inflationary and disinflationary processes are likely to be gradual. Recent theoretical
research has also attempted to explain why past inflation expectations may, after all, have a
lasting effect on current inflation.
6
Some of this literature has been inspired by Phelpss own
critique of the rational-expectations hypothesis. Fully rational expectations appear implausible
when agents have to form expectations not only about aggregate conditions, but also about other

5
See Clarida, Gal and Gertler (1999) or Woodford (2003a) for surveys of current theories on monetary policy-
making. Phelps (1978b) provides another account of his views on the problem of inflation planning.
6
This includes contributions such as Sargent (1999), Evans and Honkapohja (2001), Mankiw and Reis (2003),
Woodford (2003b), and Orphanides and Williams (2005),
8
agents beliefs about their own beliefs and so on in an infinite chain (Phelps, 1983, Frydman and
Phelps, 1983).

Microeconomic foundations for wage and price setting
Phelps (1968a, 1970b) derived aggregate wage-setting behavior from detailed modeling of the
behavior of individual agents. J obs and workers are heterogeneous and there is imperfect
information on both sides of the market. Markets are assumed to be almost atomistic, but there is
no Walrasian auctioneer that instantaneously finds the wages (and prices) that clear all markets
(as went the metaphor used in earlier general equilibrium analysis). Instead, wages are set by
firms that are able to exercise temporary monopsony power. Workers and firms meet randomly
at a rate determined by the number of unemployed workers searching for a job and the number of
vacancies, according to a function that would today be recognized as a matching function.
Phelpss work here is a precursor of the search and matching theory of unemployment, where
Peter Diamond, Dale Mortensen, and Christopher Pissarides have made especially important
contributions.
7

In Phelpss model, each firm anticipates that it can increase its hiring rate and decrease its quit
rate by raising its relative wage, i.e., its wage relative to the average of expected wages paid by
other firms. The firm tries to set a relative wage that increases with the firms desired net hiring
rate and decreases with the current rate of unemployment. Under these assumptions, Phelps
showed that there will be a unique rate of unemployment, the equilibrium rate, at which the
average firm will raise its wage offer at a rate equal to the expected rate of increase in the
average wage rate.

7
See, for example, Diamond (1984), Mortensen (1982a, b), Mortensen and Pissarides (1994), and Pissarides
(2000).
9

Phelps (1968a) contains several innovations. For the first time detailed microeconomic reasoning
concerning market interactions and the determination of wages and prices were introduced into
mainstream macroeconomic theory. While this may seem self-evident today, it was not so at the
time. On the contrary, macroeconomic relationships used in analytical models usually had the
character of broad empirical generalizations, and were not based on explicit modeling of
individual behavior. In Phelpss analysis, neither demand nor supply determined the quantity of
labor traded. Instead the market was typically characterized by the simultaneous existence of
vacant jobs and unemployed workers, an implication of the frictions captured by the matching
function.

Around the same time as Phelps presented his models of equilibrium unemployment, Milton
Friedman (economics laureate in 1976) provided his well-known critique of the Phillips curve
(Friedman, 1968). Like Phelps, Friedman emphasized the importance of inflation expectations
for the inflation-unemployment tradeoff. Friedman shared Phelpss view that unemployment
cannot be permanently reduced by expansionary demand policy and that there exists an
equilibrium rate of unemployment which is determined by the real features of the economy. In
Friedmans terminology, this rate was labeled the natural rate of unemployment.

Unlike Phelps, Friedman did not embed his discussion of inflation and unemployment in a
formal model. It was an open question whether the expectations-augmented Phillips curve should
be interpreted as a price adjustment relationship in a situation with disequilibrium between
demand and supply (as was the case in Phelpss analysis) or as an aggregate supply equation in a
competitive labor market where instantaneous wage adjustments clear the labor market at all
times. Under the latter interpretation, the causation runs from a deviation of actual from expected
10
inflation to unemployment instead of the other way around as in Phelpss work. The idea then is
that employers and workers perceptions of inflation differ. A rise in inflation induces
employers to offer higher money wages. If the rise is unanticipated by workers, the higher
money wage is interpreted by them as a rise in the real wage and will make them offer a larger
labor supply (since labor supply is taken to depend on expected rather than actual real wages).
This reduces the actual real wage (which is what matters for firms employment decisions) and
therefore leads to an expansion of employment and output.
8
Arguably due to its more realistic
underpinnings, Phelpss approach is the one that has had more influence on the subsequent
development of models of inflation and unemployment, in particular within the so-called New
Keynesian analysis of inflation and monetary policy.

Another major contribution by Phelps is his formulation of an explicit model of an imperfect
labor market with frictions, job search behavior and wage-setting firms that can explain the
determinants of equilibrium unemployment. Phelps (1968a) also represented the first explicit
analysis of the equilibrium unemployment rate as the unemployment rate consistent with
expectational equilibrium. It also provided the first formal model of so-called efficiency wages
embedded in a macroeconomic framework. The general idea of efficiency wage models is that a
firm may find it profitable to set its wage above the market-clearing level, thereby improving
worker morale (less shirking), reducing costly labor turnover or improving the quality of the pool
of job applicants. Phelps focused on the relationship between the firms relative wage and labor
turnover, an approach further developed by others, e.g., Salop (1979), as well as by Phelps
himself in his more recent work on unemployment theory; see e.g. Phelps (1994).
9

8
This is Friedmans own interpretation of his theory, as is clear from his Nobel lecture; see Friedman (1977).
9
Phelps often describes his analysis of wage setting in terms of incentive wages rather than efficiency wages.
The most well-known efficiency wage model is probably that of Shapiro and Stiglitz (1984), who stressed how
firms set wages to provide an incentive for employees not to shirk. See also Layard, Nickell and J ackman (1991).
11
In his 1968 paper, Phelps also considers the possibility that the long-run Phillips curve may be
negatively sloped at very low rates of inflation, although it is vertical at higher rates. The
explanation given is that, in such a situation, a reduction in the expected rate of wage increase
may not translate one-to-one into a reduction in the actual rate of wage increase. The reason is
that a reduction would imply cuts in the money wage level in some firms (those hit by the largest
adverse demand shocks), to which there is strong psychological resistance. Phelpss discussion
here is a precursor to later work, initiated by Akerlof, Dickens and Perry (1996), who find
empirical support for a negatively sloped long-run Phillips curve at rates of inflation below 1-2
per cent.

In collaboration with Sidney Winter, Phelps also made an important contribution to the theory of
price setting (Phelps and Winter, 1970). The model, subsequently known as the Phelps-Winter
model, seeks to explain why prices need not fully reflect short-term fluctuations in marginal
costs. Although firms produce a homogeneous good in the model, they can exercise transitory
monopoly power because consumers have imperfect information about the distribution of prices
across sellers. A firm can therefore increase its price in the short run without immediately losing
its customers, but it will have no market power in the long run (since keeping its price higher
than competitors prices will eventually imply a loss of all customers). When setting its price,
each firm trades off the gain from exploiting its short-run market power against the future
reduction in profits that will occur if a price differential to competitors leads to a loss of
customer stock. The optimal price path converges to a steady state where the price is higher than
the marginal cost (because each firm chooses to exercise some temporary monopoly power), but
lower than the static monopoly price (because this would erode the firms future market share).
A noteworthy feature of the model is that it can generate pro-cyclical movements in the real
product wage (the ratio between the wage and the product price), i.e., output may rise although
12
the firms price falls relative to the nominal wage.
10
Nowadays, the idea that business
fluctuations give rise to changes in the relationship between prices and marginal costs is
commonplace in macroeconomics. The literature includes a variety of models with different
mark-up predictions (see, e.g., Stiglitz, 1984, and Rotemberg and Woodford, 1999).

Subsequent research on inflation and unemployment
The consequences of imperfect information became a central theme in the New Classical
research agenda launched by Robert Lucas in the 1970s (Lucas, 1972, 1973, 1976). In his 1972
analysis, Lucas studied how business-cycle fluctuations result from money surprises, using the
island parable from Phelps (1969), but Lucas introduced rational expectations instead of the
adaptive expectations used by Phelps.
11
It turned out that Phelpss result that money surprises
could have temporary real effects even in the absence of nominal rigidity was robust under
rational expectations. A principal conclusion was that systematic monetary policy had no role to
play, as it would be built into inflation expectations and thus lead to price changes that would
nullify its effects. This was forcefully argued by Sargent and Wallace (1975), who claimed that a
feedback monetary policy rule could have no stabilizing effect on the economy: only errors on
the part of the central bank or unexpected changes of the rule could influence output and
employment.

The stabilizing power of monetary policy was the subject of an influential paper by Phelps and
one of his students, J ohn Taylor (Phelps and Taylor, 1977). Their objective was to examine
whether a case could be made for stabilization policy by introducing a modest degree of wage

10
The explanation is that the price mark-up over marginal cost may fall in a boom. If, instead, the price is set as
a fixed mark-up over marginal cost (or is equal to marginal cost as is the case with perfect competition), a rise in
output is associated with a fall in the real product wage. The reason is that an increase in output (employment)
implies a fall in the marginal product of labor, which tends to increase the marginal cost. With a fixed mark-up
over the marginal cost, it follows that the price will then rise relative to the wage.
11
Phelpss island parable perceives the economy as a group of islands between which information flows are
costly. To learn the wage paid on a nearby island, a worker must spend the day travelling to the island to locate a
wage offer rather than spending the day at work.
13
and price stickiness into an otherwise standard rational-expectations model.
12
The key
assumption was that monetary policy can be changed more frequently than wages and prices, so
that monetary policymakers could act on the basis of a larger information set than price and wage
setters. This appears to be an important real-world feature, as wage contracts are often long term.
In such a framework, a negative shock to aggregate demand could generate a prolonged
recession even if expectations were rational. An important result was that although systematic
policy could not affect the average size of the real output gap, it could determine the variance of
that gap, as well as the variance of inflation. The idea of a tradeoff between the variances of
different policy objectives is of primary importance in modern discussions of monetary rules
under rational expectations. The Phelps-Taylor paper is the first where such a tradeoff is derived
under rational expectations.

In other contributions, Phelps studied the consequences of unsynchronized wage contracts, i.e.,
the empirical fact that contract decisions are typically staggered over time rather than made at the
same point in time. Phelps (1970b) provided an early analysis and Phelps (1978a) showed that
such staggering could substantially increase the persistence of the real effects of monetary
disturbances. Taylor (1979) showed how staggered wage setting could be embedded in a model
of wage and output dynamics under rational expectations. Another of Phelpss students,
Guillermo Calvo, modeled staggered prices in an economy where old prices die off
asymptotically, an approach that greatly facilitated the analysis (Calvo, 1983). This particular
way of modeling price adjustments has become very common in the New Keynesian literature.

The Phelps volume (1970a) was followed two years later by a new book, Inflation Policy and
Unemployment Theory, dealing with the design of monetary policy (Phelps, 1972a). The

12
Stanley Fischer (1977) elaborated on the same basic theme, i.e., the stabilizing power of monetary policy even
under rational expectations. Fischer considered overlapping labor contracts that introduced an element of wage
stickiness into the model.
14
ambitious task was to examine the principles of optimal inflation policy and conduct a dynamic
cost-benefit analysis of monetary policy. To some extent, this book served to popularize Phelpss
previous work on the expectations-augmented Phillips curve and to draw policy conclusions
from the new theory. A lasting contribution from this book is the idea of hysteresis in the
unemployment rate: an increase in unemployment may turn out to be (partially) irreversible, as a
result of workers loss of skill and morale. Analogously, a temporary increase in employment
may bring about lasting reductions in equilibrium unemployment as a result of favorable effects
on work experience. Phelpss idea remained largely ignored for more than a decade, but came to
life again in the mid-1980s when economists struggled to understand the seemingly permanent
rise in European unemployment.
13
This research has demonstrated that unemployment
persistence is a pervasive feature of many European labor markets, although the relative
importance of different persistence mechanisms is still not fully understood.

The European unemployment problem also induced Phelps to dig deeper into the causes of
unemployment. His analyses may be found in a number of journal articles and especially in the
monograph Structural Slumps from 1994. Here Phelps turns away from the temporary
expectational disequilibrium causes of unemployment and focuses on the real (structural) causes
that determine equilibrium unemployment. This work forms part of a comprehensive research
literature that has developed over the last twenty years, to a large extent inspired by the work of
Layard, Nickell and J ackman (1991).
14
A distinctive feature of Phelpss later work is its
emphasis on the importance of capital markets for the development of unemployment. A main
hypothesis is that a higher real rate of interest tends to raise equilibrium unemployment through
negative effects on the incentives of firms to invest (in physical capital, low labor turnover,
customer stock, etc.). Although Phelps finds empirical support for the importance of the real rate

13
The adverse effects of high unemployment on human capital have been emphasized in the influential study of
European unemployment by Layard, Nickell and J ackman (1991).
14
See Calmfors and Holmlund (2000) or Blanchard (2006) for recent surveys.
15
of interest for equilibrium unemployment, there is no general consensus in the empirical
literature regarding the quantitative importance of this mechanism.

Phelpss work on capital accumulation
Phelpss perspective on the tradeoff between inflation and unemployment was fundamentally
intertemporal: low current inflation can be viewed as an investment in low inflation expectations,
thereby allowing a more favorable inflation-unemployment tradeoff in the future. To Phelps, the
intertemporal perspective was natural, since his first research interests concerned capital
accumulation. Over a ten-year period beginning in the early 1960s Phelps made a number of
significant contributions in the area of capital accumulation and economic growth.

In the late 1950s and early 1960s, a view emerged in the public debate that the aggregate US
savings rate was too low. A key issue in the discussion was how society should trade off the
consumption of current citizens against that of future citizens. Phelps emphasized the objective
of fairness among generations. Based on this view, his research examined how to best
accumulate capital. He advanced the notion of dynamic inefficiency and used it to provide an
upper bound on what the savings rate should be. Somewhat later, he returned to explicit welfare
comparisons across generations of consumers and suggested a new way of thinking about saving
decisions. The upshot was that savings rates could, in fact, be too low. In joint work with Robert
Pollak, Phelps suggested that an induced increase in the rate of saving for all generations could
be welfare-improving for all of them; see Phelps and Pollak (1968). All in all, Phelps generated a
rich set of insights about optimal capital accumulation that constituted important advances in the
field and have become cornerstones of capital theory.

16
Phelps carried out much of his analysis in the context of the growth model that is still the core of
modern growth theory: the neoclassical growth model, developed in the mid-1950s by Robert
Solow (1956, 1957; economics laureate in 1987) and Trevor Swan (1956). Using a stylized
production structure, the Solow-Swan model describes how capital accumulation and
technological progress generate output growth.
15
In the model, the economy produces one output
good, which can be used for either investment or consumption, with labor and capital as inputs.
The production function has constant returns to scale but features diminishing marginal returns
to the use of any given input. Investment is assumed to be a constant, exogenous fraction of
output. Thus, Solow and Swan merely assumed, without a normative evaluation, how much was
saved and then derived the implications for the dynamics of output. Phelpss first question can be
framed exactly in these terms: in the context of the Solow-Swan theory, what should the rate of
savings ideally be? Later, Phelps analyzed broader questions, which involved a departure from
the neoclassical one-dimensional setup: what form should savings take, e.g., how should
investments in R&D be made, and what is the role of human capital for economic growth?

Capital theory did not start with Solow and Swan, and the analysis of optimal saving did not start
with Phelps. The first fully rigorous analysis of optimal saving is the study by Frank Ramsey
(1928), who formulated the individual households saving problem as an intertemporal
maximization problem. The household chooses among different feasible consumption sequences
in order to maximize a utility function that depends on a discounted stream of utility flows from
consumption at different dates. A society-wide perspective on saving and capital accumulation,
as well as early formal contributions to capital theory and, more generally, to competitive
equilibrium theory in intertemporal models, may be found in the work of J ohn von Neumann
(1945/46), Maurice Allais (1947; economics laureate in 1988), and Edmond Malinvaud (1953).

15
In addition, Solow (1957) provided tools for an empirical evaluation of different sources of output growth, so-
called growth accounting.
17
Later, capital theory was further developed and refined by David Cass (1965, 1972) and Tjalling
Koopmans (1965; economics laureate in 1975) and others. By the late 1980s, attention turned
anew to economic growth. A vibrant literature developed capital theory towards explaining the
origins of technical change and the growth of knowledge, while also focusing on cross-country
differences in growth.
16
Some of Phelpss contributions had an immediate impact on the field,
but others, such as those on the roles of R&D and human capital, were far ahead of their time
and had their impact only after a considerable time lag.

Capital accumulation and intergenerational tradeoffs
One of Phelpss first publications is perhaps his best known contribution: his 1961 article
containing the so-called golden rule of capital accumulation. Using a Solow-Swan neoclassical
growth model, Phelps set out to find the most desirable long-run savings rate. His criterion was
the maximization of per-capita consumption in the long run, thereby restricting attention to so-
called steady states of the Solow-Swan model where the stock of capital and consumption are
constant in per-capita terms.
17
The term golden rule is a reference to the ethic of reciprocity: Do
unto others as you would have them do unto you. The constancy of the savings ratio across
generations in a steady state is thus the notion of treating others as oneself in this context. Phelps
showed that the steady state with the highest consumption can be characterized by a simple
prescription: the savings rate should equal the share of capital in national income. This is the
golden rule, which can also be stated differently: the return to capital should equal the growth
rate of output.
18
This last expression had appeared earlier in the work of Maurice Allais on

16
See Romer (1996) and Lucas (1988). For comprehensive treatments of endogenous growth theory, see Aghion
and Howitt (1999), Barro and Sala-i-Martin (2004), and Grossman and Helpman (1991).
17
In order to simplify the analysis, Phelps presumed that welfare depended only on consumption; for example,
he abstracted from the value consumers derive from leisure.
18
As an illustration, suppose that per-capita production of output y is f(k), where k is capital per worker, that total
saving is sy, where s is the fraction of income saved, that capital does not depreciate, and that population grows
at a constant rate n. In a steady state with constant capital per worker, sf(k)=nk must hold: any investment is used
to provide capital to new workers. Thus, maximization of per-capita consumption amounts to choosing s and k to
maximize (1-s)f(k) subject to sf(k)=nk, yielding f(k) = n. Under perfect competition, the market return to capital
18
intertemporal production, although Phelpss formulation came to have a greater influence on
subsequent research, since it was derived in the context of the neoclassical growth model, also
allowing a formulation of the golden rule directly in terms of the savings rate.
19

The golden-rule prescription specified a savings rate that was of the same order of magnitude as
that in U.S. data. Therefore, Phelpss analysis did not suggest a radical departure from prevailing
macroeconomic policy. But there was also a limitation in Phelpss (1961) analysis: it focused on
the best (and equitable) long-run outcome, without exploring the costs (or, possibly, benefits)
current generations would have to incur to reach this long-run state. For example, if the golden
rule called for a higher savings rate and a higher long-run capital stock, then such a steady state
could not be reached without the current generations having to forego consumption, which would
leave them worse off than future generations. Thus, the policy path required to reach a golden-
rule state may not be easy to implement politically or even desirable. Phelps, however, went on
to analyze possible circumstances under which, intergenerational concerns notwithstanding,
changes in the savings rate of the economy are desirable.

Phelps (1965) makes the argument that savings rates higher than the golden rule are not
beneficial for any generation, independently of the ways in which consumers derive utility from
consumption. The study thus examined transition paths and went beyond the earlier focus on
long-run comparisons only. Here, Phelps points to the possibility of dynamic inefficiency, an
important concept that others later examined in great detail and gave a complete characterization;

(the price of capital services), r, equals the marginal product of capital: r = f(k). Thus, optimal saving requires r
= n. Since sf(k) = nk = rk, we deduce that capital income, rk, must equal sy, and hence that s=rk/y. The results
generalize easily to allow depreciation of capital at rate and labor-augmenting technical progress at rate g. In
this case we still obtain sy = rk, where now r=n+g+. In other words, the net real interest rate, r-, should equal
the growth rate of total output, g+n.
19
See Allais (1947) and also Allais (1962), which discusses his earlier work. At about the same time as Phelps
(1961), golden-rule results were also published by Desrousseaux (1961), Robinson (1962), von Weizscker
(1962), and Swan (1964). The latter paper, presented at a conference in 1960, derives the golden rule within the
Solow-Swan growth model and expresses it, as Phelps does, in terms of the savings rate. The problem of optimal
capital accumulation is also analyzed in Malinvaud (1953).
19
see, in particular, Cass (1972). Dynamic inefficiency has become a standard concept in
contemporary normative theory of economic growth. Formally, a capital path is dynamically
inefficient if the path of saving can be changed so as to strictly increase consumption at some
point in time without lowering it at any other point in time. Thus, dynamic inefficiency implies
overaccumulation of capital and, when this occurs, it constitutes a drastic market failure in the
sense that resources are simply left on the table. In his 1965 paper, Phelps showed in particular
that whenever the long-run savings rate is above the golden-rule rate, there is dynamic
inefficiency. In his paper, Phelps acknowledged that Koopmans had helped with the proof; the
concept is often referred to as Phelps-Koopmans dynamic inefficiency. The intuition is
straightforward. A lowering of the savings rate increases consumption in the short run, since it
will take time for the capital stock, and hence production, to adjust downward significantly. But
if the initial capital stock is above the golden-rule level, then the drop in savings will also
increase consumption in the long run, since the golden rule, by definition, maximizes long-run
consumption: despite a lower long-run capital stock, consumption will increase.
20
The striking
conclusion, thus, is that regardless of which intergenerational utility perspective is adopted, a
savings rate above the golden-rule rate is too high.
21

If the savings rate lies below the golden-rule rate, an argument for higher saving has to rely on a
more careful consideration of the welfare outcomes for different generations. In joint work with

20
The argument can be made precise by the following example, where, in addition to the symbols used in
footnote 18, A is the level of technology. Consider a discrete-time economy with neither population nor
technology growth, where f(k)=Ak
. Thus, the resource constraint is c(t)+k(t+1)=Ak(t)

+(1-)k(t). Suppose that
the economy is in a steady state with a constant level of capital, k*, so that consumption is c*=A(k*)
- k*. Now
consider decumulating capital from k* to k* - e in one period and keeping it there. Consumption in all future
periods will then be A(k*- e)

- (k*- e). Differentiating with respect to e and evaluating at e=0, one deduces that
consumption must increase if A(k*)
-1
< and e is small enough. Moreover, initial consumption rises to c*+e.
Thus, if k*>(A/)
1/(1-)
, which is the golden-rule level maximizing c*, then all generations would benefit from
reduced savings.
21
Phelps was not the first to discuss efficiency of capital accumulation paths. Based on a given set of
intertemporal prices, Malinvaud (1953) and later others provided conditions for efficient aggregate savings.
However, these studies did not provide guidelines as to which prices should be used. Phelpss characterization,
in contrast, is made directly in terms of the given capital accumulation path.
20
Robert Pollak this is the route Phelps took next; see Phelps and Pollak (1968). The underlying
idea is that each consumer cares not only about his own consumption level, but also about the
consumption levels of his offspring: there is altruism, so that individuals also save to give
bequests. Based on a specific view regarding consumers valuation of their own consumption
and that of future generations, Phelps and Pollak showed that equilibrium savings rates can
sometimes be too low.

Phelps and Pollak assumed that each generation values their offsprings consumption, but not in
the same way as the offsprings do. Suppose that you care about your childs and grandchilds
consumption levels equally, but that you care more about your own, and that your child feels the
same way about his consumption relative to that of his child and grandchild. Then, Phelps and
Pollak argued, you would disagree with your childs savings choice: you would want your child
to save more than he would choose to do. Formally, these differential valuations of different
generations consumption utilities were cast in terms of discounting weights; Phelps-Pollak
discounting has more recently been dubbed quasi-hyperbolic, or quasi-geometric, discounting.
22

Consequently, it is natural to view the savings decisions of different generations of consumers as
determined in a non-cooperative game between generations. Phelps and Pollak carried out their
analysis within such a framework and showed that, absent government intervention, bad
outcomes could arise. In particular, a policy-induced increase in all generations savings rates
could increase the welfare of all generations. Even though a forced rise in your savings rate is, by
itself, not in your interest, if it is accompanied by a forced rise in your offsprings savings rates,

22
Time-inconsistent preferences had been analyzed earlier. Ramsey (1928) considers the possibility, whereas
Strotz (1956) makes the first formal analysis. Pollak (1968) comments on some aspects of Strotzs analysis.
Quasi-hyperbolic discounting, when time is discrete, is characterized by two different discounting parameters
used by a decision-maker at any time t: a (short-run) rate that applies between t and t+1, and a (long-run) rate
that applies between any two adjacent later dates. Thus, when these two parameters differ, there is disagreement.
A decision-maker at timet will attach a different relative weight to utility flows in t+s+1 relative to those in t+s
than will a decision-maker at t+s, for any positive t and s. A present bias obtains when the short-run discount
rate is higher than the long-run discount rate.
21
there is a counteracting effect, which can dominate.
23
This provides a paternalistic argument in
favor of forced savings, suggesting that governments should intervene in private consumption
decisions. This argument has been used to justify, or explain, an active role for governments in
old-age savings plans (pension systems). Recently, the Phelps-Pollak model has also been recast
in terms of the consumption choice of a single individual with time-inconsistent preferences
(say, the consumer who wants to stop smoking, only not right now). As such, the model has had
a far-reaching impact; see, e.g., Laibson (1997) and the recent psychology-and-economics
literature.

Phelpss work on optimal savings was followed by important later contributions, by others as
well as by Phelps himself. Cass (1965) and Koopmans (1965) considered the neoclassical growth
model using dynastic settings, with preferences of the form used in Ramsey (1928), and they
fully characterized optimal savings paths. These studies led to modified golden rules, where
utility discounting was incorporated into the rule.
24
Another set of studies followed Peter
Diamonds (1965) influential analysis of an overlapping-generations economy with a
neoclassical production structure, where Phelpss setting was augmented with a fully specified
population structure. Here, each generation is endowed with a utility function over consumption
streams but has no altruistic motives; consumers then maximize utility using perfectly
competitive loan markets. Diamond showed that competitive equilibria could indeed deliver
dynamic inefficiency. Phelps and Riley (1978) also considered an overlapping-generation
structure and examined capital accumulation paths that are not restricted to constant savings
rates. The intergenerational equity consideration is formalized by adopting a Rawlsian social

23
It is, for example, straightforward to show that under logarithmic utility and a savings technology that is linear
in the amount saved, equilibrium savings in the absence of government intervention are too low.
24
Abstracting from technology and population growth, a steady state resulting from intertemporal utility
maximization is characterized by a savings rate that is constant and that modifies Phelpss golden rule r = to r
= + , where is the rate at which dynasties discount their utility flows over time. Thus, given that r=f(k) is
decreasing in k, less capital accumulation is called for due to discounting.
22
welfare function, where the objective is to maximize the well-being of the generation which is
least well off.
25
The study argues that golden-rule outcomes are not generally optimal but that a
path with net capital accumulation actually can be, even though any capital accumulation moves
resources away from the initial generation.

Phelps also initiated the analysis of capital accumulation under uncertainty (Phelps, 1962).
Uncertainty, especially about the return from capital accumulation, is an important real-world
feature. Phelps examined a Ramsey-style model with an individual consumer who faces
stochastic returns to saving. Phelps used his model to gain insights into why savings rates seem
to differ among consumers with different sources of income.

Accumulation in the form of R&D and human capital
In another early contribution, Phelps examined endogenous research and development (Phelps,
1966b). Again, the perspective was not on positive analysis but on determining optimal research
investment rates. Phelps examined a neoclassical model with capital, labor, and the level of
technology as variable production factors. Within this framework, Phelps again derived, as a
special case, a golden rule of accumulation of physical capital for maximal steady-state
consumption, which is unaffected by the endogeneity of technological progress. He also derived
a golden rule of investment in research for maximal steady-state consumption. The rule
prescribes the equalization of the return to investments in technology to the rate of growth.
Phelps concluded that the golden rule could be generalized: the rates of return of all kinds of
investments should be equal to the output growth rate.
26

25
J ohn Rawlss book A Theory of Justice (1971) proposes a maximin criterion according to which the welfare
of the least well off individual in society should be maximized.
26
Inspired by the earlier work of Charles Kennedy (1964) - recently revisited in a number of studies by Daron
Acemoglu (see, e.g., Acemoglu, 2002) Drandakis and Phelps (1966) made another contribution to the analysis
of optimal research efforts by studying the choice of the direction of research: whether it should be labor-
saving or capital-saving.
23

The endogenous-growth literature that later emerged differed from Phelpss work primarily in
that it had a positive angle: it developed decentralized models of endogenous technological
change aimed at understanding growth in a broad cross section of countries.
27
Parts of this
literature (in particular, Aghion and Howitt, 1999) build directly on a theoretical model of human
capital and the diffusion of technology by Richard Nelson and Phelps (1966). The Nelson-
Phelps analysis focuses on the stock as opposed to the accumulation of human capital as a
key factor behind technology growth in the sense that an educated and knowledgeable workforce
is better able to adopt available new technology. In the empirical growth literature, the Nelson-
Phelps setting has provided a means of formalizing technological catch-up across countries,
whereby technologically less advanced countries adopt the technologies of the more advanced
ones, and do so more efficiently, the more educated the workforce in the adopting country.
28
The
analysis thus explains why data indicate that output growth is more related to the stock of human
capital than to its rate of growth. The model also helps explain why skill premia the higher
wage rates enjoyed by skilled, or educated, workers tend to be higher in times of rapid
technological change: an educated workforce is able to assimilate technological advances more
rapidly. Such reasoning has been used to interpret the recent increase in returns to education that
has taken place in many countries, in particular the US.
29
To the extent that there are important
spillovers (externalities) in the adoption of technologies, the returns to education might not be
fully reflected in the skill premium. Thus, Nelson and Phelps argued, their model suggests a
possible reason for subsidizing education.

27
Another difference is that most of the endogenous-growth literature considers cases where the long-run rate of
technology growth is endogenous. In most of Phelpss work, in contrast, since researcher-workers are assumed
to be essential for generating technical change, the rate of population growth, which is exogenous, ultimately
pins down the long-run growth rate of technology.
28
See, e.g., Benhabib and Spiegel (1994) and Barro and Sala-i-Martin (2004).
29
For a discussion of these issues, see, e.g., Hornstein et al. (2005).
24
Phelps (1966b) assumes that the long-run rate of growth of technology is ultimately pinned down
by the growth rate of the population. Thus, the maintained view, elaborated on in Phelps
(1968b), is that the accumulation of ideas fundamentally benefits from population growth. The
logic is simple: people are required in order for new ideas to be developed, but once a productive
and valuable idea is born, it can be transferred to others essentially at no cost. Thus, the amount
of new ideas per capita will grow with population size.
30
Recent contributions to growth theory
by Charles J ones (see, e.g., J ones, 1995) revisit Phelpss arguments that population growth is
central for long-run technology growth.

Phelpss other contributions
Although Edmund Phelps is best known for his work in macroeconomics, his contributions to
labor economics and public finance also deserve mention. Among other contributions he initiated
the literature on statistical discrimination, derived new results regarding the structure of optimal
income taxation, and examined the properties of an optimal inflation tax.

Phelpss ideas concerning statistical discrimination were outlined in his monograph (1972a) and
formalized in an article (1972b). Around the same time, Kenneth Arrow published equally
influential papers on statistical discrimination (Arrow, 1972a, 1972b, 1973). These studies by
Phelps and Arrow both commonly referred to as seminal contributions to the theory of
statistical discrimination emphasize that unequal treatment of equally productive workers can
arise when employers have imperfect information about individual worker characteristics. When
individual productivity is measured with error, it may be worthwhile to use group data
information on average productivities in the group to which an individual belongs so as to

30
Phelps explains the basic intuition with the following example (Phelps 1968b, pp. 511-512):
One can hardly imagine, I think, how poor we would be today were it not for the rapid population growth of
the past to which we owe the enormous number of technological advances enjoyed today If I could re-do
the history of the world, halving population size each year from the beginning of time on some random basis,
I would not do it for the fear of losing Mozart in the process.
25
improve predictions about an individual workers productivity. A consequence of such behavior
is that individuals with the same characteristics may be treated differently.

Phelpss most noteworthy contribution to public finance deals with optimal taxation of wage
income (Phelps, 1973a). Following Mirrlees (1971), Phelps assumed that whereas workers have
different wage rates per hour, the policymaker can only observe wage incomes, not wage rates
(or hours worked). A striking result of this analysis is that the marginal tax rate should approach
zero at the very top of the income distribution (even though the average tax rate may well be
high at the top). A similar finding was later obtained by Sadka (1976) and it is often referred to
as the Phelps-Sadka result.

The idea that inflation can be viewed as a tax on the holders of nominal assets has a long history
in economic doctrine. Phelps (1973b) notes that, from a public finance perspective, inflation is a
source of tax revenue for the government, and hence it should be chosen optimally in
conjunction with other taxes. He thus argued in favor of a positive (but modest) inflation rate in
order to balance the distortions from different taxes against each other.

Summary
Edmund Phelpss contributions have had a decisive impact on macroeconomic research. They
have also fundamentally influenced the practice of macroeconomic policymaking. The focus
on the intertemporal nature of macroeconomic policy tradeoffs provides a unifying theme for
his work on inflation and unemployment as well as on capital accumulation.

Toward the end of the 1960s, Phelps launched a research program with the objective of
rebuilding Keynesian macroeconomic theory based on explicit modeling of the behavior of
26
individual firms and households in a world of imperfect information and market frictions.
Phelps developed the first models of the expectations-augmented Phillips curve and
equilibrium unemployment. These models fundamentally changed the analysis of the
relationship between inflation and unemployment. They also provided a completely new
setting for the analysis of monetary (and fiscal) demand management policy. The resulting
insights changed the views, among researchers as well as policymakers, as to what such
policies could achieve and how they should achieve it.

Phelps contributions to the analysis of optimal capital accumulation demonstrated conditions
under which all generations would benefit from changes in the aggregate savings rate. He also
pioneered research on the role of human capital for technology and output growth. His
composite contributions in this area have had a profound impact on subsequent research and
have improved our understanding of the process of economic growth.
27
References

Acemoglu, D. (2002), Directed Technical Change, Review of Economic Studies, Vol. 69, 781-
810.

Aghion, P. and P. Howitt (1999), Endogenous Growth Theory, Cambridge, MA: MIT Press.

Akerlof, G., W. Dickens and G. Perry (1996), The Macroeconomics of Low Inflation,
Brookings Papers on Economic Activity, No. 1, 1-76.

Allais, M. (1947), Economie et intrt, Paris: Imprimerie National.

Allais, M. (1962), The Influence of the Capital-Output Ratio on Real National Income,
Econometrica, Vol.30, 700-728.

Arrow, K. J . (1972a), Models of J ob Discrimination, in A. H. Pascal (ed.), Racial
Discrimination in Economic Life, Lexington, MA: D. C. Heath.

Arrow, K. J . (1972b), Some Mathematical Models of Race Discrimination in the Labor
Market, in A. H. Pascal (ed.), Racial Discrimination in Economic Life, Lexington, MA: D. C.
Heath.

Arrow, K. J . (1973), The Theory of Discrimination, in O. Ashenfelter and A. Rees (eds.),
Discrimination in Labor Markets, Princeton, NJ : Princeton University Press.

Barro, R. J . and X. Sala-i-Martin (2004), Economic Growth, 2
nd
Edition, Cambridge, MA:
MIT Press.

Benhabib, J . and M. Spiegel (1994), The Role of Human Capital in Economic Development:
Evidence from Aggregate Cross Country Data, Journal of Monetary Economics, Vol. 34, 143-
173.

Blanchard, O. (2006), European Unemployment: The Evolution of Facts and Ideas, Economic
Policy, Vol.21 (J anuary), 5-59.

Calmfors, L. and B. Holmlund (2000), Unemployment and Economic Growth: A Partial
Survey, Swedish Economic Policy Review, Vol. 7(1), 107-153.

Calvo, G. (1983), Staggered Prices in a Utility-Maximizing Framework. Journal of Monetary
Economics, Vol. 12, 383-398.

Cass, D. (1965), Optimum Growth in an Aggregative Model of Capital Accumulation, Review
of Economic Studies, Vol. 32, 233-240.

Cass, D. (1972), On Capital Overaccumulation in the Aggregative, Neoclassical Model of
Economic Growth: A Complete Characterization, Journal of Economic Theory, Vol. 4, 200-
223.

28
Clarida, R., J . Gal and M. Gertler (1999), The Science of Monetary Policy: A New
Keynesian Perspective, Journal of Economic Literature, Vol. 32, 1661-1707.

Desrousseaux, J . (1961), Expansion stable et taux dintrt optimal, Annales des Mines,
November, 31-46.

Diamond, P. (1965), National Debt in a Neoclassical Growth Model, American Economic
Review, Vol. 55, 1126-1150.

Diamond, P. (1984), A Search-Equilibrium Approach to the Micro Foundations of
Macroeconomics, Cambridge, MA: MIT Press.

Drandakis, E. M. and E. S. Phelps (1966), A Model of Induced Invention, Growth and
Distribution, Economic Journal, Vol. 76, 823-840.

Evans, G. and S. Honkapohja (2001), Learning and Expectations in Macroeconomics,
Princeton, NJ : Princeton University Press.

Fischer, S. (1977), Long-Term Contracts, Rational Expectations and the Optimal Money
Supply Rule, Journal of Political Economy, Vol. 85, 191-205.

Friedman, M. (1968), The Role of Monetary Policy, American Economic Review, Vol. 58, 1-
17.

Friedman, M. (1975), Unemployment versus Inflation, Institute for Economic Affairs,
London.

Friedman, M. (1977), Nobel Lecture: Inflation and Unemployment, Journal of Political
Economy, Vol. 85, 451-472.

Friedman, M. (1987), Quantity Theory of Money, in J . Eatwell, M. Millgate and P. Newman
(eds.), The New Palgrave: A Dictionary of Economics, London: McMillan.

Frydman, R. and E. S. Phelps (1983), (eds.), Individual Forecasting and Aggregate
Outcomes:'Rational Expectations' Examined, Cambridge: Cambridge University Press.

Grossman, G. and E. Helpman (1991), Innovation and Growth in the Global Economy,

Hornstein, A., P. Krusell and G. Violante (2005), The Effects of Technical Change on Labor
Market Inequalities, in P. Aghion and S. Durlauf (eds.), Handbook of Economic Growth,
Chapter 20, Amsterdam: Elsevier.

J ones, C. (1995), R&D-Based Models of Economic Growth, Journal of Political Economy,
Vol. 103, 759-784.

Kennedy, C. (1964), Induced Bias in Innovation and the Theory of Distribution, Economic
Journal, Vol. 84, 541-547.

29
Koopmans, T. (1965), On the Concept of Optimal Economic Growth, in The Econometric
Approach to Development Planning. Amsterdam: North-Holland.

Laibson, D. (1997), Golden Eggs and Hyperbolic Discounting, Quarterly Journal of

Laidler, D. (2006), Three Lectures on Monetary Theory and Policy: Speaking Notes and
Background Papers, Working Paper 128, sterreichische Nationalbank, Vienna.

Layard R., S. Nickell and R. J ackman (1991), Unemployment: Macroeconomic Performance
and the Labour Market, Oxford: Oxford University Press.

Lerner, A. (1949), The Inflationary Process Some Theoretical Aspects, Review of
Economics and Statistics, Vol. 31, 193-200.

Lipsey, R. G. (1960), The Relation between Unemployment and the Rate of Change of
Money Wage Rates in the United Kingdom, 1862-1957: A Further Analysis, Economica, Vol.
27, 1-31.

Lucas, R. E., J r. (1972), Expectations and the Neutrality of Money, Journal of Economic
Theory, Vol. 4, 103-124.

Lucas, R. E., J r. (1973), Some International Evidence on Output-Inflation Tradeoffs,
American Economic Review, Vol. 63, 326-334.

Lucas, R. E., J r. (1976), Econometric Policy Evaluation: A Critique, in K. Brunner and A. H.
Meltzer (eds.), The Phillips Curve and Labor Markets, Carnegie-Rochester Conference Series
on Public Policy, Amsterdam: North Holland, 19-46.

Lucas, R. E., J r. (1988), On the Mechanics of Economic Development, Journal of Monetary

Malinvaud, E. (1953), Capital Accumulation and Efficient Allocation of Resources.
Econometrica, Vol. 21, 233-273.

Mankiw, G. and R. Reis (2003), Sticky Information: A Model of Monetary Nonneutrality and
Structural Slumps, in P. Aghion, R. Frydman, J . Stiglitz and M. Woodford (eds.), Knowledge,
Information, and Expectations in Macroeconomics: In Honor of Edmund S. Phelps, Princeton,
NJ : Princeton University Press.

Mirrlees, J . (1971), An Exploration in the Theory of Optimal Income Taxation, Review of
Economic Studies, Vol. 38, 175-208.

Mortensen, D. (1982a), The Matching Process as a Noncooperative Bargaining Game, in J .
McCall (ed), The Economics of Information and Uncertainty, Chicago, IL: University of
Chicago Press.

Mortensen, D. (1982b), Property Rights and Efficiency in Mating, Racing and Related
Games, American Economic Review, Vol. 72, 968-979.

30
Mortensen, D. and C. Pissarides (1994), J ob Creation and J ob Destruction in the Theory of
Unemployment, Review of Economic Studies, Vol. 61, 397-415.

Muth, J . (1961), Rational Expectations and the Theory of Price Movements, Econometrica.
Vol. 29, 315-335.

Nelson, R. and E. S. Phelps. (1966), Investments in Humans, Technological Diffusion and
Economic Growth, American Economic Review: Papers and Proceedings, Vol. 56, 69-75.

Orphanides, A. and J . C. Williams (2005), The Decline of Activist Stabilization Policy:
Natural Rate Misperceptions, Learning, and Expectations, Journal of Economic Dynamics
and Control, Vol. 29, 1927-1950.

Phelps, E. S. (1961), The Golden Rule of Accumulation: A Fable for Growthmen, American
Economic Review Vol. 51, 638-643.

Phelps, E. S. (1962), The Accumulation of Risky Capital: A Sequential Utility Analysis,
Econometrica, Vol. 30, 729-743.

Phelps, E. S. (1965), Second Essay on the Golden Rule of Accumulation, American Economic
Review Vol. 55, 793-814.

Phelps, E. S. (1966a), Golden Rules of Economic Growth, New York: W.W. Norton.

Phelps, E. S. (1966b), Models of Technical Progress and the Golden Rule of Research,
Review of Economic Studies, Vol. 33, 133-145.

Phelps, E. S. (1967), Phillips Curves, Expectations of Inflation and Optimal Unemployment
over Time, Economica, Vol. 34, 254-281.

Phelps, E. S. (1968a), Money-Wage Dynamics and Labor-Market Equilibrium, Journal of
Political Economy, Vol. 76, 678-711.

Phelps, E. S. (1968b), Population Increase, Canadian Journal of Economics, Vol. 35, 497-
518.

Phelps, E. S. (1969), The New Microeconomics in Inflation and Employment Theory,
American Economic Review: Papers and Proceedings, Vol. 59, 147-160.

Phelps, E. S. et al. (1970a), Microeconomic Foundations of Employment and Inflation Theory,
New York: W.W. Norton.

Phelps, E. S. (1970b), Money-Wage Dynamics and Labor-Market Equilibrium, in E. S Phelps
et al., Microeconomic Foundations of Employment and Inflation Theory, New York: Norton.

Phelps, E. S. (1972a), Inflation Policy and Unemployment Theory, New York: Norton.

Phelps, E. S. (1972b), The Statistical Theory of Racism and Sexism, American Economic
Review, Vol. 62, 659-661.

31
Phelps, E. S. (1973a), Taxation of Wage Income for Economic J ustice, Quarterly Journal of

Phelps, E. S. (1973b), Inflation in the Theory of Public Finance, Swedish Journal of

Phelps, E. S. (1978a), Disinflation without Recession: Adaptive Guideposts and Monetary
Policy, Weltwirtschaftliches Archiv, Vol. 100, 239-265.

Phelps, E. S. (1978b), Inflation Planning Reconsidered, Economica, Vol. 45, 109-123.

Phelps, E. S. (1983), The Trouble with Rational Expectations and the Problem of Inflation
Stabilization, in R. Frydman and E. S. Phelps (eds.), Individual Forecasting and Aggregate
Outcomes:'Rational Expectations' Examined, Cambridge: Cambridge University Press.

Phelps, E. S. (1994), Structural Slumps: The Modern Equilibrium Theory of Employment,
Interest and Assets, Cambridge, MA: Harvard University Press.

Phelps, E. S., and R. A. Pollak (1968), On Second-Best National Saving and Game-
Equilibrium Growth, Review of Economic Studies, Vol. 35, 185-199.

Phelps, E. S., and J . Riley (1978), Rawlsian Growth: Dynamic Programming of Capital
Wealth for Intergeneration 'Maximin' J ustice, Review of Economic Studies, Vol. 45, 103-120.

Phelps, E. S. and J . B. Taylor (1977), Stabilizing Powers of Monetary Policy under Rational
Expectations, Journal of Political Economy, Vol. 85, 163-190.

Phelps, E. S. and S. G. Winter (1970), Optimal Price Policy under Atomistic Competition, in
E.S. Phelps, et al., Microeconomic Foundations of Employment and Inflation Theory, New
York: Norton.

Phillips, A. W. (1958), The Relation between Unemployment and the Rate of Change of
Money Wage Rates in the United Kingdom 1861-1957, Economica, Vol. 25, 283-289.

Pissarides, C. (2000), Equilibrium Unemployment Theory, Cambridge, MA: MIT Press.

Pollak, R. A. (1968), Consistent Planning, Review of Economic Studies, Vol.35, 201-208.

Ramsey, F. P. (1928), A Mathematical Theory of Saving, Economic Journal, Vol. 38, 543-
559.

Rawls, J . (1971), A Theory of Justice, Cambridge, MA: Harvard University Press.

Robinson, J . (1962), A Neoclassical Theorem, Review of Economic Studies, Vol. 29, 219-226.

Romer, P. (1986), Increasing Returns and Long-Run Growth, Journal of Political Economy,
Vol. 94, 1002-1037.

32
Rotemberg, J . J . and M. Woodford (1999), The Cyclical Behavior of Prices and Costs, in J .
Taylor and M. Woodford (eds.), Handbook of Macroeconomics, Vol. 1B, Amsterdam: North-
Holland.

Sadka, E. (1976), On Income Distribution, Incentive Effects and Optimal Income Taxation,
Review of Economic Studies, Vol. 43, 261-268.

Salop, S. (1979), A Model of the Natural Rate of Unemployment, American Economic
Review, Vol. 69, 117-125.

Sargent, T. J . (1999), The Conquest of American Inflation, Princeton, NJ : Princeton
University Press.

Sargent, T. J . and Wallace, N. (1975). Rational Expectations. The Optimal Monetary
Instrument, and the Optimal Money Supply Rule, Journal of Political Economy, Vol. 83, 241-
54.

Shapiro, C. and Stiglitz, J . (1984), Equilibrium Unemployment as a Worker Discipline
Device, American Economic Review, Vol. 74, 433-444.

Solow, R. M. (1956), A Contribution to the Theory of Economic Growth, Quarterly Journal
of Economics, Vol. 70, 65-94.

Solow, R. M. (1957), Technical Change and the Aggregate Production Function, Review of
Economics and Statistics, Vol. 39, 312-320.

Stiglitz, J . (1984), Price Rigidities and Market Structure, American Economic Review, Vol.
74, 350-355.

Strotz, R. (1956), Myopia and Inconsistency in Dynamic Utility Maximization, Review of
Economic Studies, Vol. 23, 165-180.

Swan, T. W. (1956), Economic Growth and Capital Accumulation, Economic Record, Vol.
32, 334-361.

Swan, T. W. (1964), Of Golden Ages and Production Functions, in K. Berrill (ed.), Economic
Development with Special Reference to East Asia, New York: MacMillan and St Martins
Press, 3-16.

Taylor, J . B. (1979), Staggered Wage Setting in a Macro Model, American Economic Review,
Papers and Proceedings, Vol. 69, 108-113.

von Neumann, J . (1945-46), A Model of General Economic Equilibrium, Review of Economic
Studies, Vol. 13, 1-9.

von Weizscker, C. C. (1962), Wachstum, Zins und Optimale Investissionsquote, Basel.

33
Woodford, M. (2003a), Imperfect Common Knowledge and the Effects of Monetary Policy,
in P. Aghion, R. Frydman, J . Stiglitz and M. Woodford (eds.), Knowledge, Information, and
Expectations in Macroeconomics: In Honor of Edmund S. Phelps, Princeton, NJ : Princeton
University Press.

Woodford, M. (2003b), Interest and Prices: Foundations of a Theory of Monetary Policy,
Princeton,NJ : Princeton University Press.
Advanced information on the Bank of Sweden Prize in Economic Sciences in Memory of Alfred Nobel
10 October 2005
Robert Aumanns and Thomas Schellings
Contributions to Game Theory:
Analyses of Conflict and Cooperation
1. Introduction
Wars and other conicts are among the main sources of human misery. A minimum
of cooperation is a prerequisite for a prosperous society. Life in an anarchic state
of nature with its struggle of every man against every man is, in Thomas Hobbes
(1651) famous phrase, solitary, poor, nasty, brutish, and short.
Social scientists have long attempted to understand the fundamental causes of
conict and cooperation. The advent of game theory in the middle of the twentieth
century led to major new insights and enabled researchers to analyze the subject
with mathematical rigor. The foundations of game theory were laid out in the classic
book by John von Neumann and Oscar Morgenstern, The Theory of Games and Eco-
nomic Behavior, published in 1944. The 1994 economics laureates John Harsanyi,
John Nash and Reinhard Selten added solution concepts and insights that substan-
tially enhanced the usefulness and predictive power of non-cooperative game theory.
The most central solution concept is that of Nash equilibrium. A strategy combi-
nation (one strategy for each player) constitutes a Nash equilibrium if each players
strategy is optimal against the other players strategies.
1
Harsanyi showed that this
solution concept could be generalized to games of incomplete information (that is,
where players do not know each others preferences). Selten demonstrated that it
could be rened for dynamic games and for games where players make mistakes with
(innitesimally) small probabilities. Nevertheless, the great intellectual achievements
of these researchers would have been to little avail, had game-theoretic tools not been
applied to address salient questions about society.
The work of two researchers, Robert J. Aumann and Thomas C. Schelling, was
essential in developing non-cooperative game theory further and bringing it to bear
1
A non-cooperative game in normal form consists of a list of players, a set of strategies available
to each player, and a function that species the payo consequences to all players of each strategy
combination.
1
2
on major questions in the social sciences.
2
Approaching the subject from dierent
anglesAumann from mathematics and Schelling from economicsthey both per-
ceived that the game-theoretic perspective had the potential to reshape the analysis
of human interaction. Perhaps most importantly, Schelling showed that many famil-
iar social interactions could be viewed as non-cooperative games that involve both
common and conicting interests, and Aumann demonstrated that long-run social
interaction could be comprehensively analyzed using formal non-cooperative game
theory.
Although their writings on conict and cooperation were well received when they
appeared in the late 1950s, it took a long time before Aumanns and Schellings visions
came to be fully realized. The delay reects both the originality of their contributions
and the steepness of the subsequent steps. Eventually, and especially over the last
twenty-ve years, game theory has become a universally accepted tool and language in
economics and in many areas of the other social sciences. Current economic analysis of
conict and cooperation builds almost uniformly on the foundations laid by Aumann
and Schelling.
2. Schelling
Thomas Schellings book The Strategy of Conict (1960) launched his vision of game
theory as a unifying framework for the social sciences. Turning attention away from
zero-sum games, such as chess, where players have diametrically opposed interests, he
emphasized the fact that almost all multi-person decision problems contain a mixture
of conicting and common interests, and that the interplay between the two concerns
could be eectively analyzed by means of non-cooperative game theory. The stage
had been set by Nash (1950a,1951), who had proven that there exist (Nash) equilibria
in all games with nitely many pure strategies. Schelling took on the complementary
task of deducing the equilibria for interesting classes of games and evaluating whether
these games and their equilibria were instructive regarding actual economic and social
interaction. He did this against the background of the worlds rst nuclear arms race
and came to contribute greatly to our understanding of its implications.
2
While cooperative game theory starts out from a set of potential binding agreements and play-
ers preferences over them, non-cooperative game theory starts out from players strategy sets and
preferences over the associated outcomes.
3
2.1. Conict, commitment and coordination. Schellings earliest major con-
tribution is his analysis of behavior in bilateral bargaining situations, rst published
as an article (Schelling, 1956) and later reprinted as Chapter 2 of Schelling (1960).
Here, bargaining is interpreted broadly: besides explicit negotiationsfor example
between two countries or between a seller and a buyerthere is also bargaining
when two trucks loaded with dynamite meet on a road wide enough for one, to cite
one of Schellings characteristically graphical examples.
Bargaining always entails some conict of interest in that each party usually seeks
an agreement that is as favorable as possible. Yet, any agreement is better for both
parties than no agreement at all. Each player has to balance the quest for a large
share of the pie against the concern for agreement. When Schelling wrote his article,
economists work on bargaining had typically taken a cooperative or normative ap-
proach, by asking questions such as: what is a fair outcome? An exception was Nash,
who modeled bargaining both with a cooperative (Nash 1950b) and a non-cooperative
(Nash, 1953) approach. While Nashs formulations allow elegant mathematical analy-
ses by way of abstracting from many realistic bargaining tactics, Schelling examines
the bargaining tactics a player can use in order to tilt the outcome in his or her favor
emphasizing in particular that it may be advantageous to worsen ones own options
in order to elicit concessions from the opponent. It can be wise for a general to burn
bridges behind his troops as a credible commitment towards the enemy not to retreat.
Similarly, the owners of a rm may protably appoint a manager with limited powers
to negotiate, and a politician may gain from making public promises that would be
embarrassing to break. Such tactics work if the commitment is irreversible or can
only be undone at great cost, while commitments that are cheap to reverse will not
elicit large concessions. However, if both parties make irreversible and incompatible
commitments, harmful disagreement may follow.
Let us illustrate some of the key issues by means of a stylized and simple example.
Suppose that two countries disagree over the right to a patch of territory.
3
Each
country can choose to mobilize military force or refrain from doing so. If both mobilize
there is a high probability of war, while the probability of a peaceful agreement about
3
For more elaborate game-theoretic analyses of commitment in bargaining, see for example Craw-
ford (1982), Muthoo (1996), and G uth, Ritzberger and van Damme (2004).
4
division of the territory is low. Let the expected payo to each country be zero if both
mobilize. If instead both countries refrain from mobilization, a peaceful agreement
about division of the territory has a high probability, while the probability of war
is small. In this case, each country obtains a positive expected payo /. However,
if only one country mobilizes, it can take complete control of the territory without
war, and neither the other country nor any other party can force a military retreat
by the occupant. The aggressor obtains payo c while the losers payo is c, where
c / c 0, war thus being the worst outcome.
4
This simple mobilization game
can be described by the following payo bi-matrix, where one player (here country)
chooses a row and the other simultaneously chooses a column, with the row players
payo listed rst in each entry:
Mobilize Refrain
Mobilize 0. 0 c. c
Refrain c. c /. /
TABLE 1
This game belongs to a class of games known as Chicken, sometimes called Hawk-
Dove. Such games have three Nash equilibria: two pure and one mixed. The pure
equilibria entail mobilization by exactly one country; if one country expects the other
to mobilize, then it is optimal to refrain from mobilization. The mixed equilibrium
entails randomized mobilization by each country and thus a positive probability of
war.
The pure equilibria are plausible in situations where the two countries have some
means to coordinate on either equilibrium. For example, a small perturbation of the
game that would create even a tiny asymmetry in the payos, may be enough for
both players to expect mobilization by the player who has the most to gain from
it, thus rendering that equilibrium salient or focal. According to Schelling, it
is likely that humans are capable of such coordination in many situations, while a
purely formal analysis is likely to be unable to capture the principles of salience or
4
In some conict situations, war is less undesirable than the humiliation that may be associated
with lack of mobilization when the other country mobilizes. In such situations, a / 0 c, and
the game becomes a prisoners dilemma, with mutual mobilization as the outcome, see section 3.1.
5
focality in the game in question: One cannot, without empirical evidence, deduce
what understandings can be perceived in a non-zero sum game of maneuver any more
than one can prove, by purely formal deduction, that a particular joke is bound to be
funny is a famous quote by Schelling (1960, p.164). Instead, equilibrium selection
is an area where experimental psychology can contribute to game theory (ibid.
p.113).
5
Absent any commonly understood coordination principle, the games mixed equi-
librium appears more plausible. Each country is then uncertain about the others
move, assigning some probability j to the event that the other country will mobilize.
The Nash equilibrium probability of mobilization is j = (c/),(c/ +c), rendering
each country indierent whether to mobilize.
6
It follows that, for plausible parameter
values, the probability of war is decreasing in the losers payo c; the key to minimiz-
ing the risk of war is not only to contain the winners gain but equally importantly
to improve the losers payo.
7
Mobilizing and threatening to mobilize are not equivalent. A formal analysis of
deterrence is complicated and requires specifying a dynamic game with several stages,
but with Schellings intuition as a guide it is possible to proceed without detailed
mathematics. The study of credible deterrence through so-called second-strike
strategies takes up a major part of The Strategy of Conict. Schelling emphasizes
that investments in deterrence can become dangerous in case of false warnings as well
as when misjudging the adversarys interests and intentions.
Suppose that Country 1 can pre-commit to mobilize if Country 2 mobilizes. More
precisely: rst Country 1 chooses whether to refrain from mobilization altogether
or to commit to mobilize if and only if Country 2 mobilizes. Thereafter, Country 2
observes 1s move and decides whether or not to mobilize. If payos are as described in
5
By now, there is a sizeable experimental literature on focal points in bargaining as well as in
other games, much of it informed and inspired by Schelling. We return to the coordination problem
below.
6
Mobilizing yields expected payo (1 j)a, while refraining yields jc + (1 j)/. Equating the
two determines the equilibrium probability j.
7
By statistical independence between the two players randomization, the probability for war is
= j
2
j + (1 j)
2
i, where j is the probability for war when both countries mobilize and i < j
the probability for war when no country mobilizes. It follows that is increasing in j for all
j i, (j + i). Hence is decreasing in c if c < (a /) j,i.
6
Table 1, the (subgame perfect) equilibrium outcome will be that Country 1 makes the
mobilization commitment, and both countries refrain from mobilization. Indeed, it is
sucient that Country one commits to mobilize with a suciently high probability.
8
Such deterrence thus guarantees a peaceful outcomea balance of terror.
Suppose, moreover, that Country 1 is uncertain whether Country 2 actually prefers
war to the negotiated outcome. In the game-theoretic parlance (based on Harsanyis
work), Country 1 now has incomplete information about Country 2s payos. Should
Country 1 still commit to mobilize if country Country 2 mobilizes? Schellings analysis
reveals that the optimal commitment strategy is then often to choose a probability of
mobilization that is less than one. In other words, in the face of an enemys military
escalation, a country should threaten to let the situation slip out of hand rather
than commit to certain retaliation, or in Schellings words, make threats that leave
some things to chance. The reason is that a modest probability of war may be enough
to deter the enemys mobilization.
9
Another virtue of uncertain retaliation threats is that credibility is easier to attain
the smaller is the own expected retaliation cost. In fact, Schelling suggested that a
good way to meet enemy aggression is to engage in brinkmanship gradually step-
ping up the probability of open conict. Since each step is small, credibility can be
sustained by the anger and outrage that builds steadily against an unrelenting oppo-
nent, and since the opponent can reduce the probability of conict by relenting, the
probability of conict is kept low. As Schelling observed, most children understand
brinkmanship perfectly.
The above analysis implies that countries should keep the adversary guessing
about their response to aggression, at the same time ensuring that forceful retaliation
8
Suppose Country 1 can commit to any probability [0, 1] of retaliation if Country 2 mobilizes.
If Country 2s preferences are as in Table 1, deterrence requires that / (1 )a or, equivalently,
that 1 /,a =
.
9
Let 0 be the probability that Country 1 attaches to the possibility that Country 2 prefers to
mobilize regardless of the retaliation threat. For <
, Country 2 will still mobilize for sure, so

the payo to Country 1 is then (1 ) c, a decreasing function of . For
its expected payo

is 0(1 )c + (1 0)/, again a decreasing function of . Hence, deterrence (choosing =
) is
optimal for Country 1 if and only if 0(1
)c + (1 0)/ is at least as large as the payo c from

not retaliating ( = 0), or, equivalently, if and only if 0 (1 c,/),(1 c,a).
7
is regarded as a real option. Two other insights are also quite immediate. First,
deterrence only works when retaliatory weapons can be shielded in case of an enemy
attack; war prevention thus requires invulnerable basing of weaponssuch as missile
silosrather than protection of population centers. Second, instability is dangerous.
The balance of terror is maintained only as long as retaliation is suciently probable
and harsh compared to the gains from occupation. War can be ignited by changes in
preferences as well as in technology, and successful attempts at disarmament have to
be balanced throughout.
Schellings analysis of credible commitments demonstrated that some Nash
equilibria are more plausible than others, inspiring Reinhard Seltens subgame per-
fection renement of the Nash equilibrium concept.
10
Schellings and Seltens work
on strategic commitment initiated a lively economics literature. The analyses of
strategic investment in oligopoly markets developed by, among others, Avinash Dixit
and Schellings student Michael Spence (a 2001 laureate) are leading examples of ap-
plied work on commitment that took o in the late 1970s (see, for example, Spence,
1977, and Dixit, 1980). Their analyses show that a rm operating in an imperfectly
competitive market can increase its prots by changing its cost structure, even if its
unit production cost increases as a result. For example, a rm can credibly commit
to a high volume of output by investing in an expensive plant with low marginal
costs. Even if average costs thereby go up, losses due to inecient production can be
outweighed by the gains generated by competitors less aggressive behavior.
The literature on monetary policy institutions provides another example of the
idea of strategic commitment at work. Here, the major point is that under certain
circumstances, voters and politicians are better o delegating monetary policy to de-
cision makers with other preferences than their own. Since rms and trade unions
take the expected monetary policy into account when setting prices and wages, an in-
dependent central banker can be superior to an elected politician even if the politician
10
A Nash equilibrium in an extensive-form game is subgame perfect (Selten, 1965) if it induces a
Nash equilibrium in every subgame. Since Nash equilibrium only requires optimality on the path of
play, Nash equilibria may well rely on threats or promises that will not subsequently materialize.
Subgame perfection eliminates many such equilibria, and, in later work Selten (1975) developed a
stronger renement, perfection.
8
at each point in time would act in accordance with the current public interest.
11
Sometimes conicts of interest may appear so strong as to be insoluble. The best
strategy for an individual may result in the worst outcome for a group. The short-
run gains from cheating on an agreement might by far outweigh the short-run losses.
Schelling (1956) noted that What makes many agreements enforceable is only the
recognition of future opportunities for agreement that will be eliminated if mutual
trust is not created and maintained, and whose value outweighs the momentary gain
from cheating in the present instance. (op. cit. p. 301). Thus, if the parties take
a long perspective and do in fact interact repeatedly, their common interests may be
suciently strong to sustain cooperation. In fact, Schelling went further: Even if
the future will bring no recurrence, it may be possible to create the equivalence of
continuity by dividing the bargaining issue into consecutive parts. That is, people
can structure their relationships, by extending interaction over time, in such a way
as to reduce the incentive to behave opportunistically at each point in time.
When Schelling rst made these observations and conjectures, game theory had
not advanced far enough to allow him to articulate them precisely, far less prove
them. Gradually, however, the literature on repeated games and Folk Theorems
(discussed below) demonstrated how present cooperation can be credibly sustained
by the threat of conict in similar situations in the future. As for Schellings assertion
that it is sometimes possible to sustain agreement by decomposing one large coop-
erative action into several small ones, it took the profession more than forty years
to fully develop the formal argument. Lockwood and Thomas (2002) demonstrate in
a two-player model that private provision of public goods can often be substantially
higher if the parties can take turns contributing than if they can only make one round
of contributions each.
12
By gradually increasing their contribution, implicitly threat-
ening to stop the increase if the other does so, each party holds out a carrot to the
other. However, fully ecient contribution levels are only attainable under strong
additional assumptions, such as zero discounting (Gale, 2001) or non-smooth payo
11
Last years economics prize was awarded to Finn Kydland and Edward Prescott in part for
having identied and analyzed problems of commitment in economic policy-making. The macroeco-
nomic literature on delegation is largely inspired by their work, see e.g. Rogo (1985).
12
Admati and Perry (1991) were the rst to tackle the problem head on, but their analysis
considered a fairly special environment and only yielded weak support for Schellings conjecture.
9
functions (Marx and Matthews, 2000). These analyses can potentially explain why
progress is necessarily gradual in many areas where cooperative actions are costly
to reverse. Examples include military disarmament, environmental cooperation, and
shrinkage of production capacity in a declining market.
Gradual cooperation occurs not only among humans. The biologist John Maynard
Smith describes the mating behavior of the black hamlet, a hermaphrodite coral reef
sh which carries both sperm and eggs simultaneously (Maynard Smith, 1982, pages
159-160). When the sh mates, it engages in several rounds of egg trading where it
alternately lays eggs and fertilizes the eggs of its partner. The proposed explanation
is that it is cheaper to produce sperm than eggs, so if all eggs were being laid at
once, the sh playing the male role in this rst round might not produce any eggs
thereafter, preferring instead to play the male role again with another sh willing
to produce eggs. By saving some eggs to be used as a reward for the other shs
eggs, each sh lowers the partners incentive to defect. This is but one example from
evolutionary biology where Schellings analysis has relevance.
Schelling also studied a class of social interactions that involve little or no conict
of interest, so-called pure coordination games. These are games where all players
prefer coordination on some joint course of action and no player cares about which
coordinated course of action is taken. For example, it may not matter to a team
of workers who carries out which task, as long as the team gets its job done. In
this case, coordination may be easy if players can communicate with each other but
appears dicult without communication. By experimenting with his students and
colleagues, Schelling discovered that they were often able to coordinate rather well
without communicating even in unfamiliar games that had an abundance of Nash
equilibria. As an example, consider the game where two people are asked to select a
positive integer each. If they choose the same integer both get an award, otherwise
no award is given. In such a setting, the majority tends to select the number 1. This
number is distinctive, since it is the smallest positive integer. Likewise, in many other
settings, Schellings experimental subjects were able to utilize contextual details, joint
references, and empathy in order to identify focal equilibria.
13
It seems likely that
13
Subsequent attempts to discover fundamental coordination principles include Mehta, Starmer
and Sugden (1994a,b). Camerer (2003, Chapter 7), gives an overview of coordination experiments.
10
many social conventions and organizational arrangements have emerged because they
facilitate coordination. Inspired by Schellings analysis of coordination in common
interest games, the philosopher David Lewis specied the compelling hypothesis that
language itself has emerged as a convention (Lewis, 1969).
A nal interesting class of social decision problems are interactions in which par-
ticipants are mutually distrustful. For example, two generals may both agree that
war is undesirable, and will hence prepare for peace as long as they both think that
the other will do likewise. Yet, if one general suspects that the other is preparing
for war, then his best response may be to prepare for war as wellwhen war is less
undesirable than being occupied.
14
As Schelling (1966, page 261) notes, this idea had
already been clearly formulated by Xenophon (in the fourth century B.C.). A more
recent version of the argument is due to Wohlstetter (1959), who in turn inspired
Schelling. The analysis was advanced by Schelling (1960, Chapter 9), who expressed
it in game-theoretic terms and considered explicitly the role of uncertainty in trig-
gering aggression. To illustrate the possibility that war is caused solely by mutual
distrust, consider the following payo bi-matrix (the rst number in each entry being
the payo to the row player):
War Peace
War 2. 2 3. 0
Peace 0. 3 4. 4
TABLE 2
Each player has the choice between going to war and behaving peacefully. The two
pure-strategy Nash equilibria are (War, War) and (Peace, Peace). If players are ra-
tional, carry out their plans perfectly, and have no uncertainty about the opponents
payo, Schelling (1960, p.210) thought that peace would be the most plausible out-
come of such a game (a position that is not shared by all game theorists). However,
Schelling (1960, p.207) also contended that a small amount of nervousness about the
opponents intentions could be contagious enough to make the peaceful equilibrium
crumble: If I go downstairs to investigate a noise at night, with a gun in my hand,
For recent theoretical work, see Binmore and Samuelson (2005).
14
This was not the case in the previous example, where war was the worst outome.
11
and nd myself face to face with a burglar who has a gun in his hand, there is a
danger of an outcome that neither of us desires. Even if he prefers just to leave
quietly, and I wish him to, there is danger that he may think I want to shoot, and
shoot rst. Worse, there is danger that he may think that I think he wants to shoot.
Schelling did attempt a formal analysis of this surprise attack dilemma, but since
game theory at that time lacked a proper framework for studying games with incom-
plete information, it is fair to say that his modeling did less than full justice to his
intuition.
15
The Strategy of Conict has had a lasting inuence on the economics profession
as well as on other social sciences. It has inspired, among other things, the detailed
analysis of bargaining in historical crisis situations (see e.g. Snyder and Diesing,
1977). The book and its sequels Strategy and Arms Control (1961, coauthored with
Morton Halperin) and Arms and Inuence (1966), also had a profound impact on
military theorists and practitioners in the cold war era, played a major role in estab-
lishing strategic studies as an academic eld of study, and may well have contributed
signicantly to deterrence and disarmament among the superpowers.
16
2.2. Other contributions. Over the forty-ve years since the publication of The
Strategy of Conict, Thomas Schelling has continued to produce a series of novel and
useful ideas. We briey mention two of them here.
In a much cited article from 1971, Schelling analyzed how racially mixed societies
and neighborhoods can suddenly become segregated as the proportion of inhabitants
of one race gradually slides below a critical level. A modest preference for not form-
ing part of a minority in ones neighborhood, but not necessarily favoring dominance
of ones own race, can cause small microshocks to have drastic consequences at the
macro level. Besides providing a convincing account of an important social policy
problem, Schelling here oers an early analysis of tippingthe rapid movement
from one equilibrium to anotherin social situations involving a large number of in-
15
Recently, Baliga and Sjostrom (2004) have provided a formal analysis.
16
The secrecy surrounding military issues makes it dicult to assess the exact impact of Schellings
work on the behavior of superpowers. However, a clue is that in 1993 Schelling won the National
Academy of Sciences (U.S.) Award for Behavioral Research Relevant to the Prevention of Nuclear
War.
12
dividuals. The tipping phenomenon is pursued in several dierent contexts in another
of Schellings inuential books, Micromotives and Macrobehavior from 1978, and has
been further analyzed by other social scientists.
The next seminal set of ideas is explored in a sequence of articles on self-command,
notably Schelling (1980, 1983, 1984a, 1992).
17
Here, Schelling observes that we do
many things that we wish we would rather not do, for example smoking and drinking
too much or exercising and saving too little. He also explores the limits of self-
management and the associated challenges for public policy. Interestingly, the im-
portance of credible commitments is no smaller in this context of intrapersonal con-
icts than in the interpersonal conicts which occupied Schelling at the beginning
of his career. Over the last decade, with the rise of behavioral economics, the issue
of limited self-command has received widespread attention.
18
There are now many
papers in leading economics journals on procrastination, under-saving, and unhealthy
consumption.
In sum: the errant economist (as Schelling has called himself) turned out to be
a pre-eminent pathnder.
3. Aumann
Robert Aumann has played an essential role in shaping game theory. He has pro-
moted a unied view of the very wide domain of strategic interactions, encompassing
many apparently disparate disciplines, such as economics, political science, biology,
philosophy, computer science and statistics. Instead of using dierent constructs to
deal with various specic issuessuch as deterrence, perfect competition, oligopoly,
taxation and votingAumann has developed general methodologies and investigated
where these lead in each specic application. His research is characterized by an
unusual combination of breadth and depth. Some contributions contain involved
analysis while others are technically simple but conceptually profound. His funda-
mental works have both claried the internal logic of game-theoretic reasoning and
expanded game theorys domain of applicability.
17
The rst two of these are reprinted as Chapters 3 and 4 in Schelling (1984b).
18
For early formal analyses of such problems, see e.g. Strotz (1956) and Phelps and Pollak (1968).
13
3.1. Long-term cooperation. Among Aumanns many contributions, the study
of long-term cooperation has arguably had the most profound impact on the social
sciences. As pointed out above, a great deal of interaction is long-term in nature,
sometimes of indenite duration. Countries often have an opportunity to gain some
advantage at their neighbors expense. Competing rms may take daily or monthly
production and pricing decisions, conditioned in part on their competitors past be-
havior. Farmers may join together to manage some common resource, such as a
pasture or water source, etc. It is therefore important to study recurrent interaction
with a long horizon.
The dierence between short-term and long-term interaction is perhaps most eas-
ily illustrated by the well-known prisoners dilemma game. This is a two-person game,
where each player has two pure strategies, to cooperate (C) or defect (D). The
players choose their strategies simultaneously. Each players dominant strategy is
Dthat is, D is an optimal strategy irrespective of the others strategybut both
players gain if they both play C. When played once, the game thus admits only one
Nash equilibrium: that both players defect. However, the equilibrium outcome is
worse for both players than the strategy pair where both cooperate. An example
is given by the following payo bi-matrix, where, as before, the rst number in each
entry is the payo to the row player and the second number the payo to the column
player.
19
C 1
C 2. 2 0. 3
1 3. 0 1. 1
TABLE 3
Suppose that the same two players meet every day, playing the prisoner dilemma
over and over again, seeking to maximize the average daily payo stream over the
innite future. In this case, it can be shown that cooperation in every period is an
equilibrium outcome. The reason is that players can now threaten to punish any
19
The payos are assumed to be von Neumann-Morgenstern utilities truly capturing the motives
of the players. If the payos are monetary instead, it is perfectly possible that a rational player
would choose C out of of concern for the other players income.
14
deviation from cooperative play today by refusing to cooperate in the future. That
is, the short-term gain from defection today is more than outweighed by the reduction
in future cooperation.
In fact, Aumann (1959) proved a much more general result, concerning any su-
pergame G
that consists of the innite repetition of any given game G. Essentially,

he showed that any average payo that is feasible in the supergame and does not
violate individual rationality (see below) in the stage game G can be sustained as
a Nash equilibrium outcome in G
. Moreover, he demonstrated that the result holds

even if robustness is required with respect to joint deviations by coalitions of players.
Let us state the result more precisely. A pure strategy in G
is a decision rule that

assigns a pure strategy in G to each period and for every history of play up to that
period. The set of pure strategies in G
is thus innite and contains very complex

strategies. The main result of the paper species exactly the set of strong equilibrium
payos of G
.
20
A strong equilibrium, a solution concept due to Aumann (1959), is a
strategy prole such that no group (subset, coalition) of players can, by changing its
own strategies, obtain higher payos to all members of the group.
21
Nash equilibrium
is thus the special case in which the deviating group always consists of exactly one
player. Aumann showed that the set of strong equilibrium payos coincides with the
so-called ,-core of the game G that is being repeated. The ,-core, a version of the
core, essentially requires that no group of players can guarantee themselves higher
payoseven if the others would gang up against them.
When Aumanns result is applied to deviating groups of size one, the result is a
so-called Folk Theorem for repeated games. According to this theorem, the set of
Nash equilibrium payos of an innitely repeated game G
coincides with the set of

feasible and individually rational payos. A payo vectora list of payos, one for
each player is feasible if it is the convex combination of payo vectors that can be
obtained by means of pure strategies in G, and a payo level is individually rational
for a player if it is not less than the lowest payo in G to which the other players
can force the player down.
22
The gist of the argument is to provide strategies in
20
Aumann denes payos in G
by means of a certain limit of time averages of payos in G.

21
Not all games have such equilibria.
22
The set of individually rational payos can be dened as follows. For each (pure or mixed)
strategy combination of the other players in G, let the player in question play a (pure or mixed)
15
G
that constitute threats against deviations from strategies in G
that implement
the given payo vector.
In the prisoners dilemma considered here, the set of feasible and individually
rational payo pairs consists of all payo pairs that can be obtained as convex com-
binations of the payo pairs in Table 3 and where no payo is below 1. To see this,
rst note that each player can guarantee himself a payo of at least 1 by playing
1. Second, the four pure-strategy pairs result in payo pairs (2. 2), (1. 1), (3. 0) and
(0. 3). The set of feasible payo pairs is thus the polyhedron with these pairs as
vertices. The shaded area in Figure 1 below is the intersection of these two sets. All
these payo pairs, and no others, can be obtained as time-average payos in Nash
equilibrium of the innitely repeated play of this game.
4
0
x
y
x
y
Figure 1: The set of feasible and individually rational payo pairs in the prisoners
dilemma game in Table 3.
Applied to the game in Table 1, the Folk Theorem claims that all payo pairs
that are convex combinations of (0. 0), (c. c), (c. c) and (/. /), and where no payo is
best reply. The minimal value among the resulting payos to the latter denes the lower bound on
that players individually rational payos.
16
below c, can be obtained as time-average payos in Nash equilibrium of the innitely
repeated play of that game. In particular, the good outcome (/. /) is sustainable
despite the fact that it is not an equilibrium of the game when played once. Deviations
from prescribed play can be threatened by minmaxing the deviator, that is, the
other player randomizes between the two pure strategies in such a way as to minimize
the deviators expected payo when the latter plays his or her best reply against this
punishment. Such punishments can also sustain other outcomes as equilibria of
the innitely repeated games, for example alternating play of C and D according
to some prescribed pattern. Applied to more complex games, such punishments can
temporarily force players payos below all Nash equilibrium payo levels in the stage
game G. For example, rms in repeated quantity (Cournot) competition can punish
deviations from collusive behavior (such as implicit cartel agreements to restrain
output) by temporarily ooding the market and thereby forcing prots down to
zero.
In the 1950s, several game theorists had conjectured that rational players should
be able to cooperatefor example play C in the above prisoners dilemmaif the
game would only continue long enough (see Section 5.5 in Luce and Raia, 1957).
Its folklore avor is the reason why the result came to be referred to as a Folk
Theorem. As indicated above, Schelling (1956) denitely believed the folk wisdom
and deemed it to be empirically relevant. Still, it was Aumanns precise and general
statement and proof that laid the foundation for subsequent analyses of repeated
interactions. Later, Friedman (1971) established a useful, although partial result
for repeated games: if players discount future payos to a suciently small extent,
then outcomes with higher payos to all players than what they would receive in a
pure-strategy Nash equilibrium of the underlying stage game G can be obtained as
equilibria in the innitely repeated game.
During the cold war, between 1965 and 1968, Robert Aumann, Michael Maschler
and Richard Stearns collaborated on research on the dynamics of arms control ne-
gotiations. Their work became the foundation of the theory of repeated games with
incomplete information, that is, repeated games in which all or some of the players
do not know which stage game G is being played, see Aumann and Maschler (1966,
1967, 1968), Stearns (1967) and Aumann, Maschler and Stearns (1968). For example,
17
a rm might not know a competitors costs and a country might not know another
countrys arsenal of military weapons or the other countrys ranking of alternative
agreements. The extension introduces yet another strategic element: incentives to
conceal or reveal private information to other players. How might a person, rm or
country who has extra information utilize the advantage? How might an ignorant
player infer information known to another player by observing that players past ac-
tions? Should an informed player take advantage of the information for short-run
gains, thereby risking to reveal his information to other players, or should he conceal
the information in order to gain more in the future? Building on the work of John
Harsanyi, Aumann, Maschler and Stearns brought game theory to bear on these sub-
tle strategic issues. Their work is collected and commented upon in Aumann and
Maschler (1995).
Aumann and Shapley (1976) and Rubinstein (1976, 1979) rened the analysis of
repeated games with complete information by showing that all feasible and individ-
ually rational outcomes can also be sustained as subgame-perfect Nash equilibria. In
the context of an innitely repeated game, subgame perfection essentially requires
that the players, in the wake of a unilateral deviation from the equilibrium path of
play, have incentives to play according to the equilibrium. In particular, subgame
perfection requires that no player will ever have an incentive to deviate from punish-
ing a deviator, nor to deviate from punishing a player who deviates from punishing a
player, etc. Many Nash equilibria are not subgame perfect, and it was by no means
clear that such a seemingly stringent renement would leave intact the entire set of
Nash equilibrium payos of supergames. Indeed, as Aumann and Shapley showed, if
players discount future payos, and strive to maximize the expected present value of
their own payo stream, then the set of subgame perfect equilibrium outcomes may be
signicantly smaller than the set of Nash equilibrium outcomes. For while the Nash
equilibrium criterion does not depend on the costs of punishing deviators, the
subgame perfection criterion does. However, their generalized Folk Theorem estab-
lishes that the distinction between subgame perfect and Nash equilibrium disappears
if there is no discounting.
The theory of repeated games has ourished over the last forty years, and we
now have a much deeper understanding of the conditions for cooperation in ongoing
18
relationships. Following a characterization of optimal punishments by Abreu (1988),
it became easier to nd the set of sustainable equilibrium payos in repeated games.
Fudenberg and Maskin (1986) established Folk Theorems for subgame perfect equilib-
rium in innitely repeated games with discounting and an arbitrary (nite) number
of players. Aumann and Sorin (1989) showed that players bounded recall can shrink
the set of equilibria to those that are socially ecient, and Abreu, Dutta and Smith
(1994) essentially characterized the class of games for which the Folk Theorem claim
holds under innite repetition and discounting.
An example of subgame-perfect equilibrium in an innitely repeated game with
discounting is when : identical rms with no xed costs and constant marginal cost
c sell the same product and are engaged in dynamic price competition in a market.
Each rm announces a price in each period and consumers buy only from the rm(s)
with the lowest price, with their demand spread evenly over these rms. If this
interaction took place only once, then the resulting market price would be the same
as under perfect competition: j = c. However, when the interaction takes place
over an indenite future where prots are discounted at a constant rate, many other
equilibrium outcomes are possible if the discounting is not too severe. For example, all
rms may start out by setting the monopoly price j j and continue doing so until a
price deviation has been detected, from which period on all rms set the competitive
price j = c. Such as strategy prole constitutes a subgame perfect equilibrium if
o 1 1,:, where o (0. 1) is the discount factorthe factor by which future
prots are discounted each period.
23
The more competitors there are, the harsher
is thus the condition on the discount factorand hence the harder it is to sustain
collusion.
23
To see this, let (j) be the industry prot when all rms quote the same price j, and assume
that this function is continuous and unimodal with maximum at j = j. The strategy prole here
described constitutes a subgame perfect equilibrium if and only if
1
( j), (1 c) is (weakly) exceeds

(j) for all j < j. The rst quantity is the present value of the rms prot if it continues to set the
collusive price j, while (j) is the present value of the prot to a rm if it undercuts the collusive
price by posting a price j < j such a rm will earn zero prot in all future periods because
all rms will subsequently price at marginal cost. By continuity of the function , the required
inequality holds if and only if c 1 1,:. There is no incentive to deviate from punishment of a
deviator, should a deviation occur, since all prots are zero as soon as any rm quotes j = c.
19
Other strands of the literature examine the possibilities of long-term cooperation
when players are impatient and only have access to noisy signals about past behavior;
prominent early contributions include Green and Porter (1984) and Abreu, Pearce
and Stacchetti (1990). More recent related contributions concern long-lived players,
as well as imperfect public and private monitoring.
24
There is also a literature on
cooperation in nitely repeated games, that is, when the stage game G is repeated
a nite number of times. For example, Benoit and Krishna (1985) established Folk-
theorem-like results for repeated games with multiple Nash equilibria when the time
horizon is nite but long, and Kreps, Milgrom, Roberts and Wilson (1982) showed
that if a prisoners dilemma is repeated suciently many times it takes only a small
amount of incomplete information about payos to sustain cooperation most of the
time, although conict will break out in the last couple of rounds. Neyman (1999)
showed that cooperation in a nitely repeated prisoners dilemma is possible even
under complete information if the time horizon is not commonly known (see below for
a brief discussion of common knowledge in games). Another important contribution to
the literature on repeated games is Axelrod (1984), whose experimental tournaments
suggest that simple strategies such as tit-for-tat perform well in populations of
boundedly rational players.
All these subsequent insights owe much to Aumanns innovative and fundamental
research. When studying cooperation among agents with partly conicting inter-
ests, whether these are rms in a capitalist marketplaceas in many of the rst
applicationsor farmers sharing a common grassland or irrigation systemas in Os-
trom (1990)the theory of repeated games is now the benchmark paradigm.
The theory of repeated games helps to explain a wide range of empirical ndings,
notably why it is often harder to sustain cooperation when there are many players,
when players interact infrequently, when there is a high probability that interaction
will cease for exogenous reasons, when the time horizon is short, and when others
behavior is observed after a delay. Price wars, trade wars and other economic and
social conicts can often be ascribed to one or more of these factors. The repeated-
games framework also sheds light on the existence and functioning of a variety of
24
See Fudenberg and Levine (1994), Fudenberg, Levine and Maskin (1994), Kandori (2002), and
Ely, Horner and Olszewski (2005)
20
institutions, ranging from merchant guilds (Greif, Milgrom, and Weingast, 1994) and
the World Trade Organization (Maggi, 1999) to the maa (Dixit, 2003).
3.2. Other contributions. Aumann has made numerous important contribu-
tions to other aspects of game theory and its application to economics. Here, we only
mention a few of them.
Players knowledge about each others strategy sets, information and preferences
is of utmost importance for their choice of course of action in a game. Thus it is
natural to ask: What epistemic assumptions imply equilibrium play by rational play-
ers? Game theorists were largely silent on this fundamental question, and economists
carried out equilibrium analyses without worrying too much about it, until Aumann
established the research agenda sometimes called interactive epistemology. In his pa-
per Agreeing to disagree (1976), Aumann introduced to game theory the concept of
common knowledge, a concept rst dened by Lewis (1969). An event is common
knowledge among the players of a game if it is known by all players, if all players
know that it is known by all players, if all players know that all players know that it
is known by all players etc., ad innitum. Roughly, Aumann proved that if two play-
ers have common knowledge about each others probability assessments concerning
some event, then these assessments must be identical. Aumanns counter-intuitive
agreement result has had a considerable eect on the theoretical analysis of trade
in nancial markets, see e.g. Milgrom and Stokey (1982).
In the 1980s, Bernheim (1984) and Pearce (1984) showed that players rationality
and their common knowledge of the game and of each others rationality does not,
in general, lead to Nash equilibrium, not even in games with a unique Nash equilib-
rium. A decade later, Aumann and Brandenburger (1995) established tight sucient
epistemic conditions for Nash equilibrium play.
As mentioned above, Aumann dened the concept of strong equilibrium, which
is a renement of Nash equilibrium. In two papers, published in 1974 and 1987, he
also dened another solution concept that is coarser than Nash equilibrium: cor-
related equilibrium. Unlike Nash equilibrium, correlated equilibrium permits players
strategies to be statistically dependent, and thus Nash equilibrium emerges as the
special case of statistical independence. Such correlation is possible if players can
condition their strategy choice on correlated random variables, such as distinct but
21
related observations of the weather, a news event, or some other variable feature of
their environment. In a correlated equilibrium, each players conditioned choice is
optimal, given the others decision rules.
The set of correlated equilibrium outcomes of a complete-information game also
provides the limits to cooperation when players can communicate freely, possibly
through an impartial mediator, prior to choosing their strategies in the underly-
ing game. When each players observed random variable is a recommendation from
an impartial mediator, a correlated equilibrium is a collection of recommendations
such that no player can increase his or her expected payo by a unilateral deviation
from his or her recommendation. In the mobilization game discussed above (see Ta-
ble 1), it can be shown that there are correlated equilibria in which war is avoided
completely, while the negotiation payo pair (/. /) is attained with positive probabil-
ity. To see this, suppose that a mediator recommends exactly one of the countries
to refrain from mobilization with equal probability : for each country, and recom-
mends both to refrain from mobilization with the remaining probability, 1 2:. If
: 2/,(2/+cc) each country will refrain from mobilization if and only if it receives
this recommendation.
25
For a careful discussion of the link between the concept of
correlated equilibrium and the role of communication in games, see Myerson (1991,
Chapter 6).
Aumann (1987) showed that correlated equilibrium can be viewed as a natural
extension of Bayesian decision theory to non-cooperative games. In this interpreta-
tion, rational players (according to the denition of rationality due to Savage, 1954)
will play a correlated equilibrium if their rationality and their probabilistic priors are
common knowledge.
Aumann also made noteworthy contributions to other areas of economics; one is
his joint work on decision theory with Frank J. Anscombe (Anscombe and Aumann,
1963), another is his continuum model of perfect competition (Aumann 1964, 1966),
and a third is his joint work with Mordecai Kurz and Abraham Neyman on applica-
tions of game theory to political economy (Aumann and Kurz, 1977, Aumann, Kurz
25
If a country does not receive a recommendation, then it knows that the other country received a
recommendation to refrain, in which case mobilization is optimal. If a country receives a recommen-
dation, then the expected payo of refraining from mobilization is c + (1 2) / and this exceeds
a, the expected payo of mobilization.
22
and Neyman, 1983 and 1987).
4. Recommended readings
The work of Thomas Schelling is accessible also to non-specialists and we recommend
consulting his original publications. Aumanns writings are highly technical, but
usually also contain easily accessible discussions. See Aumann (1981) for a survey
of the repeated games literature up till then, and Aumann and Maschler (1995) for
a discussion of early work on repeated games with incomplete information. For a
readable and almost entirely non-technical introduction to game theory, see Dixit
and Nalebu (1991); this book discusses long-term cooperation in Chapter 4 and
credible commitments in Chapter 6. For comprehensive books on game theory, see
Dixit and Skeath (2004) for an introductory text and Fudenberg and Tirole (1991) and
Myerson (1991) for advanced and technical expositions. Aumanns and Schellings
personal (if not necessarily current) views on game theory, may be found in Aumann
(1985) and Schelling (1967). For more bibliographic and personal details about the
two game theorists, see Zeckhausers (1989) portrait of Schelling, and Harts (2005)
interview with Aumann.
References
[1] Abreu D. (1988): On the theory of innitely repeated games with discounting,
[2] Abreu D., D. Pearce and E. Staccchetti (1990): Towards a theory of discounted
repeated games with imperfect monitoring, Econometrica 58, 1041-1063.
[3] Abreu D., P. Dutta and L. Smith (1994):The Folk Theorem for repeated games:
A NEU condition Econometrica 62, 939-948.
[4] Admati A.R. and M. Perry (1991): Joint projects without commitment, Re-
view of Economic Studies 58, 259-276.
[5] Anscombe F.J. and Aumann R.J. (1963): A denition of subjective probabil-
ity, Annals of Mathematical Statistics 34, 199-205.
[6] Aumann R.J. (1959): Acceptable points in general cooperative :-person
games, in R. D. Luce and A. W. Tucker (eds.), Contributions to the Theory
23
of Games IV, Annals of Mathematics Study 40, 287-324, Princeton University
Press, Princeton NJ.
[7] Aumann R.J. (1964): Markets with a continuum of traders, Econometrica 32,
39-50.
[8] Aumann R.J. (1966): Existence of competitive equilibria in markets with a
continuum of traders, Econometrica 34, 3-27.
[9] Aumann R.J. (1974): Subjectivity and correlation in randomized strategies,
Journal of Mathematical Economics 1, 67-96.
[10] Aumann R.J. (1976): Agreeing to disagree, The Annals of Statistics 4, 1236-
1239.
[11] Aumann R. J. (1981): Survey of repeated games, in Essays in Game The-
ory and Mathematical Economics in Honor of Oskar Morgenstern, pages 11-42,
Wissenschaftsverlag (Mannheim).
[12] Aumann R.J. (1985): What is game theory trying to accomplish? in K.Arrow
and S. Honkapohja (eds.), Frontiers of Economics, Basil Blackwell, Oxford.
[13] Aumann R.J. (1987): Correlated equilibrium as an extension of Bayesian ra-
tionality, Econometrica 55, 1-18.
[14] Aumann R.J. and A. Brandenburger (1995): Epistemic condition for Nash equi-
librium, Econometrica 64, 1161-1180.
[15] Aumann R. J. and M. Kurz (1977): Power and taxes, Econometrica 45, 1137-
1161.
[16] Aumann R. J., M. Kurz and A. Neyman (1983): Voting for public goods,
Review of Economic Studies 677-694.
[17] Aumann R. J., M. Kurz and A. Neyman (1987): Power and public goods,
Journal of Economic Theory 42, 108-127.
24
[18] Aumann R.J. and M. Maschler (1966, 1967, 1968): Game theoretic aspects of
gradual disarmament, Repeated games with incomplete information: A survey
of recent results, and Repeated games of incomplete information, the zero-sum
extensive case, Reports ST-80, 116 and 143, Mathematica Inc., Princeton, NJ.
[19] Aumann R. J. and M. Maschler (with the collaboration of R. Stearns) (1995):
Repeated Games with Incomplete Information, MIT Press.
[20] Aumann R.J., M. Maschler and R. Stearns (1968): Repeated games of incom-
plete information: an approach to the non-zero sum case, in Report of the U.S.
Arms Control and Disarmament Agency ST-143, Chapter IV, 117-216.
[21] Aumann R.J. and L. Shapley (1974): Value of Non-Atomic Games, Princeton
University Press, Princeton NJ.
[22] Aumann R.J. and L. Shapley (1976): Long-term competition: A game-theoretic
analysis, mimeo. Hebrew University. (Reprinted in N. Megiddo, (ed.) (1994):
Essays in Game Theory in Honor of Michael Maschler, 1-15, Springer Verlag,
Berlin.
[23] Aumann R.J. and S. Sorin (1989): Cooperation and bounded recall, Games
and Economic Behavior 1, 5-39.
[24] Axelrod R. (1984): The Evolution of Cooperation, Basic Books, New York.
[25] Baliga S. and T. Sjostr om (2004): Arms races and negotiations, Review of
[26] Benot J.-P. and V. Krishna (1985): Finitely repeated games, Econometrica
53, 890-904.
[27] Bernheim B.D. (1984): Rationalizable strategic behavior, Econometrica 52,
1007-1028.
[28] Binmore K. and L. Samuelson (2004): The evolution of focal points, Games
and Economic Behavior, forthcoming.
25
[29] Camerer C. (2003): Behavioral Game Theory, Princeton University Press,
Princeton NJ.
[30] Crawford V. (1982): A theory of disagreement in bargaining, Econometrica
50, 607-637.
[31] Dixit A. (1980): The role of investment in entry deterrence, Economic Journal
90, 95-106.
[32] Dixit A. (2003): On modes of economic governance, Econometrica 71, 449-481.
[33] Dixit A. and B. Nalebu (1991): Thinking Strategically: The Competitive Edge
in Business, Politics, and Everyday Life, W.W. Norton.
[34] Dixit A. and S. Skeath (2004): Games of Strategy, 2nd ed., W.W. Norton, New
York.
[35] Ely J., J. Horner and W. Olszewski (2005): Belief-free equilibria in repeated
games, Econometrica 73, 377-415.
[36] Friedman J. (1971): A non-cooperative equilibrium for supergames, Review of
[37] Fudenberg D. and D. Levine (1994): Eciency and observability with long-run
and short-run players, Journal of Economic Theory 62, 103-135.
[38] Fudenberg D., D. Levine and E. Maskin (1994): The Folk theorem with imper-
fect public information, Econometrica 62, 997-1039.
[39] Fudenberg D. and E. Maskin (1986): The Folk theorem for repeated games
with discounting and with incomplete information, Econometrica 54, 533-554.
[40] Fudenberg D. and J. Tirole (1991): Game Theory, MIT Press.
[41] Gale D. (2001): Monotone games with positive spillovers, Games and Eco-
nomic Behavior 37, 295-320.
[42] Green E. and R. Porter (1984): Noncooperative collusion under imperfect price
information, Econometrica 52, 87-100.
26
[43] Greif A., P. Milgrom, and B.R. Weingast (1994): Coordination, commitment,
and enforcement, Journal of Political Economy 102, 745-776.
[44] G uth W., K. Ritzberger and E. van Damme (2004): On the Nash bargaining
solution with noise, European Economic Review 48, 697-713.
[45] Hart S. (2005): An interview with Robert Aumann, Macroeconomic Dynamics,
forthcoming.
[46] Hobbes, T. (1651): The Leviathan, Cambridge.
[47] Kandori M. (2002): Introduction to repeated games with imperfect monitor-
ing, Journal of Economic Theory 102, 1-15.
[48] Kreps D., P. Milgrom, J. Roberts and R. Wilson (1982): Rational cooperation
in the nitely repeated prisoners dilemma, Journal of Economic Theory 27,
245-252.
[49] Lewis D. (1969): Convention, Harvard University Press, Cambridge MA.
[50] Lockwood B. and J.P. Thomas (2002): Gradualism and irreversibility, Review
of Economic Studies 69, 339-356.
[51] Luce D. and H. Raia (1957): Decisions and Games. Wiley, New York.
[52] Maggi G. (1999): The role of multinational institutions in international trade
cooperation, American Economic Review 89, 190-214.
[53] Marx L.M. and S.A. Matthews (2000): Dynamic voluntary contribution to a
public project, Review of Economic Studies 67, 327-358.
[54] Maynard Smith J. (1982): Evolution and the Theory of Games, Cambridge Uni-
versity Press, Cambridge.
[55] Mehta J., C. Starmer and R. Sugden (1994a): The nature of salience: An
experimental investigation of pure coordination games, American Economic
Review 84, 658-673.
27
[56] Mehta J., C. Starmer and R. Sugden (1994b): Focal points in pure coordination
games: An experimental investigation, Theory and Decision 36, 163-185.
[57] Milgrom P. and N. Stokey (1982): Information, trade and common knowledge,
[58] Muthoo A. (1996): A bargaining model based on the commitment tactic,
[59] Myerson R.B. (1991): Game Theory: Analysis of Conict, Harvard University
Press, Cambridge MA.
[60] Nash, J. (1950a): Equilibrium points in n-person games, Proceedings of the
National Academy of Sciences 36, 48-49.
[61] Nash J. (1950b): The bargaining problem, Econometrica 18, 155-162.
[62] Nash J. (1951): Non-cooperative games, Annals of Mathematics 54, 286-295.
[63] Nash J. (1953): Two-person cooperative games, Econometrica 21, 128-140.
[64] Neyman A. (1999): Cooperation in repeated games when the number of stages
is not commonly known, Econometrica 67, 45-64.
[65] Ostrom E. (1990): Governing the Commons: The Evolution of Institutions for
Collective Action, Cambridge University Press, Cambridge.
[66] Pearce D.G. (1984): Rationalizable strategic behavior and the problem of per-
fection, Econometrica 52, 1029-1050.
[67] Phelps E. and R. Pollak (1968): On second-best national savings and game-
theoretic growth, Review of Economic Studies 35, 201-208.
[68] Reiter D. (1995): Exploding the power keg myth: Preemptive wars almost
never happen, International Security 20, 5-34.
[69] Rogo K. (1985): The optimal degree of commitment to an intermediate mon-
etary target, Quarterly Journal of Economics 100, 1169-1189.
28
[70] Rubinstein A. (1976): Equilibrium in supergames, Center for Mathematical
Economics and Game Theory, Hebrew University. (First part reprinted in N.
Megiddo (ed.) (1994): Essays in Game Theory in Honor of Michael Maschler,
17-28, Springer Verlag, Berlin.
[71] Rubinstein A. (1979): Equilibrium in supergames with the overtaking crite-
rion, Journal of Economic Theory 21, 1-9.
[72] Savage L. (1954): The Foundations of Statistics, Dover, Mineola NY.
[73] Schelling T.C. (1956): An essay on bargaining, American Economic Review
46, 281-306.
[74] Schelling T.C. (1960): The Strategy of Conict, Harvard University Press, Cam-
bridge MA.
[75] Schelling T.C. (1966): Arms and Inuence, Yale University Press, New Haven.
[76] Schelling T.C. (1967): What is game theory? in J.C. Charlesworth (ed.), Con-
temporary Political Analysis, Free Press, New York. (Reprinted as Chapter 10 of
Schelling, 1984.)
[77] Schelling T.C. (1971) Dynamic models of segregation, Journal of Mathematical
Sociology 1, 143-186.
[78] Schelling T.C. (1978): Micromotives and Macrobehavior, Harvard University
Press, Cambridge MA.
[79] Schelling T.C. (1980): The intimate contest for self-command, The Public
Interest 60, 94-118.
[80] Schelling T.C. (1983): Ethics, law, and the exercise of self-command, in S.M.
McMurrin (ed.): The Tanner Lectures on Human Values IV, 43-79, University
of Utah Press, Salt Lake City.
[81] Schelling T.C. (1984a): Self-command in practice, in policy, and in a theory of
rational choice, American Economic Review, Papers and Proceedings 74, 1-11.
29
[82] Schelling T.C. (1984b): Choice and Consequence, Harvard University Press,
Cambridge MA.
[83] Schelling T.C. (1992):Addictive drugs: The cigarette experience, Science 255,
430-434.
[84] Schelling T.C. and M.H. Halperin (1961): Strategy and Arms Control, Twentieth
Century Fund, New York.
[85] Selten R. (1965): Spieltheoretische Behandlung eines Oligopolmodells mit
Nachfragetr agheit - Teil I: Bestimmung des dynamischen Preisgleichgewichts,
Zeitschrift f ur die gesamte Staatswissenschaft 121, 301-324.
[86] Selten R. (1975): Re-examination of the perfectness concept for equilibrium
points in extensive games, International Journal of Game Theory 4, 25-55.
[87] Snyder G.H. and P. Diesing (1977): Conict among Nations: Bargaining, Deci-
sion Making, and System Structure in International Crises, Princeton University
Press.
[88] Spence M. (1977): Entry, investment, and oligopolistic pricing, Bell Journal
of Economics 8, 534-544.
[89] Stearns R. (1967): A formal information concept for games with incomplete
information, Report of the U.S. Arms Control and Disarmament Agency ST-
116, Chapter IV, 405-403.
[90] Strotz. R. (1956): Myopia and inconsistency in dynamic utility maximization,
[91] Zeckhauser R. (1989): Distinguished fellow: Reections on Thomas Schelling,
Journal of Economic Perspectives 3, 153-164.
[92] von Neumann J. and O. Morgenstern (1944): The Theory of Games and Eco-
nomic Behavior, Princeton University Press, Princeton NJ.
[93] Wohlstetter A. (1959): The delicate balance of terror, Foreign Aairs 37,
211-234.
11 October 2004

Information Department, P.O. Box 50005, SE-104 05 Stockholm, Sweden

Finn Kydland and Edward Prescotts
Contribution to Dynamic Macroeconomics:
The Time Consistency of Economic Policy and
the Driving Forces Behind Business Cycles

1 Introduction
Finn Kydland and Edward Prescott have been awarded the 2004 Bank of Sweden Prize
in Economic Sciences in Memory of Alfred Nobel for their fundamental contributions to
two closely related areas of macroeconomic research. The rst concerns the design of
macroeconomic policy. Kydland and Prescott uncovered inherent imperfectionscredibility
problemsin the ability of governments to implement desirable economic policies. The sec-
ond area concerns business cycle uctuations. Kydland and Prescott demonstrated how
variations in technological developmentthe main source of long-run economic growthcan
lead to short-run uctuations. In so doing, they oered a new and operational paradigm for
macroeconomic analysis based on microeconomic foundations. Kydland and Precsotts work
has transformed academic research in economics, as well as the practice of macroeconomic
analysis and policymaking.
1.1 General background
During the early postwar period, macroeconomic analysis was dominated by the view mar-
shalled by Keynes (1936). In this view, short-run uctuations in output and employment
are mainly due to variations in aggregate demand, i.e., in investors willingness to invest
and consumers willingness to consume. Moreover, macroeconomic stabilization policy can,
and should, systematically control aggregate demand so as to avoid recurring uctuations
in output. These ideas largely reected the experience from the Great Depression, when
a deep protracted trough in aggregate output, along with falling employment and capital
utilization, were observed throughout the western world. Keynesian macroeconomic analy-
sis interpreted these phenomena as failures of the market system to coordinate demand and
supply, which provided an obvious motive for government intervention.
Until the mid-1970s, the dominating Keynesian paradigm seemed quite successful in ex-
plaining macroeconomic uctuations. But real-world developments in the late 1970s revealed
serious shortcomings of the earlier analysis. It could not explain the new phenomenon of
simultaneous ination and unemployment. This so-called stagation seemed closely related
to shocks on the supply side of the economy: oil price hikes and the worldwide slowdown
of productivity growth. Such supply shocks had played only a subordinate role in the Key-
nesian framework. Moreover, conventional macroeconomic policy, based on existing theory,
was unable to cope with the new problems. Rather, monetary and scal policy appeared
to make matters worse in many countries by accommodating private-sector expectations of
high price and wage increases. This occurred despite the stated objective of governments
and central banks to maintain low and stable ination.
Keynesian models were also criticized on methodological grounds. Models used in applied
work built on broad theoretical and empirical generalizations (reduced forms) summariz-
ing the relationships governing the main macroeconomic variables, such as output, ination,
unemployment, and consumption. Robert Lucass research in the early and mid-1970s (Lu-
cas 1972, 1973, 1976) pointed to the drawbacks of this approach, in particular that the
relationships between macroeconomic variables are likely to be inuenced by economic pol-
icy itself. As a result, policy analysis based on these relationships might turn out to be
1
erroneous. Lucas concluded that the eects of macroeconomic policy could not be properly
analyzed without explicit microeconomic foundations. Only by carefully modeling the be-
havior of individual economic agents, such as consumers and rms, would it be possible to
derive robust conclusions regarding private-sector responses to economic policy. The build-
ing blocks of such an analysise.g., consumers preferences, rms technologies, and market
structuresare likely to be robust to changes in economic policy.
As the Lucas critique rapidly gained wide acceptance, development of an alternative and
operational macroeconomic framework was called for. This was a daunting task, however.
Such a new framework had to be based on solid microeconomic foundations. It also had
to give an integral role to economic policy and economic agents perceptions of how policy
is determined. The award-winning contributions by Kydland and Prescott appeared in two
joint articles, which took decisive steps forward in these respects.
1.2 The contributions in brief
Rules Rather than Discretion: The Inconsistency of Optimal Plans, from 1977, studies
the sequential choice of policies, such as tax rates or monetary policy instruments. The
key insight is that many policy decisions are subject to a fundamental time consistency
problem. Consider a rational and forward-looking government that chooses a time plan for
policy in order to maximize the well-being of its citizens. Kydland and Prescott show that
if given an opportunity to re-optimize and change its plan at a later date, the government
will generally do so. What is striking about this result is that it is not rooted in conicting
objectives between the government and its citizens, nor is it due to the ability of unrestricted
policymakers to react to unforeseen shocks. The result, instead, is simply a problematic
logical implication of rational dynamic policymaking when private-sector expectations place
restrictions on the policy decisions.
A signicant upshot is that governments unable to make binding commitments regarding
future policies will encounter a credibility problem. Specically, the public will realize that
future government policy will not necessarily coincide with the announced policy, unless
the plan already encompasses the incentives for future policy change. In other words, se-
quential policymaking faces a credibility constraint. In mathematical terms, optimal policy
decisions cannot be analyzed solely by means of control theory (i.e., dynamic optimization
theory). Instead they should be studied as the outcome of a game, where current and future
policymakers are modeled as distinct players. In this game, each player has to anticipate
the reaction of future players to current play: rational expectations are required. Kydland
and Prescott analyzed general policy games as well as specic games of monetary and scal
policymaking. They showed that the outcome in a rational-expectations equilibrium where
the government cannot commit to policy in advancediscretionary policymakingresults
in lower welfare than the outcome in an equilibrium where the government can commit.
Kydland and Prescotts 1977 article had a far-reaching impact not only on theoretical
policy analysis. It also provided a new perspective on actual policy experience, such as
the stagation problem. The analysis showed that a sustained high rate of ination may
not be the consequence of irrational policy decisions; it might simply reect an inability of
policymakers to commit to monetary policy. This insight shifted the focus of policy analysis
2
from the study of individual policy decisions to the design of institutions that mitigate the
time consistency problem. Indeed, the reforms of central banks undertaken in many countries
as of the early 1990s have their roots in the research initiated by Kydland and Prescott.
Arguably, these reforms are an important factor underlying the recent period of low and
stable ination. More broadly, the insight that time inconsistency is a general problem for
economic policymaking has shifted the focus not only towards normative research on the
optimal design of institutions such as central banks, but also towards positive research on
the interaction between economic decision-making and political institutions. It has inspired a
large cross-disciplinary literature at the intersection between economics and political science.
Time to Build and Aggregate Fluctuations, from 1982, proposed a theory of business
cycle uctuations far from the Keynesian tradition. In this article, Kydland and Prescott
integrated the analyses of long-run economic growth and short-run macroeconomic uctu-
ations, by maintaining that a crucial determinant of long-run living standards, i.e., growth
in technology, can also generate short-run cycles. Moreover, rather than emphasizing the
inability of markets to coordinate supply and demand, Kydland and Prescotts business cy-
cle model relied on standard microeconomic mechanisms whereby prices, wages, and interest
rates enable markets to clear. They thus argued that periods of temporarily low output
growth need not be a result of market failures, but could simply follow from temporarily
slow improvements in production technologies.
Kydland and Prescott showed that many qualitative features of actual business cycles,
such as the co-movements of central macroeconomic variables and their relative variabili-
ties, can be generated by a model based on supply (technology) shocks. Using uctuations
in technology growth of the same magnitude as those measured from data, Kydland and
Prescott also demonstrated that their simple model could generate quantitatively signicant
cycles. It thus appeared that technology shocks should be taken seriously as a cause of
business cycles.
From a methodological point of view, Kydland and Prescotts article answered Lucass
call for an alternative to the Keynesian paradigm. It was the rst study to characterize the
general equilibrium of a full-edged dynamic and stochastic macroeconomic model based on
microeconomic foundations. This required solving a set of interrelated dynamic optimization
problems, where consumers and rms make decisions based on current and expected future
paths for prices and policy variables, and where the equilibrium price sequences are such
that private-sector decisions are consistent with clearing markets at all points in time and
all states of the world. Kydland and Prescott showed that this challenging analysis could
be carried out in practice by extensive use of numerical methods. Their empirical approach
relied on model simulation, based on so-called calibration, and on comparing the synthetic
data from simulations with actual data. Such calibration can be regarded as a simple form of
estimation, where model parameters are assigned values so as to match the models long-run
macroeconomic features with those in the data and render the behavior of individual agents
in the model consistent with empirical microeconomic studies.
Kydland and Prescotts work on business cycles initiated an extensive research program.
Successively more sophisticated dynamic models of business cycles have been formulated,
solved numerically, and compared to data using both calibration methods and econometric
3
estimation. Kydland and Prescotts emphasis on supply shocks led researchers to reconsider
the origins of business cycles and assess the relative importance of dierent shocks. Their
results were established for well-functioning markets, while subsequent research considered
various market imperfections and examined their implications. As a result of these eorts,
current state-of-the-art business-cycle models give prominence to both supply and demand
shocks. These models rely on explicit microeconomic foundations to a much larger extent
than did earlier Keynesian models. For example, so-called new-Keynesian models, which
have become a standard tool for analyzing monetary policy, have a core similar to Kydland
and Prescotts original model, but incorporate frictions in the form of sticky prices and
wages.
Kydland and Prescotts two articles have central themes in common. Both articles view
the macroeconomy as a dynamic system, where agentsprivate agents and policymakers
make rational, forward-looking, and interrelated decisions. Both articles provide insights
into postwar developments in the world economy, in terms of private-sector or government
behavior. Both articles oer a new perspective on good macroeconomic policy, leading to a
reconsideration of policymaking institutions and a dierent approach to stabilization policy.
Separately, each of the articles spawned a large independent literature. In the following, we
describe the contributions in more detail.
2 Time consistency of economic policy
In the late 1960s and early 1970s, macroeconomic research paid particular attention to the
expectations held by private agents. A rst step was to emphasize expectations as important
determinants of economic outcomes. Friedman (1968) and Phelps (1967, 1968) based their
natural-rate theories of unemployment on the expectations-augmented Phillips curve, where
the relationship between actual ination and unemployment depends on expected ination.
A second step was to study expectations formation in more depth. Lucas (1972, 1973)
based his analysis on the rational-expectations hypothesis, according to which economic
agents make the best possible forecasts of future economic events given available information,
including knowledge of how the economy functions, and where the best possible forecast
presumes that other agents act according to the same principle, now and in the future.
Kydland and Prescotts analysis of economic policy design added a new dimension to
expectations formation. Their model endogenized government decision-making by assuming
that governments choose policy in order to maximize the welfare of their citizens. Kydland
and Prescott followed Lucas in assuming that the private sectors expectations about fu-
ture government policy are rational; they also followed Friedman and Phelps in assuming
that those expectations are important determinants of economic outcomes. Under these as-
sumptions, Kydland and Prescott showedby way of a general argument as well as specic
examplesthat government policymaking is subject to a time consistency problem.
Kydland and Prescotts 1977 paper contained several related contributions, both method-
ological and substantive. First, they pointed to the general origin of the time consistency
problem: without a commitment mechanism for future policy, the government faces an ad-
ditional constraint in policymaking because its policy has to be credible. In other words,
4
if private expectations about future policy choices are rational, a certain set of economic
outcomes are simply not attainable under discretionary policy. Second, they derived the
policy outcomes that would result in this case if both private agents and policymakers act
rationally. They characterized time-consistent equilibria without commitment and demon-
strated that such equilibria involve lower welfare than those with government commitment.
Third, they argued that more or less unalterable policy rules may be called for. This initi-
ated a discussion about institutional design aimed at creating commitment mechanisms that
enlarge the set of feasible economic outcomes and improve economic well-being. In all these
ways, Kydland and Prescotts contribution has fundamentally changed the way we think
about economic policy.
2.1 The general idea
The following simple and abstract model with two time periods, t1 and t, suces for
describing the problem. In period t 1, a government wants to attain the best possible
outcome for economic agents in period t. Economic outcomes in period t depend not only
on the policy undertaken in period t, but also on private-sector decisions made in period t1
(determining, for example, savings or wages in period t). Private-sector decisions in period
t1 depend on expectations about the period-t policy. These expectations are rationally
formed. In period t1, private-sector agents understand the determinants of government
policy in period t and base their forecasts on this knowledge. There is no uncertainty in
the model, so that rational expectations imply perfect foresight on the part of private-sector
agents.
In the case with commitment, the government chooses its period-t policy in period t1,
without the ability to change this policy later on. Clearly then, the optimal choice of period-t
policy has to take into account its eects on private-sector decisions in period t1. Because
the equilibrium period-t policy pins down the expectations in period t1 as to what this
policy will be, it inuences private-sector decisions made in period t1 that aect period-t
economic outcomes.
In the more realistic case without commitment, i.e., with discretionary policy, in period
t1 the government cannot make binding decisions over policy in period t until period
t arrives. In this case, by contrast, the period-t policy choice will not take into account
how private-sector decisions are made in period t1, because when this policy choice is
undertaken, private-sector decisions in period t1 have already been made and can no
longer be inuenced. This will generally lead to dierent period-t policy choices than in
the commitment case. As a result, economic outcomes will lead to lower welfare than under
commitment. This follows from the fact that with rational expectations, policy in period
t is perfectly anticipated, but owing to the time sequencing of decision-making, there is no
way for the government to inuence these expectations. When deciding its policy in period
t, the government thus solves an optimization problem that does not consider all the eects
of its policy choice.
To see how time inconsistency arises in this example, suppose a government that cannot
make commitments in period t1 announces an intention to adopt the same period-t policy
as the optimal policy it would (hypothetically) have selected under commitment. This an-
5
nouncement would not be credible, because when period t arrives and the government makes
its actual policy choice, it would discover that it is optimal to renege on its announcement.
2.2 Methods
Given the problem at hand, the governments policymaking problem cannot be analyzed
as an optimal-control problem, as in conventional macroeconomics. In an optimal control
problem, the optimizer chooses a control-variable sequence that maximizes an objective func-
tion subject to constraints. But in Kydland and Prescotts setup the dependence of private
behavior on (rational) private expectations about policy makes the constraints endogenous.
These constraints describe relations between current policy and current economic outcomes
thatvia private-sector expectationsare inuenced by future policy choices. This ne-
cessitates the use of game-theoretic methods in order to determine equilibrium outcomes.
Kydland and Prescott used dierent equilibrium concepts in the dierent models and ex-
amples appearing in their paper. In one, they used the equilibrium concept proposed by
Nash (1950). Another model solution was based on sequential rationality, close to Seltens
(1965) subgame-perfect equilibrium, where the expectations of all agentsall private agents
as well as the governmentare consistent with future equilibrium behavior, regardless of the
choices made today. In games with a nite time horizon, such a solution can be found by
backwards induction. Kydland and Prescott also studied a model example with an innite
time horizon showing how to use recursive methods. Such methods are particularly useful
for dening a special type of equilibria, so-called Markov-perfect equilibria, where current
actions are time-invariant functions of payo-relevant variables only.
1
2.3 Examples and applications
Kydland and Prescott dealt informally with several examples of time inconsistency. They
pointed out that government assistance in protection against natural disasters, such as oods
and earthquakes, may not be optimal ex ante, whereas it is optimal ex post. Suppose that an
uninhabited area is likely to be aected by tropical storms, and that this risk is so high that
it is not socially desirable from an ex-ante perspective for the population to settle there. The
necessary protective assistance, which only the government can undertake, is too costly. The
question then is what action the government would undertake if the area is in fact settled:
either it assists settlers in constructing protective devices to limit losses in the event of a
storm, or it refrains. When it is socially desirable to provide protection ex post, there is a
time consistency problem. If the government can commit to not providing such assistance in
the event the area is settled, the citizens will simply not settle there and the socially desirable
outcome is attained. If, on the other hand, the government cannot commit, there will be
settlement, since the citizens then know that they will receive assistance and protection, and
a socially less desirable outcome is obtained.
1
In a stationary model with an innite time horizon, a Markov-perfect equilibrium is formally found as a
xed point in function space: the equilibrium strategy (rule) has to represent optimal current behavior when
expectations of future behavior are given by the same rule. Under an innite time horizon, other types of
equilibria are also possible. For a more detailed discussion, see Maskin and Tirole (2001).
6
Another example concerns the protection of patents for technological innovations. Sup-
pose a government can commit regarding the extent to which it will protect patents in the
future. It can then optimally balance the negative eects of monopoly power for the inno-
vators receiving the patents, against the benets of creating incentives for innovation so as
to obtain ex-post monopoly power. If the government cannot commit, however, the relevant
incentives for innovation will not be taken into consideration.
In addition to an abstract general model, Kydland and Prescotts formal analysis treated
two cases: tax policy and stabilization (ination) policy. Here, we rst consider a tax-policy
problem, similar in spirit to Kydland and Prescotts but simplied for expositional reasons.
2
We then turn to Kydland and Prescotts own example of stabilization policy.
2.3.1 Optimal taxation
A government taxes a large number of identical consumers in order to nance a given level of
(per capita) expenditures, G. This government has access to two tax bases, capital income,
K, and labor income, L, and takes into account the distortions caused by taxation.
3
Each of
the (per-capita) tax bases in period t depends negatively on the tax rate that applies to it:
K(
t
) and L(
t
), where
t
and
t
are the tax rates on capital and labor income, respectively,
are both decreasing functions.
4
The government budget constraint can thus be stated as
t
K(
t
) +
t
L(
t
) = G.
Next, assume that private agents decide on the supply of capital in period t 1, whereas
they decide on the supply of labor in period t. We consider two cases.
Suppose rst that the government can set the tax rates
t
and
t
already in period
t 1, i.e., with full commitment. Optimal behavior, provided that the government wants
to maximize the representative consumers equilibrium utility, follows Ramseys well-known
elasticity principle. In particular, the optimal tax rates can be solved from the governments
budget constraint and the equation
t
1
t
K
=

t
1
t
L
,
where
x
is the elasticity of x with respect to its own (net-of-tax) price:
x
(dx/x)/(dp
x
/p
x
).
Intuitively, the government wants to equalize the distortion on the last unit of revenue raised
across the two tax bases, which implies that the less elastic tax base is assigned a higher tax
rate. What is particularly important here, however, is that the governments choice of
t
and
t
takes into account how the supply of capitalfrom the consumers savings in period
2
The presentation here is related to Fischer (1980) and is based on Persson and Tabellini (1990).
3
For simplicity, we set the prices of both capital and labor services to one.
4
The assumptions on K and L can be derived from rst principles: suppose that the consumers (quasi-
linear) utility function reads u(c
t1
)+c
t
+v(1l
t
). Here, u and v are strictly increasing and strictly concave
functions, and 1 l
t
is interpreted as leisure time, where 1 is the total time endowment. With the budget
constraints c
t1
+ k
t
= y
t1
and c
t
= (1
t
)k
t
+ (1
t
)l
t
+ y
t
, where y
t1
and y
t
represent income from
other sources in periods t 1 and t, respectively, utility maximization gives strictly decreasing capital and
labor supply functions similar to those in the text.
7
t 1depends on the choice of
t
. Thus, the Ramsey formula for tax rates represents the
optimal solution to the ex-ante taxation problem.
Suppose, by contrast, that the government cannot commit to
t
and
t
in advance. Given
some amount of savings from period t 1, how will the government set tax rates in period
t, once this period arrives? The taxation problem becomes trivial and Ramseys principle
for taxation can be applied to the ex-post taxation problem. Since the supply of capital is
entirely inelastic at this pointcapital is predeterminedall the capital income should be
taxed away before labor taxes are used! Ex post, this creates no distortions on the supply of
capital and mitigates distortions on the supply of labor. Because income from capital has a
dierent elasticity ex ante than ex post, the commitment solution is not time consistent. If
the government were given a chance to re-optimize in period t, it would change the ex-ante
optimal plan. Thus, there is a credibility problem: in period t 1, the government cannot
just announce the tax rates that solve the problem under commitment and hope that the
private sector will believe the announcement.
What is the time-consistent solution when both the government and private agents are
rational? Applying the requirement of sequential rationality discussed in Section 2.2, no
matter how much is saved in period t 1, the decisions of the government and the private
sector have to be optimal in period t. Suppose that G is suciently large, i.e., that labor
taxation is always necessary in order to nance the governments expenditures. Then, ratio-
nality in period t dictates that the ex-post tax rate on capital is 100 percent. As a result, all
consumers choose zero savings in period t 1: rational expectations of government policy
choice in t tells them that any savings are wasted from their perspective. It follows that, for
any amount of private savings,
0 +
t
L(
t
) = G,
which can be used to solve for the tax rate on labor in period t.
5
Clearly, this outcome
is worse than the outcome under commitment (as both tax rates are higher, the utility of
consumers must be lower).
2.3.2 Optimal stabilization policy
Consider a monetary policymaker who faces a trade-o between ination and unemploy-
ment. Private-sector behavior is represented by an expectations-augmented Phillips curve.
Unemployment in period t, u
t
, is given by
u
t
= u
(
t
E(
t
)),
where u
is the equilibrium (natural) rate of unemployment,

t
is the ination rate between
periods t 1 and t, E(
t
) is the ination rate expected by the private sector in period t 1,
and is a positive exogenous parameter. This is a reduced-form relation which could be
motivated, for example, by assuming that (i) the demand for labor depends negatively (and
therefore unemployment depends positively) on the real wage in period t; and (ii) nominal
wage contracts are set in advancein period t 1based on expectations of period-t prices.
5
In general, (at least) two values of
t
solve the equation
t
L(
t
) = G. As the government optimizes, it
will choose the lowest of these on the upward-sloping portion of the Laer curve.
8
Then, higher-than-expected ination lowers the real wage, labor demand increases, and
unemployment falls.
The policymakers objective is to maximize the function
S(u
t
,
t
),
where S is weakly decreasing and concave in each argument and has a maximum at the
point where u
t
< u
and
t
= 0.
6
The indierence curves representing equal values of S are
shown in Figure 1 (taken from Kydland and Prescott, 1977). The literature has subsequently
tended to use a particular quadratic form for S:
S =
1
2
(u
t
ku
)
2
1
2
2
t
,
where represents the weight the policymaker assigns to ination (relative to unemployment)
and k < 1 represents some distortion that makes the government target an unemployment
rate lower than the equilibrium rate.
7
The policymaker can use monetary policy in period t to control
t
. Without uncertainty,
as assumed here, rational expectations on the part of the private sector imply E(
t
) =
t
.
Therefore, actual unemployment always equals equilibrium unemployment, i.e., u
t
= u
.
If, in period t 1, the policymaker can commit to the ination rate in period t, then the
optimal policy is obvious. Since u
t
= u
must hold, the policymaker will choose

t
= 0. The
ex-ante optimal outcome (u
, 0) is illustrated in Figure 1 as point O.

The commitment outcome, however, is not time consistent. Ex post, when expectations
have already been formed (nominal wages are predetermined when the government chooses
period-t ination), the government nds it optimal to choose an ination rate higher than
0 if it is allowed to change its zero-ination plan. An inationary policy will reduce un-
employment, which by assumption raises welfare.
8
Ex post, optimal policy is dictated by
the condition S/u = S/: the ination rate is increased to the level at which the
marginal gain from lower unemployment equals the marginal cost of higher ination. This
outcome, characterized by an ination bias", is labeled C in Figure 1.
Using the parametric form of the model, it is easily shown that equilibrium ination
under lack of commitment is given by
9
t
=
(1 k)u
.
Thus, ination is higher when the dierence between the policymakers unemployment target
and the equilibrium rate is larger (k lower, or u
higher), the weight it assigns to ination is

6
Kydland and Prescotts assumption that a zero ination rate is optimal is not essential for the conclusions.
7
This formulation was rst used by Barro and Gordon (1983a).
8
The key assumption required here is that the marginal welfare gain from lowering unemployment below
u
should exceed the marginal welfare loss from raising ination above zero; for the functional form given,
the former is strictly positive and the latter is zero (to the rst order).
9
This expression is derived by using the rst-order condition, when the government takes E(
t
) as given,
i.e.,
t
= (u
t
ku
). Combining this condition with the Phillips curve u

t
= u
(
t
E(
t
)) and using
the rational expectations condition
t
= E(
t
), one arrives at the expression in the text.
9
lower ( lower), or the responsiveness of unemployment to real wages is greater ( higher).
Policymakers who care a great deal about unemployment, but little about ination, thus
end up without lower unemployment, but with higher ination, than do policymakers with
a stronger preference for low ination relative to low unemployment. This striking result
is entirely due to the lack of commitment. Under commitment, the ex-post temptation to
inate is absent, and rational expectations dictate the same outcome for any preference
parameters.
2.4 Subsequent research
Kydland and Prescotts contribution oered a politico-economic explanation as to why many
countries found themselves in a process of self-generating ination despite repeated promises
to combat it. In particular, Barro and Gordon (1983a) expressed and elaborated further on
this view. Based on Kydland and Prescotts analysis, they formulated a positive theory of
monetary policy and ination. According to this theory, ination is higher when equilibrium
unemployment is higher relative to the unemployment target that politicians try, but are
not able, to attain.
Barro and Gordon (1983a,b) also introduced supply shocks and stabilization policy into
the model and showed that these extensions did not alter the basic conclusions. In the
extended model, a prior commitment to adhere to a certain policy should be interpreted as
commitment to a rule that makes policy respond optimally to future macroeconomic shocks.
Such a (contingent) rule leads to better macroeconomic outcomes than discretionary policy:
it delivers lower ination and the same average unemployment rate over the cycle, as well as
the same extent of macroeconomic stabilization.
A challenging issue in research as well as in practice was whether the time-consistency
problem could be resolved. One possibility was oered by Barro and Gordon (1983b), Backus
and Drill (1985), and Tabellini (1985). These authors borrowed insights from the theory of
repeated games to show that, under some conditions, equilibria with low ination could also
arise under discretionary policymaking. According to this theory, when monetary policy-
makers invest in a good reputation by combatting ination, they inuence private-sector
expectations regarding future ination rates.
Kydland and Prescott also pointed to the potential benet of legislated rules that would
introduce a time lag between policy decisions and their implementation (similar to the lag
often applied to constitutional change). An apparent drawback of such rules is that, in prac-
tice, they would have to be quite simple; this would make it dicult to react to unforeseen
macroeconomic events that required a policy response. Thus, simple, non-contingent rules
for monetary policy, such as rules prescribing a xed rate of ination or a xed exchange rate,
may be less desirable than discretionary policy because the uctuations in output under the
simple rule would become too large. These problems prompted other researchers to focus on
institutional reforms that would improve the performance of discretionary policymaking.
In the context of monetary policy, Rogo (1985) demonstrated that a proper balance
between credibility (of low-ination policy) and exibility (stabilization) could be achieved
through delegation of monetary policy to an independent central bank. In particular, ac-
cording to Rogos analysis, if an independent central bank is managed by a conservative
10
central bankeri.e., an agent who is more ination averse than citizens in generalthen
better welfare outcomes can be achieved. In the language of the stabilization policy example
above, an agent with a higher than in the social welfare function of society should be
selected to head and independently run the central bank. This nding has been used in the
practical policy debate in many countries to recommend the independence of central banks
from direct government inuence. The theoretical presumptionthat it should be possible
to reduce ination through such institutional reform without any cost in terms of increased
average unemployment over the business cyclehas received support from empirical re-
search examining the eects of monetary policy institutions on macroeconomic outcomes
both across countries and over time.
10
Rogos idea has since been developed and rened substantially, for example in analyses of
monetary regimes with explicit ination targets and incentive contracts for central bankers
(e.g., Walsh, 1995 and Svensson, 1997). The literature has also considered a number of
extensions of the economic environment from a simple Phillips curve to dynamic models of
the macroeconomy. Much of the recent literature has focused on time consistency problems
associated with eorts to use monetary policy for the purpose of stabilizing ination and
unemployment around their target values. Such problems also arise when monetary policy
is not characterized by a general ination bias. Some of these models have analyzed the
potential for counteracting temporary increases in ination with lower costs in terms of lost
output and employment if future low-ination policies are credible (see, e.g., Clarida, Gali,
and Gertler, 1999, for a survey of this research). Many of these applications are based on
Kydland and Prescotts methods for characterizing equilibria. Other aspects of the design of
monetary institutions, such as the transparency and accountability properties of alternative
institutions, have also been analyzed from the stepping stone of Kydland and Prescotts
insights.
Time consistency problems have also been examined in other areas of government pol-
icymaking. Important early examples in the literature on scal policy are Fischers (1980)
analysis of the taxation of capital and labor as well as Lucas and Stokeys (1983) analysis of
government debt and the timing of taxes. Subsequent theoretical and empirical research has
studied how balanced-budget rules or other restrictions on scal policy inuence government
spending, budget decits, and international indebtedness.
More generally, a vibrant literature has developed in the area of political economics, bor-
rowing insights and methods from Kydland and Prescotts work. This research program
extends the analysis of credibility problems to issues of political incentives and political in-
stitutions as central determinants of policy outcomes. Signicant cross-disciplinary work
has incorporated ideas and tools from political science to analyze conicts of interest be-
tween voters and policymakers, and between dierent groups of voters or political parties
(Kydland and Prescotts examples all view the private sector through the lens of identical
representativeindividuals). This area of research extends far beyond stabilization and
taxation policy, dealing with trade policy, regulation policy, labor-market policy, economic
growth, and so on. Several recent books, aimed at graduate students in economics and po-
10
See, e.g., Grilli, Masciandro, and Tabellini (1991), Cukierman (1992), and Alesina and Summers (1993),
for early empirical analyses.
11
litical science as well as active researchers in the area, summarize this body of research (see,
e.g., Drazen, 2000, Persson and Tabellini, 2000, and Grossman and Helpman, 2001).
2.5 Broader impact
Kydland and Prescotts analysis of the time consistency problems inherent in monetary pol-
icy provides an explanation as to why many countries appeared to be trapped in a vicious
spiral of high ination during the 1970s, despite continuous declarations from governments
and central banks that ination would be brought down. Kydland and Prescotts recom-
mendation that policies should be rules-based rather than discretionary sparked a debate,
where simple rules for money growth, xed exchange rates, etc., were suggested as solutions
to the ination problem. Following the academic literature reviewed above, the practical
policy debate also changed focus. Discussions of isolated policy actions gave away to explicit
consideration of the broad institutional framework shaping the incentives of policymakers
and thus determining which policies would be credible and politically feasible.
Since the early 1990s, a number of countries have pursued radical institutional reforms of
their monetary policy frameworks, especially by increasing the independence of their central
banks regarding the operational conduct of policy to achieve the objectives set forth by
the political system. At the same time, these objectives have been stated more clearly,
usually with price stability as the primary goal. Central bank reforms in countries such as
New Zealand, Sweden, and the United Kingdom drew extensively on the conclusions from
the academic literature initiated by Kydland and Prescott, as did discussions about the
design of the new European Central Bank, ECB, in the European Union. There is also a
close connection between the academic research on this topic and the increasing reliance on
explicit ination targets among central banks in developed as well as developing countries.
2.6 Related literature
Time consistency problems in monetary policy, due to the governments desire to raise rev-
enue from surprise ination, were pointed out already in Auernheimer (1974). Calvo (1978)
examined such time consistency problems but did not derive the time-consistent solution
under discretionary decision making or analyze possible solutions. Another type of time
consistency problem appears in the literature on savings decisions initiated by Strotz (1956)
and later developed by Phelps and Pollak (1968). Here, time inconsistency enters directly
through preferences that change over time, whereas in Kydland and Prescotts analysis,
inconsistency is embedded in the constraints.
A time consistency problem more closely related to that studied by Kydland and Prescott
may be found in Buchanans (1975) discussion of the Samaritans dilemma, but again with-
out a systematic formal analysis of the time consistency problem associated with government
policy. Similar problems are mentioned in Elster (1977). A related problem also appears
in the literature on price setting by a durable-goods monopolist. The idea here is that a
monopoly producer of a new good would like all consumers to believe that the good will
be continued to be sold at a high price. Consumers with a high willingness to pay would
then purchase the good at the high initial price, after which it could be sold at a lower
12
price to the remaining consumers. The so-called Coase conjecture (1972) holds that pricing
will occur at marginal cost because consumers are forward-looking, whereby the monopolist
actually competes in price with his future selves. Formal game-theoretic analyses of this
problem have later been provided by Stokey (1981), Bulow (1982), and Gul, Sonnenschein,
and Wilson (1986).
3 The driving forces behind business cycles
The last two decades have witnessed radical changes in the theory and practice of business
cycle research, andmore generallyin the predominant views on business cycle phenom-
ena. Keynesian analysis from the early postwar period relied on a set of relationships among
aggregate variables (reduced forms) that were intended to sum up the interactions be-
tween various structural relationships. Although each such relationship was motivated by
microeconomic theory of consumer and rm behavior, it was usually not explicitly derived
from such theory. More importantly, dierent macroeconomic relationships were not based
on a common microeconomic structure when used together in applied macroeconomic anal-
ysis.
Estimated versions of such business cycle models were widely used in practical forecasting
and policy-oriented evaluations of monetary and scal policy interventions. By the mid-
1970s, the Lucas critique (Lucas, 1976) had pointed to serious problems with this approach.
Estimated reduced-form relationships could not be expected to be robust to changes in policy
regimes or in the macroeconomic environment. Macroeconomic developments emphasized
this critique when seemingly stable macroeconomic relationships, based on historical data,
appeared to break down in the 1970s. In particular, the new stagationhigh unemployment
combined with high inationplayed havoc with the Phillips curve, which had earlier seemed
to trace out a stable negative relationship between the rates of ination and unemployment.
The experiences of the 1970s also called into question the predominant idea that business
cycles are driven mainly by changes in demand. Instead, the contemporary macroeconomic
uctuations seemed to be caused largely by supply shocks, such as the drastic oil price hikes
in 1973-74 and 1979 and the worldwide slowdown in productivity growth as of the mid-1970s.
Lucas proposed formulating a new macroeconomic theory on rmer ground, i.e., on an ex-
plicit microeconomic structure instead of postulated aggregate relationships. This structure
would include assumptions about consumers and their preferences, rms and their technolo-
gies, the information of these agents, in what specic markets they interact, and so on. On
the basis of these deep parameters, general-equilibrium implications for aggregate variables
would be derived and confronted with data. Consumers preferences and rms technolo-
gies were not likely to be aected by changes in scal or monetary policy regimes or in the
macroeconomic environment, even though the behavior of consumers and rms would be
aected . Hence, quantitative analysis based on microeconomic underpinnings is likely to
be more robust to such regime changes and thus more useful in policy analysis than models
based on historical aggregate relationships.
Unfortunately, Lucass guidelines were not accompanied by an operational prescription
for implementing them. The development of an alternative macroeconomic modelling ap-
proach, which would satisfy even the minimal requirement of deriving joint predictions for the
13
main macroeconomic variables from sound microeconomic foundations, seemed a daunting
task. Such a theory would have to be dynamic to properly model investment, consumption,
and other intertemporal decisions on the basis of optimal, forward-looking behavior of rms
and households. Simple dynamic models with rational expectations certainly existed and a
research program on how to econometrically estimate such models was underway, following
the pathbreaking work by Sargent (1977, 1978, 1979). These models involved drastic simpli-
cations, however, and essentially required representing the economyor parts of itby a
few linear relationships. Around 1980, traditional (likelihood-based) econometric estimation
of dynamic, non-linear models on a rich enough form to be operationally used in quantitative
macroeconomic analysis seemed out of reach.
Kydland and Prescotts 1982 paper transformed macroeconomic analysis in several di-
mensions. Indeed, it provided a blueprint for rendering Lucass proposal operational. In their
modeling, Kydland and Prescott relied on a stochastic version of the neoclassical growth
model, which has since become the core of much macroeconomic modeling. They showed
that technology shocks, i.e., short-run variations around the positive growth trend for tech-
nology that makes economies grow in the long run, could be an important cause of output
uctuations. In todays macroeconomic models, supply shocks typically play an important
role alongside demand shocks. In their model solution, Kydland and Prescott relied on
numerical solution and computer simulation to an extent not previously implemented in eco-
nomics. Nowadays, numerical analysis of economic models is an indispensable element in the
tool kit of graduate students in economics. In their empirical implementation, Kydland and
Prescott relied on so-called calibration, a simple but informative form of estimation when
confronting new models with data. Since then, new macroeconomic theory is frequently
compared with data using these methods. In all these ways, Kydland and Prescotts work
changed not only the basic methodology of business cycle analysis, but also our perspective
on the importance of various types of shocks and their propagation mechanisms.
3.1 The general idea
We begin by outlining the general features of Kydland and Prescotts business cycle the-
ory. Next, we review their methodology, and outline a specic simple example of model
formulation, along with a brief look at empirical implementation.
Kydland and Prescott set out to integrate business cycle theory and growth theory. Since
they viewed technology shocks as potentially important sources of short-run output uctua-
tions, it seemed natural to turn to the neoclassical growth modelthe workhorse of growth
theory ever since the research of Robert Solow (1956). Another reason for using the neo-
classical growth model was related to the problem of distinguishing the short run (cycles)
from the long run (growth); as the long run, by necessity, is a sequence of short runs.
Moreover, most variables of interest in growth theory and business cycle theory coincide.
Kydland and Prescotts starting point was the fact that the U.S. economy, and many
other Western economies as well, had grown at an average annual rate of around 2 percent
for approximately 100 years, increasing output by a factor of seven. Their hypothesis was
that technology growth might be an important determinant, not only of long-term living
standards, but also of short-term uctuations, to the extent that technology growth displays
14
variations over time. One way of measuring technology growth relies on growth accounting,
another tool developed by Solow (1957). Based on certain assumptions about the working of
the economy (constant returns to scale, perfect competition, and market clearing), consistent
with the model economy studied by Kydland and Prescott, this procedure accounts for the
part of output growth due to the growth of inputs (labor and capital, notably). The residual
componentthe Solow residualis then interpreted as technology growth. Kydland and
Prescott (1982) assumed a standard deviation for technology shocks of the same magnitude
as for the Solow residuals. Measurement based on Solow residuals implies relatively large
variations in technology growth over time, a substantial part of which appear at business
cycle frequencies. More rened methods have subsequently been used (see Section 3.4 below).
Conceptually, Kydland and Prescott studied a closed-economy dynamic, stochastic gen-
eral equilibrium model with perfect competition and no market frictions. How do technology
shocks translate into output movements in this model? A positive technology shock in pe-
riod t represents a higher-than-average growth rate of total factor productivity, i.e., a large
increase in the economys ability to produce output from given supplies of capital and labor.
Higher productivity raises wages, so labor supply in period t increases as workers nd work
more protable than leisure. Thus, two eects serve to raise period t output: the direct eect
of higher productivity and the indirect eect of higher labor input. The return to capital
increases as well, but the capital stock in period t is predetermined. Thus, if the technology
shock in period t had been foreseen, the implied increase in the period-t return to capital
could also have led to higher investment in previous periods, thus raising output in period t
through a third, indirect channel.
The boost in period-t output has dynamic consequences. Part of the increase in output
is consumed, while the remainder is saved and invested. The split depends on consumers
preferences and the expected longevity of the productivity shock. Theory and microeconomic
evidence indicate a desire to smooth consumption over time, and the portion of a temporary
increase in output that is saved depends on the preference for smoothing. The less quickly
the productivity shock is expected to die out, the more protable will it be to save and
invest. Kydland and Prescott based their technology growth series on the data, which feature
signicant positive autocorrelation, thus leading to an investment response to a current shock
which is higher than if technology growth were uncorrelated over time. This raises the capital
stock in period t +1, while technology is still above trend due to autocorrelation. Incentives
for higher than normal labor supply thus remain, andif the increase in the capital stock
is large and technology shocks are suciently autocorrelatedlabor supply in period t + 1
will be more above trend than in period t, as will investment.
These dynamic eects constitute the models propagation mechanism, whereby an im-
pulse of a temporary technology shock shapes the path of future macroeconomic variables.
The mechanism is stable, i.e., the eects of an impulse eventually die out, because the tech-
nology growth process is mean-reverting and because decreasing returns to capital bring
investment back to trend.
The theory delivers time series for macroeconomic variables broadly consistent with data.
Due to the propagation mechanism, all macroeconomic aggregates display high autocorrela-
tion and high co-movement, and the volatility of investment is higher than that of output,
which is higher than that of consumption. The economy goes through booms and busts,
15
with recessions caused by lower-than-average technology growth leading workers to work
fewer hours and consumers to invest less. Calibrated with parameters from microeconomic
studies and simulated with impulses from an estimated technology growth process, Kyd-
land and Prescotts baseline model generates output uctuations that amount to around 70
percent of those observed in postwar U.S. data.
3.2 Methods
Kydland and Prescott studied a dynamic, stochastic general equilibrium model. An equi-
librium in the model is a stochastic process for quantities and prices such that (i) given the
price processes, consumers and rms choose quantities so as to maximize expected utility and
maximize prots and (ii) markets clear. Property (i) embeds rational expectations; in a full-
edged dynamic model, unbiased predictions of the future evolution of prices is an element of
optimizing behavior. Basic theorems ensuring the existence of an equilibriumwhich, math-
ematically, required solving a xed-point problem in high-dimensional spacewere already
provided in the work of Arrow and Debreu (see Debreu, 1959). However, a precise charac-
terization of an equilibrium was very dicult, due to the complexity of dynamic stochastic
analysis. Thus, Kydland and Prescotts 1982 paper made several simplications of the gen-
eral structure described by Arrow and Debreu.
Kydland and Prescott considered only one consumption good and one type of innitely
lived consumer (to be interpreted as a dynastic family: a sequence of parents and children
with altruistic preferences vis--vis ospring). Moreover, as in the standard neoclassical
growth model, Kydland and Prescott assumed only one type of production technology: an
aggregate production function, based on the inputs of capital and labor. They also as-
sumed that markets are devoid of frictions, so that any equilibrium is Pareto optimal. This
facilitated matters in the sense that standard welfare theorems allowed them to nd and
characterize the equilibrium using optimization theory. Since an equilibrium delivered the
best possible outcome for the representative consumer, they could sidestep the price mech-
anism and nd the equilibrium quantities directly by solving a social planning problem.
Based on these quantities, the equilibrium prices were then easily retrieved from the rst-
order conditions for utility and prot maximization (for details, see Section 3.3.2 below).
All of these simplications have been examined and relaxed in the subsequent literature (see
Section 3.4).
In spite of these drastic simplications, Kydland and Prescott found it necessary to use
numerical analysis to characterize the equilibrium. In so doing, they adapted available in-
sights in numerical analysis to the problem at hand and used computer-aided model solution.
Todays state-of-the-art business cycle models are substantially more complex than that an-
alyzed by Kydland and Prescott, and the numerical analysis of economic models has evolved
into a subeld of its own in economics.
11
Comparison of the model to data was another challenging task. A standard econometric
approach, i.e., to choose the models parameters so as to obtain the best possible t to the
business cycle data, could not really be used due to the complexity of the model. To generate
model output for even one set of parameter values was quite dicult and time-consuming,
11
For references and an overview, see Amman, Kendrick, and Rust (1996).
16
given that it involved numerical solution of a dynamic, stochastic optimization problem. A
search among sets of parameter values in order to obtain the best t was deemed infeasi-
ble. Instead, Kydland and Prescott adopted the method of calibration. They selected
parameter values to match a subset of moments in the data in a way that did not require
solution of the entire stochastic model. In particular, they chose parameter values to match
certain long-run macroeconomic statistics (such as average postwar interest rates and aver-
age capital-output ratios) and microeconomic data (which allowed the parameterization of
preferences).
The idea of choosing parameters in accordance with some basic facts, rather than the
business cycle properties the model was designed to explain, motivated the term calibration.
Clearly, calibration is a simple form of estimation, since the model parameters are chosen in
a well-specied algorithm to t a subset of the overall data; in this case, the estimation is
based on microeconomic and (long-run) macroeconomic data.
12
However, the method was
very practical. It allowed parameterization without solving the full model and it gave clear
and useful directions as to whether specic changes in the model could better explain the
data.
13
Nowadays, given the advances in computer technology and econometric methods,
structural estimation of this class of business cycle models can be carried out and is actively
pursued by business cycle analysts (see Section 3.4 below).
3.3 Examples and applications
We now describe in detail a simple special case of Kydland and Prescotts 1982 setup, while
briey pointing out how their more general model diers from this special case.
14
3.3.1 The setup
Time is discrete and innite: 0, 1, 2, . . . . There is one type of output good at each date, y
t
,
which can be used for consumption or investment: c
t
+ i
t
= y
t
. Capital accumulation obeys
k
t+1
= (1 )k
t
+i
t
: one unit of investment at t adds to the capital stock at t +1 and then
depreciates at a constant rate .
15
Output is produced from capital and labor according
to the production function f: y
t
= f(z
t
, k
t
, l
t
). Here, f is increasing and concave and is
homogeneous of degree one in k and l. The stochastic technology parameter, z
t
, follows an
AR(1) process, z
t+1
= z
t
+
t+1
, where (0, 1) indicates positive autocorrelation and
t
is identically and independently distributed over time with zero mean and positive variance
denoted
2
.
16
12
Early applications of similar methodologies, although typically in static models, can be found in the
empirical literature on international trade (see Shoven and Whalley, 1984). Dynamic models were used in
public nance (see Auerbach and Kotliko, 1987).
13
Long-run averages in Kydland and Prescotts stochastic model are approximately equal to the long-run
averages in the non-stochastic version of the model, which are straightforward to solve for analytically.
14
A version of this special case was examined in a multi-sectoral context by Long and Plosser (1983).
15
Kydland and Prescott assumed a so-called time-to-build technology, according to which it takes longer
than one period for investments to mature into productive capital.
16
The formulation here describes a process where the technology level and population size do not exhibit
growth. This is convenient for purposes of illustration, but it is straightforward to allow for positive trends
in technology and population.
17
There is a large number of (identical) consumers, so each consumer chooses quantities and
takes prices as given. Each consumer is innitely lived and derives utility from consumption
and leisure. Her preferences from the perspective of time 0 are described by
E[

X
t=0
t
u(c
t
, 1 l
t
)],
where , the discount factor, is positive and less than unity reecting a preference for current
consumption, and where u is a strictly increasing and strictly concave function. The total
time endowment is 1 and l
t
is the time spent working. The variables y
t
, c
t
, and so on, are
stochasticthey are driven by the stochastic process for technology z
t
and E denotes the
expectations operator.
The consumers budget constraint reads
c
t
+ k
t+1
= (1 + r
t
)k
t
+ w
t
l
t
,
where r
t
is the market return to capital before depreciation and w
t
the wage rate. Prices in
this model are thus given by the stochastic processes for r
t
and w
t
. The consumer maximizes
the utility function subject to the budget constraint holding at all dates.
Firms maximize prots under perfect competition. Given that the consumers determine
capital accumulation, this implies solving the static optimization problem of choosing k
t
and
l
t
to maximize f(z
t
, k
t
, l
t
)r
t
k
t
w
t
l
t
. Because f is homogeneous of degree one, equilibrium
prots are zero in all time periods..
3.3.2 Analysis
The rst-order necessary conditions for consumer and rm optimization govern how the
model works. Consumers have an intertemporal rst-order condition which determines the
consumption-savings choice:
u
1
(c
t
, 1 l
t
) = E[u
1
(c
t+1
, 1 l
t+1
)(1 + r
t+1
)].
The marginal utility loss of lower consumption in period t must equal the expected discounted
value of the return to higher investment, in terms of next periods utility. Consumers also
have an intratemporal condition which delivers the labor-leisure choice:
u
1
(c
t
, 1 l
t
)w
t
= u
2
(c
t
, 1 l
t
).
The marginal loss of working one more unit of time in terms of lost leisure (the right-hand
side) must thus equal the wage times the marginal utility gain of higher earnings.
Firms prot maximization conditions simply state that marginal products equal real
factor prices: r
t
= f
2
(z
t
, k
t
, l
t
) and w
t
= f
3
(z
t
, k
t
, l
t
). Thus, these prices are easy to nd as a
function of the input quantities and the current value of the technology parameter.
The equilibrium of this model takes the form of stochastic processes for quantities and
prices that satisfy all the equilibrium conditions. As in their work on time consistency,
Kydland and Prescott used recursive analysis whereby equilibrium processes are expressed
18
as functions of the economys state variable, which in this case is (z
t
, k
t
): the current
shock and capital stock. The equilibrium solution for labor will be given by a function h
l
:
l
t
= h
l
(z
t
, k
t
). Thus, this function tells us how much labor will be supplied in equilibrium
for any value of the current shock and any value of the capital stock. We let h
c
denote
the solution for consumption: c
t
= h
c
(z
t
, k
t
). The two functions h
l
and h
c
are sucient for
describing the equilibrium. As an example, if the current state is (z
t
, k
t
), the capital stock
in period t +1 must satisfy k
t+1
= (1 )k
t
+f(z
t
, k
t
, h
l
(z
t
, k
t
)) h
c
(z
t
, k
t
), which is itself a
function only of the current state. As another example, the wage rate in period t must equal
f
3
(z
t
, k
t
, h
l
(z
t
, k
t
)), which is also a function only of the current state.
In general, it is not possible to nd explicit forms for h
l
and h
c
. The next step in
the analysis is therefore to use numerical analysis in order to approximate these functions
under some suitable specic parametric-form assumptions on f and u and on the remaining
parameters of the model.
In a very special case of the model, it is possible to obtain a full analytical (as opposed
to numerical) characterization of the equilibrium. If
u(c, 1 l) = (1 ) log c + log(1 l),
f(z, k, l) = zk
l
1
,
and = 1, then it can be veried that all the equilibrium conditions are satised when labor
supply is constant over time, i.e., h
l
(z
t
, k
t
) = A, and consumption is a constant fraction of
outputso that h
c
(z
t
, k
t
) = Bz
t
k
t
A
1
for some constants A and B.
17
Model output from the specic case which allows a closed-form solution is not directly
comparable to business cycle data, because the assumption of full depreciation implies that
a period in the model is very long. In the data, physical capital depreciates at a rate on the
order of 10 percent per year, and full depreciation occurs after perhaps 20 years. Business
cycle models require the period length to be much shorter, and the typical assumption has
been to let a period be a quarter, or a year. Moreover, although the special case of the
model has a propagation mechanism similar to that discussed in Section 3.1, labor supply
does not respond to shocks. But if the rate of depreciation is assigned a realistic value
(which necessitates using numerical model solution), then a labor supply mechanism with
the properties discussed above appears.
3.3.3 Calibration
We have already touched on the importance of specic parameters for the properties of the
model. As outlined above, Kydland and Prescott did not us an econometric approach, but
instead chose to calibrate their model. That is, they chose all the parameters based on
microeconomic data and long-run macroeconomic data, rather the business cycle frequencies
in the data. How, then, does the procedure of calibration work in the case of this model?
18
As already indicated, the length of a time period has to be set and, typically, a time period in
17
To show this, simply insert the asserted functional forms for consumption and labor into the rst-order
conditions and verify that these conditions hold at all points in time.
18
For details on the calibration undertaken, see, e.g., Prescott (1986).
19
the model is interpreted to represent one quarter. Following Kydland and Prescott, long-run
data (for example, postwar averages from the U.S. if the model is intended to describe this
economy) are then used for the following variables: the average capital-output ratio (about 8,
when output is measured on a quarterly basis, which helps pin down the rate of depreciation
), the average quarterly real interest rate (about 0.015, which pins down the discount rate
), the average capital share of income (about 0.35, which pins down the coecient in the
production function), and the average amount of hours worked (about 0.2 as a fraction of
total hours, which pins down the weight on leisure in utility in the case of the logarithmic
utility function).
In other words, a version of the equilibrium conditions is used where all variables grow at
a constant rate over timeto depict a long-run, or steady-state, situation. The parameter
values are then solved for, given the above long-run statistics. Typically, a somewhat more
general utility function than the logarithmic function is considered. This permits parame-
terization of the curvatures with respect to consumption and leisure such that elasticities
of intertemporal substitution and labor supply conform to those obtained in microeconomic
studies. Finally, the key parameters of the technology process, z, are chosen to conform to
the properties of an estimated process for the Solow residual, matching to its rst-order
autocorrelation and (the variance of) to its variance.
3.3.4 Quantitative evaluation
Once, all of the parameter values have been assigned, the model is fully specied and can be
solved numerically. Given the solution for the equilibrium functions h
l
and h
c
, output and
other variables of interest can be simulated by selecting an initial capital stock and drawing
stochastic shocks from their given statistical distribution, while using the functions h
l
and h
c
to generate series for all variables of interest. These series can then be compared to business
cycle data.
As an illustration, consider the setup exemplied above, which is also analyzed in Cooley
and Prescott (1995).
19
The calibration is carried out as outlined, where the functional form
for utility is u(c, l) =
(c
1
l
)
1
1
1
. This form implies that the consumers attitudes toward
riskas dened by the coecient of relative risk aversionis constant over time and equal
to . As in the general discussion above, is determined on the basis of time-use studies.
The parameter is chosen on the basis of microeconomic studies of risk-taking. The studies
discussed in Prescott (1986) suggest a value for around 1, which in fact implies that the
utility function reduces to the logarithmic function above. Next, in line with the results of
growth accounting, the process for technology shocks is taken to have a serial correlation
coecient ( above) of 0.95 and a standard deviation ( above) of 0.7 percent.
The data are quarterly data from the postwar U.S. economy.
20
Prior to comparing the
simulated output from the model to data, the business cycle component of the data has
to be extracted. Business cycles are normally thought of as uctuations around a growth
19
The framework here and in Cooley and Prescott (1995) are identical except that the latter study allows
long-run growth in population and technology. Because these factors only have a marginal inuence on the
results, they are omitted here.
20
The sample begins in the rst quarter of 1954 and ends with the second quarter of 1991.
20
Some statistics from the model (data)
Output Consumption Investment Hours Labor productivity
Percent standard deviations
1.35 (1.72) 0.33 (0.86) 5.95 (8.24) 0.77 (1.59) 0.61 (0.55)
Correlation with output
1 (1) 0.84 (0.83) 0.99 (0.91) 0.86 (0.99) 0.98 (0.03)
Table 1
path that occur with a frequency of three-ve years.
21
Thus, the data are de-trended, or
ltered, so as to suppress their low-frequency (trend) components, while retaining those
in the business cycle range. This can be accomplished in several ways. Cooley and Prescott
(1995) used a form of dierence lter.
22
Some key statistics of the de-trended data from
their study and the associated model output are summarized in Table 1.
23
The table shows that the variation in output in the model is somewhat lower than in the
U.S. data, but the discrepancy is not large. The model predicts that consumption varies much
less and investment much more than output. This is a result of consumption smoothing.
The variation in hours in the model is less than half of that in the data. This discrepancy
is largely due to the uctuation in the number of employed workers in the data; the simple
model here focuses on variations in the number of hours worked per worker (the intensive
margin) and abstracts from variations in labor-force participation and unemployment (the
extensive margin). Productivity in the model varies somewhat more than the measure of
productivity found in the data, which is based on average compensation per hour.
The table also shows strikingly high correlations between output and other macroeco-
nomic variables. The reason is that there is only one shockone source of uncertaintyin
the model. The correlations between consumption and output and between investment and
output are quite similar in the model and in the data. As for the correlation between hours
and output, the model does not generate a co-movement as high as that observed in the U.S.
economy. This is another consequence of the rudimentary modeling of labor supply in this
21
See, e.g., Burns and Mitchell (1946).
22
The specic lter used is the one introduced in Hodrick and Prescott (1980).
23
The model was simulated 100 times, each with a sample length of 150 time periods. The numbers in
the table are averages across simulations. The series for output, consumption, and investment are based
on real (1982$) quantities; output is GNP, consumption is consumption on nondurables and services, and
investment is gross private domestic investment. Hours are total hours of work based on the U.S. Household
Survey. As in the model, labor productivity equals average total compensation per hour, measured from the
U.S. National Income and Product Accounts.
21
simple model. Hourly compensation is around zero in the data whereas the model generates
a correlation close to one. This again reects that the model has only one shock and that
labor supply operates only on the intensive margin.
The implications of the model for the dynamic correlations of dierent variables over the
cycle can also be compared to data. For example, the correlation between output in adjacent
quarters is 0.70 in the model; in the data, it is 0.85.
24
3.4 Subsequent research
Kydland and Prescotts methodology has been broadly adopted, and adapted to a variety
of frameworks for analyzing business cycles. The vast literature that has followed may be
organized into ve dierent categories: (1) extensions aimed at evaluating the consequences
when the stark simplifying assumptions, such as that of only one type of consumer or ab-
stracting from monetary issues, are relaxed; (2) analyses relying on other impulses, such as
monetary shocks or international shocks to the terms of trade; (3) studies considering dif-
ferent propagation mechanisms, such as those implied by imperfections in credit, labor and
goods markets; (4) investigations focusing on improved measurement of technology shocks
and other variables; and (5) studies aimed at improving the empirical analysis in the direction
of structural estimation.
Category 1. A number of simplifying assumptions made by Kydland and Prescott have
been relaxed and the main results have been found to be quite robust. These include homo-
geneity of rms and households (captured by the assumptions of a representative rm and a
representative household, as opposed to heterogeneity of rms and households in dierent di-
mensions), perfect altruism across generations, the absence of idiosyncratic and uninsurable
consumer risk, the aggregation of goods into a single composite good, perfect substitutabil-
ity between consumption and investment goods, perfect competition in goods markets, no
role for money (as opposed to, say, a demand for money based on so-called cash-in-advance
constraints), the closed economy, and the exogeneity of technology improvements.
25
The propagation and serial correlation properties of the model are aected when allowing
for costs of changing/reallocating the capital stock and the labor force over time and across
sectors. These properties are also aected when nonconvex technologies, such as indivisi-
bilities in labor supply, are introduced to generate adjustment on the extensive margin of
employment, which is missing in the simple model of Section 3.3. Extensions of the model
to include costs of rapidly adjusting capital and labor in response to shocks help replicate
certain asymmetries that characterize the business-cycle data. For instance, recessions are
steeper and shorter than booms, estimates of responses to technology shocks suggest that
hours worked respond strongly but with a lag, and the components underlying the employ-
ment processjob creation and job destructionhave very dierent time-series properties,
with job destruction far more variable than job creation (see, e.g., Gilchrist and Williams,
24
For a more complete evaluation of the models predictions for the structure of leads and lags, see Cooley
and Prescott (1995).
25
For an early survey of the literature, see Cooley (1995); for a more recent account, see King and Rebelo
(2000).
22
2000 and Campbell and Fisher, 2000). Moreover, consideration of variable utilization of fac-
tors of production has been shown to signicantly increase the amplitude of the economys
response to shocks (for a general discussion of this issue, see King and Rebelo, 2000).
Category 2. Many studies have suggested alternatives to aggregate technology shocks
as the source of cycles. Real impulses include shocks to relative world prices for raw materi-
als, shocks to the mechanism for wage setting implying changes in rms costs (cost-push
shocks), shocks to scal policy and government expenditures, or technology shocks that pri-
marily aect a sector of the economy, rather than aggregate output. Demand shocks include
shocks to consumer preferences, e.g., bursts of impatience, even though the diculty of
empirically measuring the size of such shocks makes them hard to assess. Preference shocks
are perhaps the closest counterpart to those which Keynes perceived as driving the economy,
although they may not represent animal spirits of investors or consumer condence.
One branch of the literature deals with shocks to information, i.e., news about upcoming
events, such as technology improvement.
Yet another branch of the literature examines business cycles from an international per-
spective. Some researchers have studied the reaction of small open economies to international
shocks, such as terms-of-trade or interest-rate shocks; see Mendoza (1991) and, for an early
application to the case of Sweden, Lundvik (1992). Other researchers have instead consid-
ered the transmission of business cycles across similar economies; for the rst contribution
to this line of research, see Backus, Kehoe, and Kydland (1992).
Category 3. Many extensions have been motivated by the belief that a macroeconomic
theory where the basic ingredients of the business cycle can be captured in a perfect-market
framework misses important real-world features. Naturally, labor-market frictions that give
rise to equilibrium unemployment expand the domain of the model, but they do not nec-
essarily imply very dierent time-series properties for output and investment. The same
applies when monopolistic elements are added to competition in goods markets.
26
Despite
the added frictions, the basic methodology in this research is typically in line with Kydland
and Prescotts approach, however. For example, the models of credit-market imperfections
in Bernanke and Gertler (1989) or Kiyotaki and Moore (1995) are explicitly founded on
microeconomics and derive market frictions from rst principles. Similarly, search and
eciency-wage models of unemployment make assumptions about agents behavior derived
from microeconomic underpinnings. Some of these models feature qualitatively dierent
propagation mechanisms than those found in Kydland and Prescotts work. Thus, Kiyotaki
and Moore (1995) nd pronounced cyclical responses of output to single shocks, whereas
Kydland and Prescotts theory tends to give monotonic responses.
Category 4. Renewed measurement of technology shocks has been undertaken using
several methods. One such method builds on Solows growth accounting (used for example in
Prescott (1986) and Kydland and Prescott (1988)), but substantially relaxes the underlying
assumptions. First, accounting has been conducted on a more disaggregated level, i.e.,
relaxing the assumption of an aggregate production function. Second, the assumption that
inputs are well measured has been relaxed by allowing (unobserved) factor utilization to
26
For search frictions, see Merz (1995) or Andolfatto (1996) and for an eciency-wage formulation, see
Danthine and Donaldson (1995). For an early analysis of a real business cycle model with monopolistic
competition, see Hornstein (1993).
23
vary and by econometrically estimating the movement of factor utilization over the cycle
using instrumental variables techniques. Third, the assumption of constant returns to scale
and perfect competition has been relaxed: based on estimating the size of markups over
marginal costs, the same type of accounting method can still be used. These improvements
on Kydland and Prescotts initial measures are discussed in Basu and Fernald (1997, 2000).
A summary of the ndings is that estimates of short-run uctuations in technology remain
signicant; whether they have the specic eect on the economy predicted by Kydland and
Prescotts original model is a more open question.
Another, and quite dierent, approach to measuring technology shocks is to estimate a
stochastic, vector-autoregressive (VAR) system of aggregate time series containing output,
measured labor productivity, and a set of other variables. In the VAR system, the shocks
driving the observed variables are then classied as supply (technology) shocks if they have
a permanent impact on labor productivity, which technology shocks should have; any re-
maining shocks are classied as demand shocks (see Blanchard and Quah, 1989). For this
approach, see, e.g., Gali (1999) and Fisher (2002). The results of this approach appear sen-
sitive to specication details, but call into question the generality of Kydland and Prescotts
original ndings.
Category 5. Econometric estimation has gradually been adopted. As mentioned earlier,
traditional econometric estimation of a fully micro-founded business cycle model was not re-
ally an option when Kydland and Prescott made their original contribution, and may not
have been very productive in the early stages of the research program. But over time, as the
theory has become richer, computers have become more powerful, and econometric methods
have become more advanced, the situation has changed (category (5) above). Econometric
estimation of the stochastic growth model began with Altug (1989), but was undertaken in a
linearized version of the business cycle model. Later developments estimated key parts of the
modelsuch as rst-order conditions for savings and labor supplyusing so-called general-
ized method of moments estimators (see, e.g., Hansen, 1982). Full structural estimation of
stochastic nonlinear dynamic general equilibrium models has now been undertaken as well.
Smets and Wouters (2003) provide a recent example of successful full-edged (Bayesian)
estimation of a model for the Euro area based on Kydland and Prescotts work.
An important new literature, which deserves special mention, encompasses several of
the extensions discussed above. Commonly referred to as new-Keynesian business cycle
research, this literature examines monetary models of business cycles based on frictions in the
process of price and/or wage adjustment (see e.g., Rotemberg and Woodford, 1997, Clarida,
Gali, and Gertler, 2000, and Dotsey, King, and Wolman, 1999). These new-Keynesian
models are built around a core very similar to Kydland and Prescotts model, but they also
include microeconomic assumptions on the costs of changing prices for rms, which typically
are assumed to interact under monopolistic competition. Price-setting and/or wage-setting
decisions are explicitly modeled as forward-looking and based on rational expectations. This
renders the analytical structure similar to that in Kydland and Prescotts original work and
the new-Keynesian researchers have also borrowed their analytical tools. The resulting
models can give rise to a Phillips-curve. Monetary shocks produce potentially large eects
on output, and monetary policy can produce, or stabilize, short-run uctuations. Models
of this type have also proved very useful for more complex analyses of the time consistency
24
problems of monetary policy uncovered by Kydland and Prescott.
New-Keynesian frameworks have been applied to both positive and normative analysis
of dierent monetary rules and institution design. Since the underlying theory rests on
explicit microeconomic assumptions, the evaluation of policy experiments is straightforward:
simulating the model for dierent policy scenarios, the resulting welfare levels for dierent
agents can easily be compared. Thus, the modeling allows not only qualitative, but also
quantitative welfare statements. Such policy evaluations are also attractive in view of the
Lucas critique. Since the models are formulated on the basis of deep parameters, they should
be more robust to the conduct of policy than aggregate reduced-form relations. In sum, the
new-Keynesian approach has synthesized earlier Keynesian analysis with the real business
cycle analysis originating in Kydland and Prescotts work.
3.5 Broader impact
Kydland and Prescotts 1982 paper transformed the academic research on business cycles.
Extensions and renements in the subsequent literature improved the ability of the original
model to match the macroeconomic data and allowed meaningful analyses of macroeco-
nomic policy. Models used in actual policy contexts have increasingly adopted Kydland and
Prescotts methodology. Gradually, many models used in policy organizations and central
banks have come to incorporate microeconomic foundations in the form of rational savings
and labor supply behavior combined with rational expectations. One typical procedure has
been to formulate deterministic versions of Kydland and Prescotts modelto be used for
medium-term counterfactual analysesand to add an ad-hoc stochastic structure that allows
for rich short-run dynamics. Today, some organizations have complete operational versions
of Kydland-Prescott-style business cycle models incorporating full short-term dynamics.
27
An alternativealthough closely relatedapproach has been to look for ways of summa-
rizing full-edged Kydland-Prescott models using (approximate) reduced-form systems that
are easy to analyze and therefore convenient tools in policy analysis (see, e.g., Woodford,
2003 and Schmitt-Groh and Uribe, 2003). Computer technology has dened the frontier
of Kydland and Prescotts research program, and the rapid recent advances in this technol-
ogy have greatly expanded the ability to solve and estimate highly complex versions of new
business cycle theories.
3.6 Related literature
The core of the real business cycle model is a neoclassical growth model with optimal sav-
ings decisions. Cass (1965) and Koopmans (1965) added optimal savings decisions to Solows
neoclassical setup, although they did not model labor supply. Stochastic shocks were intro-
duced into the optimal growth model by Brock and Mirman (1972), but their model was not
given an equilibrium interpretation and was not used to analyze technology-driven short-run
cycles. In their early contribution to the literature initiated by Kydland and Prescott, Long
and Plosser (1983) examined co-movement across sectors due to aggregate technology shocks
and coined the term real business cycles. Bruno and Sachs (1979) analyzed supply shocks,
27
For an application to short-term forecasting, see del Negro and Schorfheide (2004).
25
but not in the context of a fully dynamic and stochastic macroeconomic model based on mi-
croeconomic foundations. More generally, Kydland and Prescotts approach to business cycle
analysis is linked to early articles by Frisch (1933) and Slutsky (1937), which showed how an
economys adjustment to a sequence of random shocks can give rise to cyclical uctuations
reminiscent of business cycles.
4 Recommended reading
Although rather technically demanding, Kydland and Prescotts original articles from 1977
and 1982 are highly recommended. Several of the early articles on time consistency and
macroeconomic policy are reprinted in Persson and Tabellini (1994). For a survey of this
research, see Drazen (2000). For readings on Kydland and Prescotts business cycle theory,
see the volume by Cooley (1995), which contains a series of surveys on dierent aspects of
real business cycles. King and Rebelo (2000) provide an up-to-date comprehensive review,
while Prescott (1986) gives a non-technical introduction to the topic.
5 References
Alesina, A. and L. Summers (1993),Central bank independence and macroeconomic perfor-
mance: Some comparative evidence, Journal of Money, Credit and Banking, 25, 151-162.
Altug, S. (1989), Time-to-build and aggregate uctuations: Some new evidence, Interna-
tional Economic Review, 10, 889-920.
Amman, H., D. Kendrick and J. Rust (eds.) (1996), Handbook of Computational Economics,
North-Holland: Amsterdam.
Andolfatto, D. (1996), Business cycles and labor market search, American Economic Re-
view, 86, 112-132.
Auerbach, A. and L. Kotliko (1987), Dynamic Fiscal Policy, Cambridge University Press:
Cambridge.
Auernheimer, L. (1974), The honest governments guide to the revenue from the creation
of money, Journal of Political Economy, 82, 598-606.
Backus, D. and J. Drill (1985), Rational expectations and policy credibility following a
change in regime, Review of Economic Studies, 52, 211-221.
Backus, D., P. Kehoe, and F. Kydland (1992), International real business cycles, Journal
of Political Economy, 101, 745-775.
Barro, R. and D. Gordon (1983a), A positive theory of monetary policy in a natural-rate
model, Journal of Political Economy, 91, 589-610.
26
Barro, R. and D. Gordon (1983b), Rules, discretion and reputation in a model of monetary
policy, Journal of Monetary Economics, 12, 101-120.
Basu, S. and J. Fernald (1997), Returns to scale in U.S. production: Estimates and impli-
cations, Journal of Political Economy, 105, 249-283.
Basu, S. and J. Fernald (2000), Why is productivity procyclical? Why do we care?, Federal
Reserve Bank of Chicago Working Paper 2000-11.
Bernanke, B. and M. Gertler (1989), Agency costs, net worth, and business uctuations,
American Economic Review, 79, 14-31.
Blanchard, O. and D. Quah (1989), The dynamic eects of aggregate demand and supply
disturbances, American Economic Review, 79, 654-673.
Brock, W. and L. Mirman (1972), Optimal economic growth and uncertainty: The dis-
counted case, Journal of Economic Theory, 4, 479-515.
Bruno, M. and J. Sachs (1979), Supply vs. demand approaches to the problem of staga-
tion, NBER Working Paper #382.
Buchanan, J. (1975), The Samaritans dilemma, in E. Phelps (red.), Altruism, Morality
and Economic Theory, Russell Sage Foundation.
Bulow, J. (1982), Durable goods monopolists, Journal of Political Economy, 90, 314-32.
Burns, A. and A. Mitchell (1946), Measuring Business Cycles, National Bureau of Economic
Research: New York.
Calvo, G. (1978), On the time consistency of optimal policy in a monetary economy,
Econometrica, 46, 1411-1428.
Campbell, J. and J. Fisher (2000), Aggregate employment uctuations with microeconomic
asymmetries, American Economic Review, 90, 1323-1345.
Cass, D. (1965), Optimum growth in an aggregative model of capital accumulation, Review
of Economic Studies, 32, 233-240.
Clarida, R., J. Gali and M. Gertler (1999), The science of monetary policy: A new Keyne-
sian perspective, Journal of Economic Literature, 37, 1661-1707.
Clarida, R., J. Gali and M. Gertler (2000), Monetary policy rules and macroeconomic
stability: Evidence and some theory, Quarterly Journal of Economics, 115, 147-180.
Coase, R. (1972), Durability and monopoly, Journal of Law and Economics, 15, 143-149.
Cooley, T. (ed.) (1995), Frontiers of Business Cycle Research, Princeton University Press:
Princeton.
27
Cooley, T. and E. Prescott (1995), Economic growth and business cycles, in Cooley, T.
(ed.), Frontiers of Business Cycle Research, Princeton University Press: Princeton, 1-38.
Cukierman, A. (1992), Central Bank Strategy, Credibility and Independence: Theory and
Evidence, MIT Press: Cambridge, MA.
Danthine, J.P. and Donaldson, J. (1995), Non-Walrasian economies, in T. Cooley (ed.),
Frontiers of Business Cycle Analysis, Princeton University Press: Princeton, 217-242.
Debreu, G. (1959), Theory of Value, An Axiomatic Analysis of Economic Equilibrium, Wiley:
New York.
del Negro, M. and F. Schorfheide (2004), Priors from general equilibrium models for VARs,
forthcoming, International Economic Review.
Dotsey, M., R. King, and A. Wolman (1999), State dependent pricing and the general
equilibrium dynamics of money and output, Quarterly Journal of Economics, 114, 655-690.
Drazen, A. (2000), Political Economy in Macroeconomics, Princeton University Press: Prince-
ton.
Elster, J. (1977), Ulysses and the sirens: A theory of imperfect rationality, Social Science
Information, 16, 469-526.
Fischer, S. (1980), Dynamic inconsistency, cooperation, and the benevolent dissembling
government, Journal of Economic Dynamics and Control, 2, 93-107.
Fisher, J. (2002), Technology shocks matter, Federal Reserve Bank of Chicago Working
Paper 2002-14.
Friedman, M. (1968), The role of monetary policy, American Economic Review, 58, 1-17.
Frisch, R. (1933), Propagation problems and impulse problems in dynamic economies, in
Economic Essays in Honour of Gustav Cassel, Allen and Unwin: London.
Gali, J. (1999), Technology, employment, and the business cycle: Do technology shocks
explain aggregate uctuations?, American Economic Review, 89, 249-271.
Gilchrist, S. and J. Williams (2000), Putty-clay and investment: a business cycle analysis,
Grilli, V., D. Masciandro and G. Tabellini (1991), Political and monetary institutions and
public nance in the industrial countries, Economic Policy, 13, 342-392.
Grossman, G. and E. Helpman (2001), Special Interest Politics, MIT Press: Cambridge, MA.
and London.
28
Gul, F., H. Sonnenschein, and R. Wilson (1986), Foundations of dynamic monopoly and
the Coase conjecture, Journal of Economic Theory, 39, 155-190.
Hansen, L. (1982), Large sample properties of generalized method of moments estimators,
Hodrick, R. and E. Prescott (1980), Postwar U.S. business cycles: An empirical investiga-
tion, Journal of Money, Credit and Banking, 29:1, 1-16.
Hornstein, A. (1993), Monopolistic competition, increasing returns to scale, and the impor-
tance of productivity shocks, Journal of Monetary Economics, 31, 299-316.
Keynes, J. M. (1936), The General Theory of Employment, Interest and Money. McMillan:
London.
King, R. and S. Rebelo (2000), Resuscitating real business cycles, in J. Taylor and M.
Woodford (eds.), Handbook of Macroeconomics, North-Holland: Amsterdam, 927-1007.
Kiyotaki, N. and J. Moore (1995), Credit cycles, Journal of Political Economy, 105, 211-
248.
Koopmans, T. (1965), On the concept of optimal economic growth, The Economic Ap-
proach to Development Planning, North-Holland: Amsterdam.
Kydland, F. and E. Prescott (1977), Rules rather than discretion: The inconsistency of
optimal plans, Journal of Political Economy, 85, 473-490.
Kydland, F. and E. Prescott (1982), Time to build and aggregate uctuations, Economet-
rica, 50, 1345-1371.
Kydland, F. and E. Prescott (1988), The workweek of capital and its cyclical implications,
Journal of Monetary Economics, 21, 343-360.
Long, J. B. and C. Plosser (1983), Real business cycles, Journal of Political Economy, 91,
39-69.
Lucas, R. (1972), Expectations and the neutrality of money, Journal of Economic Theory,
4, 103-124.
Lucas, R. (1973), Some international evidence on output-ination tradeos, American
Economic Review, 63, 326-334.
Lucas R. (1976), Econometric policy evaluation: A critique, Journal of Monetary Eco-
nomics, supplement, 19-46.
Lucas, R. and N. Stokey (1983), Optimal scal and monetary policy in an economy without
capital, Journal of Monetary Economics, 12, 55-94.
29
Lundvik, P. (1992), Foreign demand and domestic business cycles: Sweden 1891-1987,
Chapter 3 in Business Cycles and Growth, Monograph Series No. 22, Institute for Interna-
tional Economic Studies, Stockholm University.
Maskin, E. and J. Tirole (2001), Markov perfect equilibrium, I: Observable actions, Journal
of Economic Theory, 100, 191-219.
Mendoza, E. (1991), Real business cycles in a small open economy, American Economic
Review 81, 797-818.
Merz, M. (1995), Search in the labor market and the real business cycle, Journal of
Monetary Economics, 36, 269-300.
Nash, J. (1950), Non-cooperative Games, Ph.D. dissertation, Department of Mathematics,
Princeton University: Princeton.
Persson, T. and G. Tabellini (1990), Macroeconomic Policy, Credibility and Politics, Har-
wood Academic Publishers: Chur.
Persson, T. and G. Tabellini (1994), Monetary and Fiscal Policy, Vol. 1 Credibility, MIT
Press: Cambridge, MA.
Persson, T. and G. Tabellini (2000), Political Economics: Explaining Economic Policy, MIT
Press: Cambridge, MA..
Phelps, E. (1967), Phillips curves, expectations of ination and optimal unemployment over
time, Economica, 34, 254-81.
Phelps, E. (1968), Money-wage dynamics and labor-market equilibrium, Journal of Polit-
ical Economy, 76, 678-711.
Phelps, E. and R. Pollak (1968), On second-best national saving and game-equilibrium
growth, Review of Economic Studies, 35, 185-199.
Prescott, E. (1986), Theory ahead of business cycle measurement, Federal Reserve Bank
of Minneapolis Quarterly Review, 10, Fall, 9-22.
Rogo, K. (1985), The optimal degree of precommitment to an intermediate monetary
target, Journal of International Economics, 18, 1169-1190.
Rotemberg, J. and M. Woodford (1997), An optimization-based econometric framework for
the evaluation of monetary policy, NBER Macroeconomics Annual, 12, 297-346.
Sargent, T. (1977), The demand for money during hyperinations under rational expecta-
tions, International Economic Review, 18, 59-92.
30
Sargent, T. (1978), Estimation of dynamic labor demand schedules under rational expec-
tations, Journal of Political Economy, 86, 1009-1044.
Sargent, T. (1979), A note on maximum likelihood estimation of the rational expectations
model of the term structure, Journal of Monetary Economics, 5, 133-143.
Schmitt-Groh, S. and M. Uribe (2003), Simple optimal implementable monetary and scal
rules, NBER working paper #10253.
Selten, R. (1965), Spieltheoretische Behandlung eines Oligopolmodells mit Nachtragefragheit,
Zeitschrift fr die gesamte Staatswissenschaft, 121, 301-324 and 667-689.
Shoven, J. and J. Whalley (1984), Applied general-equilibrium models of taxation and
international trade: An introduction and survey, Journal of Economic Literature, 22, 1007-
1051.
Slutsky, E. (1937), The summation of random causes as the source of cyclic processes,
Smets, F. and R. Wouters (2003), An estimated dynamic stochastic general equilibrium
model of the Euro area, Journal of the European Economic Association, 1, 1123-1175.
Solow, R. (1956), A contribution to the theory of economic growth, Quarterly Journal of
Economics, 70, 65-94.
Solow, R. (1957), Technical change and the aggregate production function, Review of
Economics and Statistics, 39, 312-320.
Stokey, N. (1981), Rational expectations and durable goods pricing, Bell Journal of Eco-
nomics, 12, 112-128.
Strotz, R. (1956), Myopia and inconsistency in dynamic utility maximization, Review of
Economic Studies, 23, 165-180.
Svensson, L. (1997), Optimal ination targets, conservative central bankers and linear in-
ation contracts, American Economic Review, 87, 98-114.
Tabellini, G. (1985), Accommodative monetary policy and central bank reputation, Gior-
nali degli Economisti e Annali di Economia, 44, 389-425.
Walsh, C. (1995), Optimal contracts for central bankers, American Economic Review, 85,
150-67.
Woodford, M. (2003), Interest and Prices: Foundations of a Theory of Monetary Policy,
Princeton University Press: Princeton.
31
8 October 2003

Phone: +46 8 673 95 00, Fax: +46 8 15 56 70, E-mail: info@kva.se, Website: ww.kva.se

Time-series Econometrics:
Cointegration and Autoregressive
Conditional Heteroskedasticity

Time-series Econometrics:
Cointegration and Autoregressive Conditional Heteroskedasticity
1. Introduction
Empirical research in macroeconomics as well as in nancial economics is largely
based on time series. Ever since Economics Laureate Trygve Haavelmos work
it has been standard to view economic time series as realizations of stochastic
processes. This approach allows the model builder to use statistical inference in
constructing and testing equations that characterize relationships between eco-
nomic variables. This years Prize rewards two contributions that have deep-
ened our understanding of two central properties of many economic time series
nonstationarity and time-varying volatility and have led to a large number of
applications.
Figure 1.1: Logarithm (rescaled) of the Japanese yen/US dollar exchange rate (de-
creasing solid line), logarithm of seasonally adjusted US consumer price index (increas-
ing solid line) and logarithm of seasonally adjusted Japanese consumer price index
(increasing dashed line), 1970:1 2003:5, monthly observations
Nonstationarity, a property common to many macroeconomic and nancial
time series, means that a variable has no clear tendency to return to a constant
value or a linear trend. As an example, Figure 1.1 shows three monthly series:
the value of the US dollar expressed in Japanese yen, and seasonally adjusted
consumer price indices for the US and Japan. None of these series, of which
the price series are considerably smoother than the exchange rate, seems to be
stationary, i.e., return to a xed value or uctuate around a linear trend (in
which case the deviations from trend are stationary). Other aggregate variables,
such as gross national product, consumption, employment, and asset prices share
this property. It is therefore proper to assume that they have been generated by
nonstationary processes and follow stochastic trends.
An important objective of empirical research in macroeconomics is to test
hypotheses and estimate relationships, derived from economic theory, among such
aggregate variables. The statistical theory that was applied well into the 1980s
in building and testing large simultaneous-equation models was based on the
assumption that the variables in these models are stationary. The problem was
that statistical inference associated with stationary processes is no longer valid if
the time series are indeed realizations of nonstationary processes.
This diculty was not well understood among model builders three decades
ago. This is no longer the case, and Clive Granger can be credited with this
change. He has shown that macroeconomic models containing nonstationary
stochastic variables can be constructed in such a way that the results are both
statistically sound and economically meaningful. His work has also provided the
underpinnings for modeling with rich dynamics among interrelated economic vari-
ables. Granger has achieved this breakthrough by introducing the concept of
cointegrated variables, which has radically changed the way empirical models of
macroeconomic relationships are formulated today.
The second central property of economic time series, common to many -
Figure 1.2: Daily returns of the yen/dollar exchange rate and the corresponding 20-day
moving average of the squared changes, 1986-1995
2
nancial time series, is that their volatility varies over time. Consider a nancial
return series such as the rate of change of a daily exchange rate or stock index.
As an example, the upper panel of Figure 1.2 contains the rst dierence of the
series in Figure 1.1 measured at the daily frequency. The lower panel, which
displays a 20-day (four trading weeks) moving average of the squared returns,
clearly illustrates how high-volatility periods alternate with periods of relative
calm.
Volatility of returns is a key issue for researchers in nancial economics and
analysts in nancial markets. The prices of stocks and other assets depend on
the expected volatility (covariance structure) of returns. Banks and other nan-
cial institutions make volatility assessments as a part of monitoring their risk
exposure. Up until the 1980s, both researchers and practitioners in nancial
markets used models in which volatility was assumed constant over time. As
Figure 1.2 illustrates, however, volatility may vary considerably over time: large
(small) changes in returns are followed by large (small) changes. The modeling
and forecasting of volatility are therefore crucial for nancial markets.
Research on volatility models was initiated by Robert Engle who, in the early
1980s, developed a new concept that he termed autoregressive conditional het-
eroskedasticity, and acronymized ARCH. Since their advent, models built around
this concept have become an indispensable tool for nancial analysts, bankers
and fund managers throughout the world. For two decades, Robert Engle has
remained at the forefront of research on volatility modelling and made several
outstanding contributions in this area.
2. Cointegrated economic variables
Macroeconomists build time-series models for testing economic theories, for fore-
casting, and for policy analysis. Such models are constructed and applied by
economists at universities, economic research institutes and central banks. There
is a long tradition of building large macroeconomic models with hundreds of equa-
tions and variables. More recently, small models with only a handful of equations
and variables have become more common. Since many of the time series model
builders use are best viewed as nonstationary, exploiting such series requires both
a new approach and statistical inference dierent from the traditional inference
developed for applications to stationary series.
In this section, we describe Clive Grangers contributions that lead up to the
concept of cointegration and its applications. We begin by dening the concept
and the statistical theory related to it, including the so-called Granger represen-
tation theorem. This is followed by a description of the two-step method used
to test for cointegrating relationships and estimate equation systems with cointe-
grated variables. A number of extensions of the basic concept of cointegration are
briey highlighted. We end by discussing applications. Empirical research on the
3
purchasing power parity (PPP) hypothesis is used to illustrate how cointegration
may not only change empirical analysis, but also give it a new dimension.
2.1. Cointegration: basic denition
Returning to the time series in Figure 1.1, it is appropriate to assume that they
have been generated by nonstationary stochastic processes. For a long time it
was common practice to estimate equations involving nonstationary variables in
macroeconomic models by straightforward linear regression. It was not well un-
derstood that testing hypotheses about the coecients using standard statistical
inference might lead to completely spurious results. In an inuential paper, Clive
Granger and his associate Paul Newbold (Granger and Newbold (1974)) pointed
out that tests of such a regression may often suggest a statistically signicant
relationship between variables where none in fact exists. Granger and Newbold
reached their conclusion by generating independent nonstationary series, more
precisely random walks.
1
They regressed these series on each other and observed
the value of the t-statistic of the coecient estimate calculated under the as-
sumption that the true value of the coecient equals zero. Despite the fact that
the variables in the regression were independent, the authors found that the null
hypothesis of a zero coecient was rejected much more frequently than stan-
dard theory predicts. At the same time, they observed that the residuals of the
estimated equation displayed extremely strong positive autocorrelation.
2
These results indicated that many of the apparently signicant relationships
between nonstationary economic variables in existing econometric models could
well be spurious. This work formed an initial step in Grangers research agenda
of developing methods for building more realistic and useful econometric models.
Statisticians working with time-series models suggested a simple solution to
the spurious regressions problem. If economic relationships are specied in
rst dierences instead of levels, the statistical diculties due to nonstationary
variables can be avoided because the dierenced variables are usually stationary
even if the original variables are not. Economic theories, however, are generally
formulated for levels of variables rather than for dierences. For example, theo-
ries of consumption postulate a relationship between the levels of consumption,
income, wealth and other variables and not their growth rates. A model re-
lating the rst dierences of these variables would typically not make full use
of these theories. An alternative approach would involve removing a linear time
trend from the variables and specifying the empirical relationship between them
1
Assume that {
}, = 0, 1, 2, ..., is a sequence of independent stochastic variables with

mean zero and variance
2
and let
t
=
t
=0

. Sequence {
t
}, t = 0, 1, 2, ..., is then a random
walk (without drift).
2
Granger and Newbold (1974) reached their conclusions by simulation. The asymptotic
distribution theory valid for their experiment was worked out more than a decade later by
Phillips (1986). A compact presentation of these developments can be found in Granger (2001).
4
using detrended variables. Removing (separate) time trends assumes, however,
that the variables follow separate deterministic trends, which does not appear
realistic, given the awkward long-run implications. Dynamic econometric models
based on linearly detrended variables may thus be able to characterize short-term
dynamics of economic variables but not their long-run relationships. The same
is true for models based solely on rst dierences.
Clive Grangers solution to this problem may be illustrated by the simplest
possible regression equation:
y
t
= + x
t
+
t
, (2.1)
where y
t
is the dependent variable, x
t
the single exogenous regressor, and {
t
}
a white-noise, mean-zero sequence. Granger (1981) argues that in order to be
meaningful, an equation has to be consistent in the sense that a simulation
of the explanatory right-hand side should produce the major properties of the
variable being explained. For example, if y
t
is a seasonal variable, then x
t
has
to be seasonal, if
t
is to be white noise. To develop the idea further, Granger
(1981) dened the concept of degree of integration of a variable. If variable z
t
can be made approximately stationary by dierencing it d times, it is called
integrated of order d, or I(d). Weakly stationary random variables are thus I(0).
Many macroeconomic variables can be regarded as I(1) variables: if z
t
I(1),
then z
t
I(0). Note that I(1) variables dominate I(0) variables; in a linear
combination of variables the variation of the former overwhelms the variation of
the latter. To illustrate, if z
t
I(1) and w
t
I(0), then z
t
+ w
t
I(1).
Consider again equation (2.1) and assume that both x
t
I(1) and y
t

I(1). Then, generally y
t
x
t
I(1) as well. There is, however, one important
exception. If
t
I(0), then y
t
x
t
I(0), i.e., the linear combination y
t
x
t
has the same statistical properties as an I(0) variable. There exists only one such
combination so that coecient is unique.
3
In this special case, variables x
t
and
y
t
are called cointegrated. More generally, if a linear combination of a set of I(1)
variables is I(0), then the variables are cointegrated. This concept, introduced
in Granger (1981), has turned out to be extremely important in the analysis of
nonstationary economic time series. A generalization to I(d) variables, where d
is no longer an integer, is also possible, in which case the linear combination of
cointegrated variables has to be I(d d
0
), where d
0
> 0.
3
Uniqueness can be shown as follows. Suppose there are two cointegrating relations between
the I(1) variables y
t
and x
t
:
y
t
=
j
x
t
+
jt
, j = 1, 2,
1
=
2
.
Subtracting the second from the rst yields
(
2

1
)x
t
=
1t

2t
.
The left-hand side of this equation is I(1) whereas the right-hand side as a dierence of two
I(0) variables is I(0). This is a contradiction unless
1
=
2
, in which case
1t
=
2t
.
5
The importance of cointegration in the modeling of nonstationary economic
series becomes clear in the so-called Granger representation theorem, rst formu-
lated in Granger and Weiss (1983). In order to illustrate this result, consider the
following bivariate autoregressive system of order p :
x
t
=
p
j=1
1j
x
tj
+
p
j=1
1j
y
tj
+
1t
y
t
=
p
j=1
2j
x
tj
+
p
j=1
2j
y
tj
+
2t
,
where x
t
and y
t
are I(1) and cointegrated, and
1t
and
2t
are white noise. The
Granger representation theorem says that in this case, the system can be written
as:
x
t
=
1
(y
t1
x
t1
) +
p1
j=1
1j
x
tj
+
p1
j=1
1j
y
tj
+
1t
y
t
=
2
(y
t1
x
t1
) +
p1
j=1
2j
x
tj
+
p1
j=1
2j
y
tj
+
2t
, (2.2)
where at least one of parameters
1
and
2
deviates from zero. Both equations
of the system are balanced, that is, their left-hand and right-hand sides are of
the same order of integration, since y
t1
x
t1
I(0).
Suppose that y
t
x
t
= 0 denes a dynamic equilibrium relationship between
the two economic variables, y and x. Then y
t
x
t
is an indicator of the degree
of disequilibrium. The coecients
1
and
2
represent the strength of the dise-
quilibrium correction, and the system is now said to be in error-correction form.
A system characterized by these two equations is thus in disequilibrium at any
given time, but has a built-in tendency to adjust itself towards the equilibrium.
Thus, an econometric model cannot be specied without knowing the order of
integration of the variables. Tests of the unit root (nonstationarity) hypothesis
were developed by Fuller (1976), Dickey and Fuller (1979, 1981), Phillips and
Perron (1988), and others.
4
When these tests are applied to each of the three
time series in Figure 1.1, the null hypothesis of a unit root cannot be rejected.
But a unit root is rejected for the rst dierences of the series. The series can
thus be regarded as realizations of stochastic I(1) variables.
It should be mentioned that linear combinations of nonstationary variables
had appeared in dynamic econometric equations prior to Grangers work on coin-
tegration. Phillips (1957), who coined the term error correction, and Sargan
(1964) were forerunners. The well-known consumption equation in Davidson,
Hendry, Srba and Yeo (1978), the so-called DHSY model, disseminated the idea
4
Dierencing a variable with a unit root removes the unit root.
6
among macroeconomists. In their paper, based on quarterly UK series, a lagged
dierence of c
t
y
t
, where c
t
is the logarithmof private consumption of nondurable
goods and y
t
private income, represented the error-correction component. These
authors did not, however, consider the statistical implications of introducing such
components into their models.
5
2.2. Cointegration: estimation and testing
The concept of cointegration would not have become useful in practice without
a statistical theory for testing for cointegration and for estimating parameters of
linear systems with cointegration. Granger and Robert Engle jointly developed
the necessary techniques in their classical and remarkably inuential paper, Engle
and Granger (1987), where the theory of cointegrated variables is summed up
and extended. The paper contains, among other things, a rigorous proof of the
Granger representation theorem.
Engle and Granger (1987) consider the problem of testing the null hypothesis
of no cointegration between a set of I(1) variables. They estimate the coecients
of a static relationship between these variables by ordinary least squares and
apply well-known unit root tests to the residuals. Rejecting the null hypothesis
of a unit root is evidence in favor of cointegration. The performance of a number
of such tests is compared in the paper.
More recently, it has become possible to test the null hypothesis that the
estimated linear relationship between the I(1) variables is a cointegrating rela-
tionship (errors in the regression are stationary) against the alternative of no
cointegration (errors are nonstationary). Tests of this hypothesis were developed
by Shin (1994), based on a well-known stationarity test in Kwiatkowski, Phillips,
Schmidt and Shin (1992), as well as by Saikkonen and Luukkonen (1993), Xiao
and Phillips (2002), and others.
Another fundamental contribution in Engle and Granger (1987) is the two-
stage estimation method for vector autoregressive (VAR) models with cointegra-
tion. Consider the following VAR model of order p:
x
t
=
x
t1
+
p1
j=1
j
x
tj
+
t
(t = 1, ..., T) (2.3)
where x
t
is an n 1 vector of I(1) variables,
is an n n matrix such that

the n r matrices and have rank r,
j
, j = 1, ..., p 1, are n n para-
meter matrices, and
t
is an n 1 vector of white noise with a positive denite
covariance matrix. If 0 < r < n, the variables in x
t
are cointegrated with r
cointegrating relationships
x
t
. Stock (1987) had shown that under certain reg-
ularity conditions, the least squares estimator

of is consistent and converges
5
For a discussion of the DHSY model using cointegration analysis, see Hendry, Muellbauer
and Murphy (1990).
7
to the true value at the rapid rate T
1
(T is the number of observations); for
this reason

is called superconsistent. Using this result, Granger and Engle
showed that the maximum likelihood estimator of the remaining parameters
and
j
, j = 1, ..., p 1, obtained by replacing by

, has the same asymptotic
distribution as the estimator based on the true value .
If the variables in x are cointegrated, the parameters of (2.3) can thus be
estimated in two stages: begin by estimating or, more precisely, the cointe-
grating space ( up to a multiplicative constant) using a form of least squares.
Then, holding that estimate xed, estimate the remaining parameters by maxi-
mum likelihood. The estimators of and
j
, j = 1, ..., p 1, are consistent and
asymptotically normal. Hypotheses involving these parameters and their values
can be tested using standard statistical inference.
The results in Engle and Granger (1987) opened the gates for a ood of appli-
cations. They enhanced the popularity of VAR models developed by Sims (1980)
to oer an alternative to simultaneous-equation models. Sims had emphasized the
use of unrestricted VAR models as a means of modelling economic relationships
without unnecessary assumptions. On the other hand, a VAR model with cointe-
gration is often based on the idea of a long-run, or moving equilibrium, dened
by economic theory and characterized by vector
x
t1
in (2.3) . The short-term
dynamics represented by the parameter matrices
j
, j = 1, ..., p 1, are free
from restrictions. So is the strength-of-adjustment matrix that describes the
contribution of the degree of long-run disequilibrium to the adjustment process
towards the moving target or equilibrium.
Engle and Grangers two-step method represented a decisive breakthrough
in the modeling of economic relationships using nonstationary cointegrated time
series. Among later developments, the work of Johansen (1988, 1991) deserves
special mention. Johansen derived the maximum likelihood estimator of or,
more precisely, the space spanned by the r cointegrating vectors in (2.3) using
reduced rank regression.
6
He also derived sequential tests for determining the
number of cointegrating vectors. Johansens method can be seen as a second-
generation approach, in the sense that it builds directly on maximum likelihood
estimation instead of partly relying on least squares.
2.3. Extensions to cointegration
Granger and Engle, with various co-authors, have also extended the concept
of cointegration to seasonally integrated variables. In applied work, it is quite
common to render a time series with strong seasonal variation stationary by
seasonal dierencing. For example, if x
t
is a nonstationary quarterly series, its
seasonal dierence
4
x
t
= x
t
x
t4
may be I(0). If two nonstationary seasonal
6
In practice, vector is normalized by xing one of its elements so that its estimate, as well
as the estimate of , are unique.
8
series x
t
and y
t
can be made I(0) by seasonal dierencing and there exists a linear
combination y
t
x
t
I(0), then the two series are called seasonally cointegrated.
Such series were rst analyzed by Hylleberg, Engle, Granger and Yoo (1990).
Granger and Lee (1990) dened the concept of multicointegration, which can
be a useful tool when modeling stock-ow relationships. Suppose x
t
and y
t
are
cointegrated. Then the cumulative sum of deviations from the cointegrating
relation y
t
x
t
= 0 is necessarily an I(1) variable. If this new variable is
cointegrated with one of the original cointegrated variables, x
t
or y
t
, then the
latter two are multicointegrated.
7
In many cases, deviations from equilibrium are explained by transaction and
information costs. Granger and Swanson (1996) demonstrate how such costs may
be incorporated into models with cointegrated variables and how this may give
rise to a nonlinear error correction model. Adjustment costs are often lumpy,
however, and an adjustment will not occur unless the deviation from the equilib-
rium or desired value exceeds a certain threshold value. Granger and Swanson
show how such mechanisms, analyzed for instance in (S, s) models for adjusting
inventories and in menu cost models for price adjustment, can be incorporated
into models with cointegrated variables. Statistical theory for this type of cointe-
gration was rst worked out by Balke and Fomby (1997), who called it threshold
cointegration. For recent developments in this area, see Lo and Zivot (2001).
2.4. Areas of application
Cointegration has become a common econometric tool for empirical analysis in
numerous areas, where long-run relationships aect currently observed values:
current consumption is restricted by expected future income, current long-term
interest rates are determined by expected short-term rates, and so on. In such
areas, potential cointegrating relationships can be derived from economic theory,
tested, and if there is indeed cointegration incorporated into econometric
models.
The wealth-consumption relationship is an example where the interplay be-
tween theory and practice is changing our view of the real world. The traditional
view in many textbooks is that an increase in wealth causes a rise in consumption
roughly in proportion to the real interest rate. This magnitude might also appear
reasonable in terms of the so-called life-cycle model of consumption and savings.
If it were true, the uctuations on stock and housing markets would have a very
strong impact on consumption.
The traditional view, however, relies on econometric studies and simulated
7
For example, sales and production in an industry may be I(1) and cointegrated, in which
case their dierence, the change in inventory, is an I(0) variable. Then, the level of inventory
(initial level plus cumulated changes) will be I(1). With a target level of inventory, dened as
a xed proportion of sales, inventory and sales would be cointegrated, This, in turn, makes the
original variables, sales and production, multicointegrated.
9
models that do not distinguish suciently between temporary and permanent
perturbations in wealth. A very recent study by Lettau and Ludvigson (2003)
shows that consumption, labor income and wealth have to be cointegrated if
households observe an intertemporal budget constraint in their consumption be-
havior. After ascertaining that this theoretical starting-point agrees with their
data, the authors estimate an error-correction model that gives two results. First,
a majority of perturbations in wealth are temporary and related mainly to uctu-
ations in the stock market and second, such temporary perturbations have little
eect on consumption, both in the short and long run.
Well-known cointegration studies based on economic theory include Camp-
bell and Shiller (1987) who study bubbles in asset prices, Campbell and Shiller
(1988), Cochrane (1994) and Lettau and Ludvigson (2001) who investigate the
predictability of stock prices, Campbell (1987) who tests the hypothesis that con-
sumption is determined by permanent income, King, Plosser, Stock and Watson
(1991), who consider the role of permanent productivity shocks in the postwar
US economy, Johansen and Juselius (1990) who study demand for money, and
Hall, Anderson and Granger (1992) who consider the term structure of interest
rates.
The study of exchange rates and prices allows us to illustrate, in a simple
way, how cointegration may transform empirical analysis and reveal new ways of
investigating old problems. A basic proposition, formulated long ago by Cassel
(1922), is that exchange rates adjust so as to maintain purchasing power parity
(PPP): the price of a bundle of goods, expressed in common currency, should be
the same across countries. (See Froot and Rogo (1995) and Sarno and Taylor
(2002) for surveys of the literature on PPP, exchange rates and prices.) Assuming
two countries, we have
S
t
P
t
= P
t
(2.4)
where S
t
is the exchange rate between the domestic and the foreign currency, and
P
t
and P
t
are the price levels of domestic and foreign goods in local currency.
Put dierently, the real exchange rate S
t
P
t
/P
t
is assumed to be constant and
equal to one.
At rst, this proposition seems completely at odds with the observed huge
swings in exchange rates and relative stability of ination rates; see e.g., the
series in Figure 1.1. The gure indicates that in the 1970s, contrary to the
simple PPP hypothesis, the yen gained strength against the dollar while the
Japanese ination rate was higher than the US rate. Developments have been
less contradictory thereafter, but the relation between the uctuating exchange
rate and the stable relative price levels nevertheless appears rather weak.
The lack of a clear-cut one-to-one relationship between the exchange rate and
relative prices should not seem surprising. First, in the short run, the exchange
rate is primarily aected by expectations and capital movements. Consequently,
it may take time for trade to smooth out deviations from PPP: at best, PPP
10
is a long-run relationship. Second, deviations from PPP could be explained by
some goods being nontradable, as no obvious mechanism would eliminate price
dierences between such goods. Third, transaction and shipping costs require
price dierences to be of a certain minimum size before they are removed by
trade.
Empirical studies of the PPP hypothesis initially focussed on Cassels (1922)
simple formulation. PPP was tested by using regression models of the form
s
t
= a + b(p
t
p
t
) +
t
(2.5)
where s
t
is the logarithm of the exchange rate between the home and foreign
currency, and p
t
and p
t
are the logarithms of the price levels of home and foreign
goods, respectively. As seen from (2.4) , the null hypothesis of PPP is equivalent
to a = 0 and b = 1 in (2.5). Frenkel (1978, 1981), and many others, estimated
equations of this type. In these articles, equation (2.5) was estimated by ordinary
least squares and the hypothesis b = 1 tested using the standard t-statistic.
Strong rejection of the PPP hypothesis was a common result, at least for data
from industrialized countries during the post-Bretton Woods period with oating
exchange rates. Moreover, the rejections showed no clear pattern: estimates of b
deviated from 1 both upwards and downwards. These tests are awed, however,
as they do not take into account the prospective nonstationarity of exchange rates
and price levels, i.e., the possibility that
t
I(1). Furthermore, static equations
of type (2.5) cannot separate (possibly strong) short-run deviations from PPP
and long-run adjustments of the exchange rate towards an equilibrium.
The next generation of studies explicitly treat PPP as a long-run relationship,
implying that deviations from PPP should follow a stationary process. This
amounts to assuming a = 0 and b = 1 in (2.5) and checking whether the residual,
that is, the deviation from a xed real exchange rate, is an I(0) process. When
such tests were applied to post-Bretton Woods time series, they did not reject the
unit root hypothesis; see e.g., Meese and Rogo (1988), and thus did not support
PPP even as a long-run relationship. An explanation oered in the literature
was that the time series had not been long enough for the unit-root tests to have
power. When the series are extended backwards a century or more, the unit-root
hypothesis is indeed generally rejected; see for example Frankel (1986).
If PPP is valid in the long run, howquick is the adjustment towards it? Edison
(1987) examined this question, applying an error-correction framework to series
covering a century or more. This meant estimating equations of the type
s
t
= a + b(s
t1
p
t1
+ p
t1
) + g(p
t1
p
t1
) +
t
, (2.6)
where b measures the rate of convergence towards PPP. Typical results suggest
half-lives of deviations from PPP of between 3 and 7 years. The short-run dy-
namic behavior of the exchange rate can then be explained as a combination of
transitory shocks and gradual adjustment towards PPP.
11
These second-generation studies assume that there exists a cointegrating rela-
tionship between s
t
, p
t
and p
t
with cointegrating vector = (1, 1, 1), obtained
from the simple form of PPP. It seems realistic, however, to allow for a trending
real exchange rate, due to dierent trends in the relative prices of traded vs.
nontraded goods (the so-called Balassa-Samuelson eect).
8
The relevant cointe-
grating vector in such a situation is = (1, ,
where parameters and
, if they deviate from unity, may reect dierent price trends for traded and
nontraded goods.
9
To consider this extension, begin by running the regression
s
t
= + p
t
t
+
t
(2.7)
and testing nonstationarity (or stationarity) of the errors. Stationary errors sup-
port the more general version of PPP. Next, if the errors are found to be sta-
tionary, the whole dynamic system can be estimated as discussed in Engle and
Granger (1987). This procedure gives an idea of whether and how the exchange
rate uctuates around the PPP equilibrium.
Anumber of third-generation studies using this approach have been conducted
since the late 1980s. A common nding is that the null hypothesis of no cointe-
gration is rejected more frequently than the null of a unit root in s
t
p
t
+ p
t
.
The results are still sensitive to the observation period: post-Bretton Woods data
tend to produce estimates of and
far from unity, while data from the entire

twentieth century produce values closer to one.
The three time series in Figure 1.1 serve to demonstrate how the validity of
the PPP hypothesis may be examined using cointegration analysis. Estimating
the parameters of equation (2.7) yields
s
t
= 6.63 + 0.44p
t
1.07p
t
+
t
. (2.8)
The estimated residuals from (2.8) are graphed in Figure 2.1. They appear to
uctuate around zero, which suggests stationarity, but display considerable per-
sistence, which suggests the opposite. Formal scrutiny is required.
Culver and Papell (1999) tested the PPP hypothesis for a number of countries,
with cointegration as the null hypothesis against the alternative of nonstationar-
ity, using the test of Shin (1994). Following their approach here, cointegration is
8
According to the Balassa-Samuelson eect, poor countries grow faster than rich countries,
and consumer prices in the former increase faster than they do in the latter. A reason is that
faster growth in poor economies is primarily due to productivity growth in the tradable goods
sector where prices tend to be equal across all countries. Strong growth in productivity leads
to large wage increases in the non-tradables sector as well; in this sector, however, productivity
can be expected to grow at about the same rate in all countries. Large wage increases then
imply large increases in prices of non-tradables in poor countries. The latter price increases, in
turn, raise the growth rate of the aggregate consumer price index in these countries.
9
This requires the assumption that the foreign price level p
t
is exogenous in the sense that
there is no feedback to it from the other two variables. This restricts the considerations to a
system with two equations, so that there can be at most one cointegrating vector.
12
Figure 2.1: Residuals of regression (2.8)
not rejected at the 5% level. If we conclude that s
t
, p
t
and p
t
can be regarded
as cointegrated, the vector

= (1, 0.44, 1.07) estimated from (2.7) is a cointe-
grating vector. It is not close to (1, 1, 1) as would be required for the restricted
version of PPP, but the data provide weak support for a general version of PPP
between Japan and the US during the period 1975-2003.
10
3. Modeling volatility
Many nancial economists are concerned with modeling volatility in asset returns.
Portfolio-choice theory attempts to derive optimal portfolios as functions of vari-
ances and covariances of returns. The capital asset pricing model (CAPM) and
other asset-pricing models show how investors are rewarded for taking system-
atic risks, i.e., risks related to the covariance between their own and the market
portfolio or other non-diversiable factors. Option-pricing formulas give prices of
options and other derivative instruments in terms of volatility of the underlying
asset. Banks and other nancial institutions apply so-called value-at-risk mod-
els to assess risks in their marketable assets. For all these purposes, modeling
volatility or, in other words, the covariance structure of asset returns, is essential.
Financial economists have long since known that volatility in returns tends to
cluster and that the marginal distributions of many asset returns are leptokurtic,
which means that they have thicker tails than the density of the normal distri-
bution with the same mean and variance. Even though the time clustering of
returns was known to many researchers, returns were still modeled as indepen-
dent and identically distributed over time. Examples include Mandelbrot (1963)
and Mandelbrot and Taylor (1967) who used so-called stable Paretian distrib-
utions to characterize the distribution of returns.
11
Robert Engles modelling
10
The results are sensitive to the choice of null hypothesis, however. Augmented Dickey-
Fuller tests of the unit-root hypothesis, see Dickey and Fuller (1981), give t-values from 2.3
to 2.5, which are not small enough to reject the null hypothesis at the 0.1 signicance level
(the critical value equals 2.58).
11
For a thorough account of stable Paretian distributions and their use in nance and econo-
13
of time-varying volatility by way of autoregressive conditional heteroskedasticity
(ARCH) thus signied a genuine breakthrough.
We begin this section by introducing Engles basic ARCH model, some of
its generalizations and a brief application. This is followed by a discussion of
the so-called ARCH-in-mean (ARCH-M) model where conditional rst moments
in applications, usually expected asset returns are systematically related to
conditional second moments modeled by ARCH. Multivariate generalizations of
ARCHmodels are also considered, as well as new parameterizations of ARCH. We
conclude by considering value-at-risk analysis, where ARCH plays an important
role.
3.1. Autoregressive conditional heteroskedasticity
In time series econometrics, model builders generally parameterize the conditional
mean of a variable or a vector of variables. Suppose that at time t we observe
the stochastic vector (y
t
, x
t
)
where y
t
is a scalar and x
t
a vector of variables such
that some of its elements may be lags of y
t
. This implies the following predictive
model for variable y
t
:
y
t
= E{y
t
|x
t
} +
t
, (3.1)
where the conditional mean E{y
t
|x
t
} typically has a parametric form,
12
E{
t
|x
t
} =
E
t
= 0, and E
t
s
= 0 for t = s, that is, {
t
} is a sequence of uncorrelated ran-
dom variables with mean zero. When estimating the parameters in E{y
t
|x
t
}, it is
typically assumed that the unconditional variance of the error term
t
is constant
or time-varying in an unknown fashion. (Indeed, we assumed constant variances
and covariances in our presentation of estimation of cointegrated systems.)
Engle (1982) considered the alternative assumption that, while the uncondi-
tional error variance if it exists is constant, the conditional error variance is
time-varying. This revolutionary notion made it possible to explain systematic
features in the movements of variance over time and, a fortiori, to estimate the
parameters of conditional variance jointly with the parameters of the conditional
mean. The literature is wholly devoid of earlier work with a similar idea.
Engle parameterized the conditional variance of
t
in model (3.1) such that
large positive or negative errors
t
were likely to be followed by another large error
of either sign and small errors by a small error of either sign. More formally, he
assumed that
t
can be decomposed as
t
= z
t
h
1/2
t
, where {z
t
} is a sequence of
iid random variables with zero mean and unit variance, and where
h
t
= var(
t
|F
t
) =
0
+
q
j=1
2
tj
. (3.2)
metrics, see Rachev and Mittnik (2000).
12
Alternatively, the conditional mean is a nonparametric function of x
t
, in which case the
focus is on estimating its functional form.
14
In (3.2) ,
t
= y
t
E{y
t
|x
t
}, and the information set F
t
= {
tj
: j 1},
0
> 0,
and
j
0, j = 1, ..., q.
Equation (3.2) denes the ARCH model introduced in Engle (1982), where
the conditional variance is a function of past values of squared errors. In this
classic paper, Engle developed the estimation theory for the ARCH model, gave
conditions for the maximum likelihood estimators to be consistent and asymp-
totically normal, and presented a Lagrange multiplier test for the hypothesis of
no ARCH (equal conditional and unconditional variance) in the errors
t
.
3.2. Extensions and applications
In practice,
2
t
tends to have a relatively slowly decaying autocorrelation func-
tion among return series of suciently high observation frequency, such as daily
or weekly series. An adequate characterization of this stylized fact requires an
ARCH model with a long lag q. But if the right-hand side of (3.2) is modied by
adding lags of the conditional variance h
t
(one lag is often enough), the resulting
model can be formulated with only a small number of parameters and still display
a slowly decaying autocorrelation function for
2
t
. Not long after the publication
of the ARCH paper, Engles graduate student Tim Bollerslev introduced such a
model and called it the generalized ARCH (GARCH) model. This model has the
following form, see Bollerslev (1986),
h
t
=
0
+
q
j=1
2
tj
+
p
j=1
j
h
tj
. (3.3)
The rst-order (p = q = 1) GARCH model, also suggested independently by
Taylor (1986), has since become the most popular ARCH model in practice.
Compared to Engles basic ARCH model, the GARCH model is a useful technical
innovation that allows a parsimonious specication: a rst-order GARCH model
contains only three parameters. However, it does not add any conceptually new
insight.
The application in Engle (1982) involved macroeconomic series such as the
ination rate, but Engle quickly realized that the ARCH model was useful in -
nancial economics, as well. Indeed, when considering time series of price changes,
Mandelbrot (1963) had already observed that ... large changes tend to be fol-
lowed by large changes of either sign and small changes tend to be followed
by small changes ..., but he did not go on to model the returns as time depen-
dent. If the conditional mean is assumed constant, then the ARCH model can
be used to characterize return series that contain no linear dependence but do
display clustering of volatility. As the ARCH model is also suitable for modeling
leptokurtic observations, it can be used to forecast volatility. This, in turn, may
be crucial for investors who want to limit the riskiness of their portfolio.
The upper panel of Figure 3.1 shows the daily logarithmic returns (rst dif-
ferences of the logarithms of daily closing prices) of the Standard and Poor 500
15
Figure 3.1: Upper panel: Logarithmic daily returns of the Standard and Poor 500
stock index, May 16, 1995 April 29, 2003. Lower panel: Estimated conditional
variance for the same index from the rst-order GARCH model (3.4).
stock index from May 16, 1995, to April 29, 2003 2000 observations in all.
Volatile periods do in fact alternate with periods of relative calm, as in the case
of the daily exchange rate series in Figure 1.2.
Fitting a rst-order GARCH model to the series in Figure 3.1 under the
assumption that the errors z
t
are normal (and
t
thus conditionally normal) yields
h
t
= 2 10
6
+ 0.091
2
t1
+ 0.899h
t1
. (3.4)
The sum of
1
+

1
is close to one, which is typical in applications. Condition
1
+
1
< 1 is necessary and sucient for the rst-order GARCH process to
be weakly stationary, and the estimated model (3.4) satises this condition. The
lagged conditional variance h
t1
has coecient estimate 0.9, meaning that 90% of
a variance shock remains the next day, and the half-life equals six days. The lower
panel of Figure 3.1 shows the estimated conditional variance h
t
over time. Spikes
in the graph have a relatively slowly decreasing right-hand tail, which shows that
volatility is persistent. Another noteworthy observation is that during turbulent
periods, conditional variance is several times higher than its basic level. This
suggests that the turbulence will have practical implications for investors when
forecasting of the volatility of a stock index or a portfolio. For a recent survey
on volatility forecasts, see Poon and Granger (2003).
16
What is parameterized in (3.2) and (3.3) is conditional variance. An alter-
native proposed by Schwert (1990) is to parameterize the conditional standard
deviation h
1/2
t
. This is important in forecasting with GARCH models. If the loss
function of the forecaster is based on the mean absolute error instead of the more
common root mean square error, it is natural to use a model that parameterizes
the conditional standard deviation and not the conditional variance. A more
general model would contain both alternatives as special cases. Such a model,
called the asymmetric power GARCH model, was introduced by Ding, Granger
and Engle (1993). They parameterized a general conditional moment h
t
where
> 0. When Ding et al. (1993) estimated a power-GARCH(1,1) model for a
long daily return series of the S&P 500 index, they obtained

= 1.43, and both
null hypotheses, = 1 and = 1/2, were rejected. The GARCH(1,1) model has
nevertheless retained its position as the overwhelmingly most popular GARCH
model in practice.
3.3. ARCH-in-mean and multivariate ARCH-in-mean
ARCH and GARCH models are ecient tools for estimating conditional second
moments of statistical distributions i.e., variances and covariances. A great deal
of nancial theory deals with the connection between the second moments of asset
returns and the rst moments (expected asset returns). It seemed self-evident to
extend the ARCH model to explicitly characterize this connection. Such a model,
Engles rst application of ARCH to nance, can be found in Engle, Lilien and
Robins (1987). Engle and his co-authors consider a two-asset economy with a
risky asset and a risk-free asset. They assume that risk is measured as a function
of the conditional variance of the risky asset. As a result, the price oered by risk-
averse agents uctuates over time, and the equilibriumprice determines the mean-
variance relationship. This suggests including a positive-valued monotonically
increasing function of conditional variance in the conditional mean equation. In
its simplest form this yields
r
t
= + g(h
t
) +
t
, (3.5)
where r
t
is the excess return of an asset at time t, g(h
t
) is a function of the
conditional variance h
t
, and h
t
is dened as in (3.2). Engle et al. (1987) chose
g(h
t
) = h
1/2
t
, > 0, that is, a multiple of the conditional standard deviation
of
t
. Equations (3.5) and (3.2) jointly dene an ARCH-in-mean model. The
authors applied the model to explain monthly excess returns of the six-month US
Treasury bill. Assuming the risk-free asset was a three-month Treasury bill, they
found a signicant eect from the estimated risk component

h
1/2
t
on the excess
return r
t
.
In nancial theory, however, the price of an asset is not primarily a function
of its variance but rather of its covariance with the market portfolio (CAPM)
17
and other non-diversiable risk factors (Arbitrage Price Theory). To apply the
ARCH-in-mean asset pricing model to the pricing of several risky assets thus
implied modelling conditional covariances. Instead of the standard CAPM where
agents have common and constant expectations of the means and the variances
of future returns, this generalization leads to a conditional CAPM, where the
expected returns are functions of the time-varying covariances with the market
portfolio.
Bollerslev, Engle and Wooldridge (1988) constructed such a multivariate GARCH
model. Let r
t
be the n1 vector of real excess returns of assets at time t and
t
the corresponding vector of value weights. According to the CAPM, the condi-
tional mean vector of the excess returns is proportional to the covariance between
the assets and the market portfolio:
t
= H
t
t1
, (3.6)
where H
t
= [h
ijt
] is the nn conditional covariance matrix, h
ijt
is the conditional
covariance between asset i and asset j at time t and is a constant. Accord-
ing to (3.6) , expected returns of assets change over time with variations in the
covariance structure. In other words, the so-called -coecient in CAPM is time-
varying. Matrix H
t
is parameterized in such a way that each conditional variance
and covariance has its own equation. As postulated by Bollerslev et al. (1988),
this leads to the following multivariate GARCH-in-mean model:
r
t
=
0
+ H
t
t1
+
t
and
13
vech(H
t
) = +
q
j=1
A
j
vech(
tj
tj
) +
p
j=1
B
j
vech(H
tj
). (3.7)
With three assets (n = 3), system (3.7) consists of six equations, for three con-
ditional variances and three conditional covariances. To keep down the number
of parameters in (3.7) , Bollerslev et al. (1988) made the simplifying assumptions
that p = q = 1 and that A
1
and B
1
are diagonal matrices. They then applied the
model to quarterly data for three assets: bills (six-month Treasury bill), bonds
(twenty-year Treasury bond) and stocks (NYSE index including dividends). The
results show, among other things, that the conditional variances and covariances
are strongly autoregressive. The hypothesis that the conditional covariance ma-
trix H
t
is constant over time is clearly rejected, which implies that the vector of
beta coecients in the CAPM should be time-varying.
13
The vech-operator chooses the observations in every column that lie above or on the main
diagonal and stacks them on top of each other, beginning with the rst column. For instance,
let
A =

a
11
a
12
a
21
a
22
.
Then vech(A) = (a
11
, a
12
, a
22
)
.
18
In order to mitigate the practical problem inherent in estimating a large num-
ber of parameters, Engle (1987) suggested a model that was later applied in Engle,
Ng and Rothschild (1990). In this factor-ARCH model the vector of asset returns
r
t
has the following denition:
r
t
= B
t
+
t
, (3.8)
where r
t
has a factor structure. In equation (3.8) , B is an n k matrix of factor
loadings, that is, parameters to be estimated,
t
is a k 1 vector of unobservable
factors, and k is expected to be much smaller than n. This implies that the
dynamic behavior of a large number of asset returns is characterized by a small
number of common factors. Assume now that the errors
t
have a constant
conditional covariance matrix and that the factors have a diagonal conditional
covariance matrix
t
. Assuming that
t
and
t
are uncorrelated leads to the
following conditional covariance matrix of r
t
:
cov(r
t
|F
t1
) = +B
t
B
= +
k
j=1
jt
j
, (3.9)
where
t
= diag(
1t
, ...,
kt
) and
j
is the jth column of B. If each
jt
is assumed
to have an ARCH-type structure, the parameters of this factor-ARCH model can
be estimated. Estimation is simplied by assuming that the portfolios (their value
weights) are known, since this implies knowing the elements of
j
. Estimation
and implications of the factor-ARCH model are discussed in Engle et al. (1990).
They applied a one-factor model (k = 1 in (3.9)) to the pricing of Treasury bills
of dierent maturities.
Independently of Engle and his co-authors, Diebold and Nerlove (1989) de-
veloped a similar model that was successfully applied to modeling commonality
in volatility movements of seven daily exchange rate series. Factor models of this
type have also been used to study linkages between international stock markets.
3.4. Other developments
Since its inception, the statistical theory of ARCH models has been extended
and applications abound. Hundreds of applications to nancial time series had
already been listed in a survey by Bollerslev, Chou and Kroner (1992), and the
number has continued to grow steadily. Several authors have contributed to the
estimation theory for these models and derived conditions for consistency and
asymptotic normality of maximum likelihood estimators in both univariate and
multivariate ARCH and GARCH models.
Robert Engle himself has enhanced the expanding literature. One of the
problems in multivariate GARCH modeling was to ensure that the matrix H
t
of
conditional covariances be positive denite for every t. Engle and Kroner (1995)
dened a parsimonious GARCH model where this assumption is satised that
19
has become popular among practitioners. Engle (2002a) has suggested another
multivariate GARCH model with this property, the so-called dynamic condi-
tional correlation GARCH model. Independently of Engle, a similar model was
developed by Tse and Tsui (2002). This model is an extension of the constant
conditional correlation GARCH model of Bollerslev (1990).
Engle and Ng (1993) devise misspecication tests for GARCH models, which
is an important development. They also introduce a new concept, the news
impact curve. The idea is to condition at time t on the information available at
t 2 and thus consider the eect of the shock
t1
on the conditional variance
h
t
in isolation. Dierent ARCH and GARCH models can thus be compared by
asking how the conditional variance is aected by the latest information, the
news. For example, the news impact curve of the GARCH(1,1) model has the
form
h
t
= A +
1
2
t1
where A =
0
+
1
2
(
2
is the unconditional variance of
t
). This curve is
symmetric with respect to
t1
= 0. Other GARCH models have asymmetric
news impact curves; see Engle and Ng (1993) and Ding et al. (1993) for examples
and discussion. According to such models, a positive and an equally large negative
piece of news do not have the same eect on the conditional variance.
Engles original idea has also spawned dierent parameterizations. The most
commonly applied is the exponential GARCH model of Nelson (1991), where
the logarithm of conditional variance has a parametric form. This was the rst
asymmetric GARCH model. While ordinary GARCH models require parameter
restrictions for conditional variance to be positive for every t, such restrictions
are not needed in the exponential GARCH model.
Another model that deserves mention in this context is the autoregressive sto-
chastic volatility model. It diers from GARCH models in that the logarithm of
the conditional variance is itself a stochastic process. A rst-order autoregressive
stochastic volatility process, rst suggested by Taylor (1982), has the form
ln h
t
= + ln h
t1
+
t
where h
t
is a positive-valued conditional variance variable and {
t
} is a se-
quence of independent, identically distributed random variables with mean zero
and constant variance. The stochastic volatility model of
t
has an inherent
technical complication. It does not have a closed form because it contains two
unobservable random processes: {z
t
} due to the decomposition
t
= z
t
h
1/2
t
, and
{
t
}. Recently, stochastic volatility models have attracted considerable attention
along with the development of eective numerical estimation methods for their
parameters; see the surveys by Ghysels, Harvey and Renault (1996) and Shephard
(1996).
Building on the theory of ARCH models, Engle recently considered new mod-
els for the empirical analysis of market microstructure. The idea is to apply a
20
GARCH-like model to transaction data on durations between trades, which is
feasible because duration is a positive-valued variable in the same way as the
squared error
2
t
in the ARCH model. In two contributions, Engle and Russell
(1998) and Engle (2000), dening the so-called autoregressive conditional dura-
tion (ACD), Engle initiated a new literature to clarify the behavior of individual
agents in stock markets. These papers have generated a remarkable amount of
interest and new papers on ACD models have appeared in rapid succession.
3.5. Application to value at risk
In addition to their use in asset pricing, ARCH and GARCH models have also
been applied in other areas of nancial economics. The pricing of options and
other derivatives, where the variance of the underlying asset is a key parameter,
is an obvious area of application; see Noh, Engle and Kane (1994).
ARCH and GARCH models have also become popular and indispensable tools
in modern risk management operations. Nowadays banks, other nancial institu-
tions and many large companies use so-called value-at-risk analysis. Value-at-risk
models are also used to calculate capital requirements for market risks accord-
ing the so-called Basle II rules; see, for example, Basle Committee on Banking
Supervision (1996). To understand the concept, consider an investor with an
asset portfolio. The investor wants to predict the expected minimum loss, L
min
,
on this portfolio that will occur at a given, small probability over the holding
period. The predicted value of L
min
, the value at risk, measures the riskiness
of the portfolio. Turning this around, the prediction is that the loss will be no
greater than L
min
with probability 1 . This concept is a natural measure for
risk control, for example in cases where a bank regulator wants to ensure that
banks have enough capital for the probability of insolvency within, say, the next
month not to exceed .
The attraction of value at risk is that it reduces the market risk associated
with a portfolio of assets to an easily understandably number. The loss can be
calculated by assuming that the marginal distribution of returns is constant over
time, but in view of the evidence this does not seem realistic. If the return
distribution is time-varying, however, a model is required to predict the future
values of the conditional moments characterizing the distribution. If the latter
is assumed to be conditionally normal, then the rst two moments, the mean
and the variance, completely characterize the distribution. GARCH models are
widely used for estimating the variance of the conditional return distribution
required to calculate the expected loss (their use can be extended to the non-
normal case as well). Practitioners often use an exponentially weighted moving
average
h
t
= (1
1
)
2
t1
+
1
h
t1
, 0 <
1
< 1,
which is a special case of (3.3) and, more precisely, of the so-called integrated
GARCH model introduced by Engle and Bollerslev (1986).
21
Manganelli and Engle (2001) survey the many approaches to computing the
value at risk. Variants of GARCH models have been important components of
this development. For instance, Engle and Manganelli (1999) have introduced a
so-called conditional autoregressive value at risk model, which is based on the
idea of directly modelling the quantile () of the distribution that is of interest.
As a simple example of the use of GARCH models in value-at-risk analysis,
consider an investor with an S&P 500 index the series in Figure 3.1 portfolio of
one million dollars. Assume that she applies the estimated GARCH(1,1) model
(3.4) with normal errors. The investor wants to estimate the amount below
which her loss will remain with probability 0.99 the next day the stock exchange
is open if she retains her portfolio. Consider two points in time: September
1, 1995 (Friday), when the conditional variance estimated from (3.4) attains its
minimum and July 31, 2002, when it obtains its maximum.
14
The maximum
loss predicted from the GARCH model for September 5, 1995 (Tuesday, after
Labor Day), equals $ 12,400, whereas the corresponding sum for August 1, 2002,
is $ 61,500. The dierence between the sums illustrates the importance of time-
varying volatility and of ARCH as a tool in the value-at-risk analysis.
4. Other contributions
Both Engle and Granger have made valuable contributions in several areas of
time-series econometrics. In addition to collaborating closely with Granger to
develop tests for cointegration and estimation techniques for models with coin-
tegrated variables, Engle has also done important work on exogeneity, a key
concept in econometric modeling (Engle, Hendry and Richard (1983) and Engle
and Hendry (1993)). Granger has left his mark in a number of areas. His de-
velopment of a testable denition of causality (Granger (1969)) has spawned a
vast literature. He has also contributed to the theory of so-called long-memory
models that have become popular in the econometric literature (Granger and
Joyeux (1980)). Furthermore, Granger was among the rst to consider the use
of spectral analysis (Granger and Hatanaka (1964)) as well as nonlinear models
(Granger and Andersen (1978)) in research on economic time series. His contri-
butions to the theory and practice of economic forecasting are also noteworthy.
Granger and Morgenstern (1970) is an early classic in this area, while Granger and
Bates (1969) may be regarded as having started the vast literature on combining
forecasts.
14
This example is, of course, articial in the sense that the GARCH model is estimated for
a period ending April 29, 2003. In practice, the investor could only use a GARCH model
estimated for the observations available at the time the Value at Risk is calculated.
22
5. Summary and hints for further reading
Since its inception, cointegration has become a vast area of research. A remark-
able number of books and articles dealing with theoretical aspects as well as
applications have been published. Cointegration has become a standard topic in
econometrics textbooks. Engle and Granger (1991) is a collection of key articles,
including some of the references in this paper. Books by Banerjee, Dolado, Gal-
braith and Hendry (1993), Johansen (1995) and Hatanaka (1996) consider the
statistical theory underlying cointegration analysis. Watson (1994) is a survey
that has a general vector autoregressive model with nonstationary variables as
its starting-point. A broad technical overview of the statistical theory of nonsta-
tionary processes including cointegration can be found in Tanaka (1996).
Autoregressive conditional heteroskedasticity has, like cointegration, gener-
ated an extensive literature, and most time series and econometrics texts nowa-
days include an exposition of the topic. Engle (1995) is a collection of key papers.
Surveys include Bollerslev, Engle and Nelson (1994), Diebold and Lopez (1995),
Palm (1996) and Shephard (1996). The book by Gouriroux (1996) discusses
both statistical theory and nancial applications. Finally, Engle (2002b) oers a
glimpse into the future.
The work of Clive Granger on nonstationary time series and that of Robert
Engle on time-varying volatility has had a pervasive inuence on applied eco-
nomic and nancial research. Cointegration and ARCH, and the methods the
two scholars have developed around these concepts, have indelibly changed the
way econometric modeling is carried out.
23
References
Balke, N. and Fomby, T. B.: 1997, Threshold cointegration, International Eco-
nomic Review 38, 627645.
Banerjee, A., Dolado, J., Galbraith, J. W. and Hendry, D. F.: 1993,
Co-Integration, Error-Correction, and the Econometric Analysis of Non-
Stationary Data, Oxford University Press, Oxford.
Basle Committee on Banking Supervision: 1996, Amendment to the Capital Ac-
cord to Incorporate Market Risks, Basle.
Bollerslev, T.: 1986, Generalized autoregressive conditional heteroskedasticity,
Bollerslev, T.: 1990, Modelling the coherence in short-run nominal exchange
rates: A multivariate generalized ARCH model, Review of Economics and
Statistics 72, 498505.
Bollerslev, T., Chou, R. Y. and Kroner, K. F.: 1992, ARCH modeling in nance:
A review of the theory and empirical evidence, Journal of Econometrics
52, 559.
Bollerslev, T., Engle, R. F. and Nelson, D. B.: 1994, ARCH models, in R. F.
Engle and D. L. McFadden (eds), Handbook of Econometrics, Vol. 4, North-
Holland, Amsterdam, pp. 29593038.
Bollerslev, T., Engle, R. F. and Wooldridge, J.: 1988, A capital asset-pricing
model with time-varying covariances, Journal of Political Economy 96, 116
131.
Campbell, J. Y.: 1987, Does saving anticipate declining labor income? An al-
ternative test of the permanent income hypothesis, Econometrica 55, 1249
1273.
Campbell, J. Y. and Shiller, R. J.: 1987, Cointegration and tests of present value
models, Journal of Political Economy 95, 10621088.
Campbell, J. Y. and Shiller, R. J.: 1988, The dividend-price ratio and expecta-
tions of future dividends and discount factors, Review of Financial Studies
1, 195228.
Cassel, G.: 1922, Money and Foreign Exchange After 1914, MacMillan, New
York.
Cochrane, J. Y.: 1994, Permanent and transitory components of GNP and stock
prices, Quarterly Journal of Economics 109, 241266.
24
Culver, S. E. and Papell, D. H.: 1999, Long-run purchasing power parity with
short-run data: Evidence with a null hypothesis of stationarity, Journal of
International Money and Finance 18, 751768.
Davidson, J. E. H., Hendry, D. F., Srba, F. and Yeo, S.: 1978, Econometric
modelling of the aggregate time-series relationship between consumers ex-
penditure and income in United Kingdom, Economic Journal 88, 661692.
Dickey, D. A. and Fuller, W. A.: 1979, Distribution of the estimators for au-
toregressive time series with a unit root, Journal of the American Statistical
Association 74, 427431.
Dickey, D. A. and Fuller, W. A.: 1981, Likelihood ratio statistics for autoregres-
sive time series with a unit root, Econometrica 49, 10571072.
Diebold, F. X. and Lopez, J. A.: 1995, Modelling volatility dynamics, in K. D.
Hoover (ed.), Macroeconometrics. Developments, Tensions, and Prospects,
Kluwer, Boston, pp. 427466.
Diebold, F. X. and Nerlove, M.: 1989, The dynamics of exchange rate volatility:
A multivariate latent factor ARCH model, Journal of Applied Econometrics
4, 121.
Ding, Z., Granger, C. W. J. and Engle, R. F.: 1993, A long memory property
of stock market returns and a new model, Journal of Empirical Finance
1, 83106.
Edison, H.: 1987, Purchasing power parity in the long run: A test of the dol-
lar/pound exchange rate 1890-1978, Journal of Money, Credit, and Banking
19, 376387.
Engle, R. F.: 1982, Autoregressive conditional heteroscedasticity with estimates
of the variance of United Kingdom ination, Econometrica 50, 9871007.
Engle, R. F.: 1987, Multivariate GARCH with factor structures-cointegration in
variance, unpublished paper, Department of Economics, University of Cali-
fornia, San Diego.
Engle, R. F.: 2000, The econometrics of ultra-high-frequency data, Econometrica
68, 122.
Engle, R. F.: 2002a, Dynamic conditional correlation: Asimple class of multivari-
ate generalized autoregressive conditional heteroscedasticity models, Journal
of Business and Economic Statistics 20, 339350.
Engle, R. F.: 2002b, New frontiers for ARCH models, Journal of Applied Econo-
metrics 17, 425446.
25
Engle, R. F. and Bollerslev, T.: 1986, Modeling the persistence of conditional
variances, Econometric Reviews 5, 150.
Engle, R. F. (ed.): 1995, ARCH. Selected Readings, Oxford University Press,
Oxford.
Engle, R. F. and Granger, C. W. J.: 1987, Co-integration and error-correction:
Representation, estimation and testing, Econometrica 55, 251276.
Engle, R. F. and Granger, C. W. J. (eds): 1991, Long-Run Economic Relation-
ships. Readings in Cointegration, Oxford University Press, Oxford.
Engle, R. F. and Hendry, D. F.: 1993, Testing super exogeneity and invariance,
Engle, R. F., Hendry, D. F. and Richard, J.-F.: 1983, Exogeneity, Econometrica
51, 277304.
Engle, R. F. and Kroner, K. F.: 1995, Multivariate simultaneous generalized
ARCH, Econometric Theory 11, 122150.
Engle, R. F., Lilien, D. M. and Robins, R. P.: 1987, Estimating time-varying risk
premia in the term structure: The ARCH-M model, Econometrica 55, 391
407.
Engle, R. F. and Manganelli, S.: 1999, CAViaR: Conditional autoregessive value
at risk by regression quantiles, Working paper 99-20, Department of Eco-
nomics, University of California, San Diego.
Engle, R. F. and Ng, V. K.: 1993, Measuring and testing the impact of news on
volatility, Journal of Finance 48, 17491777.
Engle, R. F., Ng, V. K. and Rothschild, M.: 1990, FACTOR-ARCH covariance
structure: Empirical estimates for treasury bills, Journal of Econometrics
45, 213237.
Engle, R. F. and Russell, J. R.: 1998, Autoregressive conditional duration: A new
model for irregularly spaced transaction data, Econometrica 66, 11271162.
Frankel, J.: 1986, International capital mobility and crowding out in the U.S.
economy: Imperfect integration of nancial or goods markets?, in R. Hafer
(ed.), How Open is the US Economy, Lexington Books, Lexington.
Frenkel, J.: 1978, Quantifying capital mobility in the 1980s, in D. Bernheim and
J. Shoven (eds), National Saving and Economic Performance, University of
Chicago Press, Chicago.
26
Frenkel, J.: 1981, The collapse of purchasing power parity during the 1970s,
European Economic Review 16, 145165.
Froot, K. A. and Rogo, K.: 1995, Perspectives on PPP and long-run real ex-
changer rates, in G. Grossman and K. Rogo (eds), Handbook of Interna-
tional Economics, Vol. 3, Elsevier, Amsterdam, pp. 16471688.
Fuller, W. A.: 1976, Introduction to Statistical Time Series, Wiley, New York.
Ghysels, E., Harvey, A. C. and Renault, E.: 1996, Stochastic volatility, in G. S.
Maddala and C. R. Rao (eds), Handbook of Statistics 14: Statistical Methods
in Finance, Elsevier, Amsterdam.
Gouriroux, C.: 1996, ARCH Models and Financial Applications, Springer,
Berlin.
Granger, C. W. J.: 1969, Investigating causal relations by econometric models
and cross-spectral methods, Econometrica 37, 424438.
Granger, C. W. J.: 1981, Some properties of time series data and their use in
econometric model specication, Journal of Econometrics 16, 121130.
Granger, C. W. J.: 2001, Spurious regressions in econometrics, in B. H. Bal-
tagi (ed.), A Companion to Theoretical Econometrics, Blackwell, Oxford,
pp. 557561.
Granger, C. W. J. and Andersen, A. P.: 1978, Introduction to Bilinear Time
Series Models, Vandenhoeck and Ruprecht, Gttingen.
Granger, C. W. J. and Bates, J.: 1969, The combination of forecasts, Operations
Research Quarterly 20, 451468.
Granger, C. W. J. and Hatanaka, M.: 1964, Spectral Analysis of Economic Time
Series, Princeton University Press, Princeton, NJ.
Granger, C. W. J. and Joyeux, R.: 1980, An introduction to long-memory time
series models and fractional dierencing, Journal of Time Series Analysis
1, 1530.
Granger, C. W. J. and Lee, T.-H.: 1990, Multicointegration, in G. F. Rhodes, Jr
and T. B. Fomby (eds), Advances in Econometrics: Cointegration, Spurious
Regressions and Unit Roots, JAI Press, New York, pp. 1784.
Granger, C. W. J. and Morgenstern, O.: 1970, Predictability of Stock Market
Prices, Heath, Lexington, MA.
Granger, C. W. J. and Newbold, P.: 1974, Spurious regressions in econometrics,
27
Granger, C. W. J. and Swanson, N. R.: 1996, Further developments in the
study of cointegrated variables, Oxford Bulletin of Economics and Statis-
tics 58, 374386.
Granger, C. W. J. and Weiss, A. A.: 1983, Time series analysis of error-correction
models, in S. Karlin, T. Amemiya and L. A. Goodman (eds), Studies in
Econometrics, Time Series and Multivariate Statistics, in Honor of T.W.
Anderson, Academic Press, San Diego, pp. 255278.
Hall, A. D., Anderson, H. M. and Granger, C. W. J.: 1992, A cointegration
analysis of treasury bill yields, Review of Economics and Statistics 74, 116
126.
Hatanaka, M.: 1996, Time-Series-Based Econometrics. Unit Roots and Cointe-
grations, Oxford University Press, Oxford.
Hendry, D. F., Muellbauer, J. N. J. and Murphy, A.: 1990, The econometrics
of DHSY, in J. D. Hey and D. Winch (eds), A Century of Economics. 100
Years of the Royal Economic Society and the Economic Journal, Blackwell,
Oxford, pp. 298334.
Hylleberg, S., Engle, R. F., Granger, C. W. J. and Yoo, B. S.: 1990, Seasonal
cointegration, Journal of Econometrics 44, 215238.
Johansen, S.: 1988, Statistical analysis of cointegration vectors, Journal of Eco-
nomic Dynamics and Control 12, 231254.
Johansen, S.: 1991, Estimation and hypothesis testing of cointegration vectors in
Gaussian vector autoregressive models, Econometrica 59, 15511580.
Johansen, S.: 1995, Likelihood-Based Inference in Cointegrated Vector Autore-
gressive Models, Oxford University Press, Oxford.
Johansen, S. and Juselius, K.: 1990, Maximum likelihood estimation and infer-
ence on cointegration - with application to the demand for money, Oxford
Bulletin of Economics and Statistics 52, 169210.
King, R. G., Plosser, C. I., Stock, J. H. and Watson, M. W.: 1991, Stochastic
trends and economic uctuations, American Economic Review 81, 819840.
Kwiatkowski, D., Phillips, P. C. B., Schmidt, P. and Shin, T.: 1992, Testing the
null hypothesis of stationarity against the alternative of a unit root: How
sure are we that economic series have a unit root?, Journal of Econometrics
54, 159178.
Lettau, M. and Ludvigson, S.: 2001, Consumption, aggregate wealth, and ex-
pected stock returns, Journal of Finance 56, 815850.
28
Lettau, M. and Ludvigson, S.: 2003, Understanding trend and cycle in asset
values: Reevaluating the wealth eect on consumption, Working paper 9848,
National Bureau of Economic Research.
Lo, M. C. and Zivot, E.: 2001, Threshold cointegration and nonlinear adjustment
to the law of one price, Macroeconomic Dynamics 5, 533576.
Mandelbrot, B.: 1963, The variation of certain speculative prices, Journal of
Business 36, 394419.
Mandelbrot, B. and Taylor, H.: 1967, On the distribution of stock price dier-
ences, Operations Research 15, 10571062.
Manganelli, S. and Engle, R. F.: 2001, Value at risk models in nance, Working
Paper 75, European Central Bank.
Meese, R. and Rogo, K.: 1988, Was it real? The exchange rate dierential
relation over the modern oating exchange rate period, Journal of Finance
43, 933948.
Nelson, D. B.: 1991, Conditional heteroskedasticity in asset returns: A new
approach, Econometrica 59, 347370.
Noh, J., Engle, R. F. and Kane, A.: 1994, Forecasting volatility and option
pricing of the SP 500 index, Journal of Derivatives 2, 1731.
Palm, F. C.: 1996, GARCH models of volatility, in G. Maddala and C. Rao (eds),
Handbook of Statistics 14: Statistical Methods in Finance, Vol. 14, Elsevier,
Amsterdam, pp. 209240.
Phillips, A. W.: 1957, Stabilization policy and the time forms of lagged responses,
Economic Journal 67, 265277.
Phillips, P. C. B.: 1986, Understanding spurious regressions, Journal of Econo-
metrics 33, 311340.
Phillips, P. C. B. and Perron, P.: 1988, Testing for a unit root in a time series
regression, Biometrika 75, 335346.
Poon, S.-H. and Granger, C. W. J.: 2003, Forecasting volatility in nancial
markets: A review, Journal of Economic Literature 41, 478539.
Rachev, S. and Mittnik, S.: 2000, Stable Paretian Models in Finance, Wiley,
Chichester.
Saikkonen, P. and Luukkonen, R.: 1993, Testing for a moving average unit root in
autoregressive integrated moving average models, Journal of the American
Statistical Association 88, 596601.
29
Sargan, J. D.: 1964, Wages and prices in the United Kingdom: A study in econo-
metric methodology, in P. E. Hart, G. Mills and J. N. Whittaker (eds),
Econometric Analysis for National Economic Planning, Butterworths, Lon-
don.
Sarno, L. and Taylor, M. P.: 2002, Purchasing power parity and the real exchange
rate, IMF Sta Papers 49, 65105.
Schwert, G. W.: 1990, Stock volatility and the crash of 87, Review of Financial
Studies 3, 77102.
Shephard, N. G.: 1996, Statistical aspects of ARCH and stochastic volatility, in
D. R. Cox, D. V. Hinkley and O. E. Barndor-Nielsen (eds), Time Series
Models. In Econometrics, Finance and Other Fields, Chapman and Hall,
London, pp. 167.
Shin, Y.: 1994, A residual-based test of the null of cointegration against the
alternative of no cointegration, Econometric Theory 10, 91115.
Sims, C. A.: 1980, Macroeconomics and reality, Econometrica 48, 148.
Stock, J. H.: 1987, Asymptotic properties of least squares estimators of cointe-
grating vectors, Econometrica 55, 10351056.
Tanaka, K.: 1996, Time Series Analysis. Nonstationary and Noninvertible Dis-
tribution Theory, Wiley, New York.
Taylor, S. J.: 1982, Financial returns modeled by the product of two stochastic
processes - a study of the daily sugar prices 1961-75, in O. D. Anderson (ed.),
Time Series Analysis: Theory and Practice, North-Holland, Amsterdam,
pp. 203226.
Taylor, S. J.: 1986, Modelling Financial Time Series, Wiley, Chichester.
Tse, Y. K. and Tsui, A. K. C.: 2002, A multivariate generalized autoregressive
conditional heteroscedasticity model with time-varying correlations, Journal
of Business and Economic Statistics 20, 351362.
Watson, M. W.: 1994, Vector autoregressions and cointegration, in R. F. En-
gle and D. L. McFadden (eds), Handbook of Econometrics, Vol. 4, North-
Holland, Amsterdam, pp. 28442915.
Xiao, Z. and Phillips, P. C. B.: 2002, A CUSUM test for cointegration using
regression residuals, Journal of Econometrics 108, 4361.
30
Advanced information on the Prize in Economic Sciences 2002, 17 December 2002


Foundations of Behavioral and Experimental Economics:
Daniel Kahneman and Vernon Smith

Until recently, economics was widely regarded as a non-experimental science that had to rely
on observation of real-world economies rather than controlled laboratory experiments. Many
commentators also found restrictive the common assumption of a homo oeconomicus
motivated by self-interest and capable of making rational decisions. But research in
economics has taken off in new directions. A large and growing body of scientific work is
now devoted to the empirical testing and modification of traditional postulates in economics,
in particular those of unbounded rationality, pure self-interest, and complete self-control.
Moreover, todays research increasingly relies on new data from laboratory experiments
rather than on more traditional field data, that is, data obtained from observations of real
economies. This recent research has its roots in two distinct, but converging, traditions:
theoretical and empirical studies of human decision-making in cognitive psychology, and tests
of predictions from economic theory by way of laboratory experiments. Today, behavioral
economics and experimental economics are among the most active fields in economics, as
measured by publications in major journals, new doctoral dissertations, seminars, workshops
and conferences. This years laureates are pioneers of these two fields of research.

Human decision-making deviates in one way or another from the standard assumptions of the
rationalistic paradigm in economics. If such deviations from rationality and self-interest were
small and purely idiosyncratic, they would on average cancel out, and economic theory would
not be too wide off the mark when predicting outcomes for large aggregates of agents.
Following the lead of Vernon Smith, early studies of alternative market mechanisms by
experimental economists can be viewed as tests of the hypothesis of idiosyncratic deviations
2
from standard economic theory. If deviations from rationality and self-interest were
systematic, however, this would call for a revision of economic theory itself. Following the
lead of Daniel Kahneman and the late Amos Tversky, early studies of human decision-making
by cognitive psychologists can be seen as testing hypotheses of systematic deviations from
rationality.

This text begins by addressing Vernon Smiths contributions to the field of experimental
economics. It then considers Daniel Kahnemans findings in the field now known as
behavioral economics. The final sections summarize these contributions and their
importance, and offer some suggestions for further reading.

1. Foundations of experimental economics

Traditionally, economics has been viewed as a non-experimental science that had to rely
exclusively on field data:

Economics cannot perform the controlled experiments of chemists or biologists
because [it] cannot easily control other important factors. Like astronomers or
meteorologists, [it] generally must be content largely to observe. (Samuelson and
Nordhaus, 1985, p. 8)

Many perceived this as an obstacle to the continued development of economics as a science.
Unless controlled experiments could be carried out, tests of economic theory would remain
restricted. Solely on the basis of field data, it is difficult to decide whether and when a theory
fails, and to pinpoint the aspects responsible for this failure. The feedback channel between
theory and observation under controlled circumstances where new experimental findings
suggest new theories and new theories suggest new experiments seemed to be largely
unavailable to economics.

The establishment of a growing research field called experimental economics has radically
challenged this view.
1
Under controlled laboratory conditions, experimentalists study human
behavior in situations that, in simplified and pure forms, mimic those encountered in markets
3
and other forms of economic interaction. The extent to which the results of such experiments
can be generalized to market situations is still under debate. But the notion that laboratory
results concerning microeconomic behavior can crucially inform the development of
economic theory is basically the same as the notion that laboratory results concerning small-
scale phenomena in physics (such as those pertaining to elementary particles and
thermodynamics) can crucially inform the development of theoretical physics (with regard to
the universe or the weather).
2

Experimental research in economics has early predecessors. More than fifty years ago,
Chamberlin (1948) attempted to test the neoclassical theory of perfect competition by way of
experiments, and 1994 economics laureate Reinhard Selten conducted early experimental
studies of price formation in oligopoly markets, the first paper being Sauerman and Selten
(1959). There are also early studies on the predictive power of game theory in an
experimental setting, by John Nash also a 1994 economics laureate with colleagues
(Kalish, Milnor, Nash and Nehrig, 1954) and by Flood (1959). Furthermore Siegel and
Fouraker (1960) and Fouraker and Siegel (1963) reported experimental results on bargaining.

Without any doubt, however, the main researcher in the experimental tradition is Vernon
Smith. Smith not only made the most important early contributions, but has also remained a
key figure in the field to date. He has educated and collaborated with a large number of
younger researchers in experimental economics. The most prominent of these is Charles
Plott, who has also made important contributions to the field.

1
New panel data sets and advancements in econometrics, such as those recognized in the awards to Heckman
and Mc Fadden in 2000, have also substantially improved the potential for convincing causal inference from
observational data.
2
External validity requires that the results uncovered in the laboratory be valid across time and space. This may
be a stronger assumption in economics than in meteorology or astrophysics, but theories about the atmosphere or
the big bang that build on experimental results also have to resort to the same kind of assumption.
4

1.1. Market mechanisms
Vernon Smiths most significant work concerns market mechanisms. He laid the groundwork
for this research area in innovative experiments with competitive markets (Smith, 1962), in
tests of different auction forms (Smith, 1965, 1976b, Coppinger, Smith and Titus, 1980), and
in the design of the so-called induced-value method (Smith, 1976a).

Smiths first experimental article (Smith, 1962) was inspired by Chamberlins (1948)
classroom experiments. Chamberlin, who was Smiths teacher at Harvard at the time, had let
participants engage in pairwise bargaining, acting as buyers and sellers of a fictitious good.
Chamberlin regarded his experimental results as a falsification of the standard neoclassical
model of a market under perfect competition (that is, with price-taking and rational agents).

Smith realized that Chamberlins results would be more compelling if the participants were
placed in a setting more similar to a real market. He thus set up an experiment where subjects
were divided into groups of potential sellers and buyers in a so-called double oral auction, a
market mechanism used in many financial and commodity markets. Subjects were randomly
assigned the roles of seller and buyer, and each seller was given one unit of the good to be
sold, and a reservation price for this unit. If the reservation price was v for the unit, the
seller was not allowed to sell below that price, while she would earn pv dollars by selling at
a price p>v. A sellers reservation price v was her own private information. Similarly, each
buyer was assigned a private reservation price w, the highest price at which he was allowed to
buy a unit. Purchases at a price p<w resulted in earnings of w-p dollars. Based on the
distribution of reservation prices which he had chosen, Smith could draw a supply and a
demand schedule and locate the competitive equilibrium price as their intersection. The
subjects, by contrast, did not have this information and were thus not able to compute the
theoretical equilibrium price. Much to his surprise, Smith found that the actual trading prices
came close to the theoretical equilibrium price, hence supporting the theory that the
experiments were initially supposed to reject.

The result from one of his experiments is illustrated in Figure 1 (Smith, 1962, Chart 1, p.
113). The left-hand panel shows the demand and supply schedules induced by the given
distribution of reservation prices. The schedules intersect at p = 2.00, which is thus the
competitive equilibrium price. The right-hand panel shows the trading prices in five
5
successive trading periods, as well as the standard deviation of the price distribution in each
period, expressed as a percentage of the theoretical equilibrium price (the number in the
diagram). As is seen in this diagram, most trading prices were close to the theoretical
prediction, and the standard deviation fell over time as the prices converged towards the
theoretical prediction.

Figure 1
Smith concluded that
there are strong tendencies for a competitive equilibrium to be
attained as long as one is able to prohibit collusion and to maintain
absolute publicity of all bids, offers, and transactions. Changes in
the conditions of supply and demand cause changes in the volume of
transaction per period and the general level of contract prices. These
latter correspond reasonably well with the predictions of competitive
price theory. (Smith, 1962, p. 134).

Smith and other researchers subsequently carried out a series of similar experiments to check
whether this agreement with theory was a mere coincidence. Later experiments continued to
confirm Smiths original result. In joint work, Plott and Smith (1978) obtained the same
general result, but added an important twist: market institutions do matter. Specifically,
6
they compared the outcomes when sellers and buyers were allowed to change prices
continuously during a trading period (Smiths original design) with the outcomes when they
had to post a price for an entire trading period. The latter design turned out to result in a
slower convergence towards the theoretical equilibrium price. The experimental approach, as
opposed to collecting field data, was essential in driving home this result; it made it possible
to hold constant the market environment (in this case the distribution of reservation prices)
while varying the market institution (in this case the rules for price adjustment) in a
controlled fashion.

In almost any market experiment, a clear test of the hypothesis in question requires
controlling for the subjects preferences. This is a major difficulty, as selling and buying will
generally be influenced by the subjects idiosyncratic evaluations of gains and losses,
evaluations that are not directly observable to the researcher. This problem was first
addressed by Chamberlin (1948), who suggested a method for resolving it, essentially by
providing the subjects with the right monetary incentives. This so-called induced-value
method was developed further by Smith (1976a)
3
, and has now become a standard tool in
experimental economics.

In order to illustrate this method, consider a subject assigned the role of buyer in a market for
a homogeneous good (where all units are identical). Suppose that the experimentalist wants
this subject to express a certain demand function D . That is, at any price p , the subject
should be willing to buy precisely q = D(p) units. But the experimentalist does not know the
subjects utility of wealth, u(w). Smiths method induces the desired demand function by
rewarding the subject with R(q) - pq dollars for any quantity q bought at price p , where R
is a suitably chosen reward function. According to economic theory, the subject will choose
the quantity q such that her marginal benefit from increasing q equals her marginal cost of
doing so, that is, such that R(q) = p.
4
As long as the unknown utility function u is
increasing and concave, her demand will coincide with the desired demand function if, for
any relevant price, the inverse derivative of the reward function R is set equal to the desired
demand function, that is, if (R)
1
(p) = D(p) for all relevant prices p. Similar methods have
been applied ever since in the experimental literature.

3
Smith had sketched this method in an earlier working paper (Smith, 1973).
4
If the quantity q maximizes the subjects utility of wealth, u(R(q)-pq), then the first-order condition
7

1.2. Tests of auction theory
Auction theory has emerged as one of the most successful developments in microeconomic
theory and game theory since the early 1960s. A number of precise theoretical results for a
variety of auction forms were developed by the late economics laureate William Vickrey,
followed by a number of younger researchers (see Krishna, 2002, for an overview). Smith
initiated the experimental testing of many of these propositions, and has published extensively
on the subject (see, for example, Smith, 1976b, Coppinger, Smith and Titus, 1980, and Cox,
Robertson and Smith, 1982). Moreover, he pioneered the use of controlled laboratory
experiments as wind tunnel tests of new auction designs for which precise theoretical
predictions are hard to obtain before they are used in practice (see section 1.3).

As the term is commonly understood, auctions may seem of little importance for real-world
economies. However, by proceeding from simpler to more complex auction forms, theory has
deepened our understanding of the functioning of many real-world markets. Even some of the
simpler auction forms studied in theory are widely used in practice, particularly in the context
of deregulation and privatization of natural monopolies, public procurement, the sale of
government bonds, etc.

Central to Smiths experimental work on auctions are the established theoretical predictions
for certain auction forms used in the sale of a single object. Such auctions are traditionally
classified into four types. In an English or ascending auction, buyers announce their bids
sequentially and in an increasing order, until no higher bid is submitted. In a Dutch or
descending auction, a high initial bid by the seller is gradually lowered in fixed steps at fixed
times regulated by a clock, until some buyer shouts buy, whereupon the clock stops. Both
of these auctions are usually oral, and the trading price is the last (first) bidders bid. In the
other two auction forms, all bidders instead simultaneously submit their bids in sealed
envelopes and the unit for sale is allocated to the highest bidder. In the first-price sealed-bid
auction, this bidder pays his or her bid to the seller; while in the second-price sealed-bid
auction, this bidder pays only the second highest bid.

Microeconomic theory also distinguishes between auctions with private and common values.
In both cases, the value to each buyer is treated as a random variable. In the case of private

u(w)(R(q)-p)= 0 has to be met, granted u is differentiable.
8
values, these valuations are statistically independent across the population of potential bidders
the value to a buyer is his or her purely idiosyncratic valuation of the object. In common-
value auctions, by contrast, the value to the buyers also has a common component, such as a
resale market value or the conditions in some related market (examples include spectrum
auctions and telecommunication markets).

Economic theory makes the following three predictions in the case of private values: (1)
English and second-price auctions are equivalent, in terms of who will (probabilistically)
obtain the item and the expected revenue to the seller. This result follows from individual
rationality (more precisely, from assuming that bidders do not use weakly dominated
strategies). (2) Dutch and first-price auctions are equivalent, a result which follows from the
more restrictive assumption of Nash equilibrium behavior, that is, individual rationality
combined with interpersonally consistent expectations. (3) All four auction forms are
equivalent if all buyers are risk neutral (that is, if they are indifferent between participating in
an actuarially fair lottery and obtaining the expected lottery prize for sure; see also section 2).

Smith carried out many experiments once more, controlling for demand and supply
conditions, while varying the market institution in order to empirically test these and other
theoretical predictions.
5
In order to generate private values, each bidder was given a
randomly and independently drawn number, v, which was kept private to the bidder. If the
bidder won the auction and paid the price p, this subject would earn the monetary amount p
v. In regard to prediction (1) above, Smith discovered that English and sealed-bid second-
price auctions indeed produce similar experimental outcomes, just as theory says. As for (2),
Dutch and sealed-price first-price auctions did not give rise to equivalent outcomes, in
contrast with theory. In the case of (3), he found that models which presume that buyers have
identical attitudes toward risk could be rejected. Furthermore, he found that the average sale
price was higher in English and sealed-bid second-price auctions than in sealed-bid first-price
auctions, and that the latter yielded higher average selling prices than Dutch auctions.

Of these results, one of the most unexpected was that Dutch and sealed-bid first-price auctions
turned out to be unequivalent. Two theoretical explanations have been suggested. One is that
utility depends not only on the monetary outcome but also on the suspense of waiting in the
9
Dutch auction, the other that bidders underestimate the increased risk associated with waiting
in the Dutch auction. These and other possible reasons for the observed non-equivalence
between the two auctions are explored in Smith (1991b).

1.3. The laboratory as a wind tunnel

Smith, as well as Plott, initiated the use of the laboratory as a wind tunnel (a laboratory
setup used to test prototypes for aircraft) in order to study the performance of proposed
institutional mechanisms for deregulation, privatization, and the provision of public goods.
These mechanisms are usually so complex that existing theory does not provide precise
predictions, which makes the experimental method particularly useful. In a series of studies
(Smith, 1979a-c, 1980, and Coursey and Smith, 1984) he studied the design of incentive-
compatible mechanisms for the provision of public goods. In these experiments, Smith tested
the effectiveness of mechanisms proposed by economic theorists, as well as some of his own
variants. Smith has also done experimental work on mechanisms to allocate airport time slots
by means of computer-assisted markets (Bulfin, Rasenti and Smith, 1982, and McCabe,
Rasenti and Smith, 1989) and on alternative organizations of energy markets (Rasenti, Smith
and Wilson, 2001).

1.4. Experimental methodology
Apart from substantive results on markets and auctions, Smiths work has had an enormous
methodological impact. His seminal American Economic Review article Experimental
economics: Induced-value theory(Smith, 1976a), provided a practical and detailed guide to
the design of economic experiments in the laboratory and a motivation for these guidelines
(see also Smith, 1982). In recent years, this paper has served as a paradigm for experimental
scholars in economics.

The experimental method developed by Smith deviates from the experimental approach used
in psychology (cf. section 2). It emphasizes the importance of providing subjects with
sufficient monetary incentives, in order to outweigh the distorting effects of decision costs.

5
Smith (1976b) is a seminal paper on this topic. See also Coppinger, Smith and Titus (1980), who seem to have
been the first to test these propositions in a comparison of all four types of auction, and Cox, Roberson and
Smith (1982).
10
Smiths method also emphasizes the importance of designing experiments as repeated trials,
so that the subjects can become familiar with and understand the experimental situation.

In many respects, the differences vis--vis psychologically oriented methods are a matter of
focus. Whereas psychologists have been predominantly interested in individual behavior,
Smith designed his original experiments mainly to analyze market outcomes. Genuine
differences of opinion about the appropriate methodology have not subsided, however. To
some extent, they reflect two different approaches to understanding human behavior, as
further discussed in section 2 (see Smith, 1991a, and Loewenstein, 1999, for different sides of
the debate).
6

Be that as it may, Smiths approach to experimentation constitutes a vital contribution, of
relevance not only for economists but also for other social scientists. For instance, Plotts
experiments on decision-making in committees (Fiorina and Plott, 1978) followed much the
same approach and generated an extensive experimental literature in political science.

2. Foundations of behavioral economics

Nearly half a century ago, Edwards (1954) introduced decision-making as a research topic for
psychologists, outlining an agenda for future research. Allais (1953a,b) outlined a
psychology-based positive theory of choice under uncertainty, while Simon (1956) proposed
an approach to information processing and decision-making based on bounded rationality. But
research in cognitive psychology did not come into its own until Daniel Kahneman and Amos
Tversky (deceased in 1996) published their findings on judgment and decision-making.
Although adhering to the tradition of cognitive psychology, Kahnemans research has equally
well been directed towards economists. Many of his articles have been published in
economics journals; one article, Kahneman and Tversky (1979), even has the highest citation
count of all articles published in Econometrica, by many considered the most prestigious
journal in economics. Given the barriers to communicating across traditional disciplines,
considerable effort has gone into building a bridge between research in economics and
psychology. Nowadays, there are in fact two bridges between these disciplines one built

6
The importance of monetary incentives or repetition obviously depends on the hypothesis that the experiment is
supposed to test. Incentives may also affect different cognitive functions in distinct ways (Nilsson, 1987).
11
around experimental methods and the other around theoretical modeling. Both serve as the
basis for the current wave of work in behavioral economics. Before discussing Kahnemans
specific contributions, the next section outlines some differences between conceptions of
decision-making in economics and psychology.

2.1. Decision-making in economics and psychology
Economists typically assume that market behavior is motivated primarily by material
incentives, and that economic decisions are governed mainly by self-interest and rationality.
In this context, rationality means that decision-makers use available information in a logical
and systematic way, so as to make optimal choices given the alternatives at hand and the
objective to be reached. It also implies that decisions are made in a forward-looking way, by
fully taking into account future consequences of current decisions. In other words, so-called
extrinsic incentives are assumed to shape economic behavior.

In psychology, especially cognitive psychology, a human being is commonly regarded as a
system, which codes and interprets available information in a conscious and rational way.
But other, less conscious, factors are also assumed to govern human behavior in a systematic
way. It is this more complex view where intrinsic incentives help shape human behavior
that has come to penetrate recent developments in economic theory.

Economists have traditionally treated a decision-makers preferences over available
alternatives as fixed and given. The decision-maker is assumed to form probabilistic beliefs
or expectations about the state of nature and the effects of her actions, and to process available
information according to statistical principles. More precisely, standard economic theory
relies on the expected-utility maximization approach founded by von Neumann and
Morgenstern (1944) and extended by Savage (1953). Here, it is presumed that for every
decision-maker there exists some real-valued function u, defined on the relevant set X of
outcomes x
1
,x
2
,x
I
, such that if one available action a results in probabilities p
i
over the
outcomes x
i
(for i=1,,I) and another available action b results in probabilities q
i
over the
same outcomes, then the decision-maker (strictly) prefers action a to action b if and only if
12
the statistically expected value of this utility function u is greater under a than under b.
7

Formally, the criterion for choosing a is thus

i
p
i
u(x
i
) >
i
q
i
u(x
i
) . (1)

Hence, given existing market conditions, which define the choice set available to the decision-
maker, the cognitive process is reduced to a problem of expectation formation and
maximization. The decision-maker is thus assumed to behave as if she correctly assigned
probabilities to relevant random events and chose an action that maximized the expected
value of her resulting utility.

By contrast, cognitive psychologists consider an interactive process where several factors may
influence a decision in a non-trivial way. These components include perception, which
follows its own laws, as well as beliefs or mental models for interpreting situations as they
arise. Intrinsic motives, such as emotions the state of mind of the decision-maker and
attitudes stable psychological tendencies to relate to a given phenomenon in ones
environment may influence a decision. Moreover, the memory of previous decisions and
their consequences serves as a critical cognitive function that also has a strong influence on
current decision-making. Given this complex view, human behavior is regarded as locally
conditioned to a given situation. Typically, behavior is adaptive; it is dependent on the
context and transitory perceptual conditions.

These differences between psychology and traditional economics also show up in research
methodology. While experiments in economics often emphasize the generality of a situation
and comprise monetary rewards and repeated trials, psychologists try to capture intrinsic
motivations and the mental processes at work in a particular decision situation, what has been
termed the framing of a decision problem.

Extensive behavioral evidence, collected by Kahneman and others through surveys and
experiments, calls the assumption of economic rationality into question, at least in complex
decision situations. A number of studies have uncovered a non-trivial amount of deviations
from the traditional model of rational economic behavior. For example, real-world decision-
makers do not always evaluate uncertain prospects according to the laws of probability, and

7
To be exact, the function u is not a utility function: such functions map decision alternatives (here actions) into
13
sometimes make decisions that violate the principles of expected-utility maximization
outlined above. Kahnemans major contributions concerning judgment and decisions under
uncertainty are discussed in the following.

2.2. Judgment under uncertainty: heuristics and biases
Kahneman and Tversky discovered how judgment under uncertainty systematically departs
from the kind of rationality postulated in traditional economic theory (Kahneman and
Tversky, 1972, 1973, Tversky and Kahneman, 1971, 1973, 1974, 1982). A basic notion
underlying much of Kahneman and Tverskys early research is that people in general are
frequently unable to fully analyze situations that involve economic and probabilistic
judgments. In such situations, human judgment relies on certain shortcuts or heuristics, which
are sometimes systematically biased.

One fundamental bias is that individuals appear to use a law of small numbers, attributing the
same probability distribution to the empirical mean value from small and large samples,
thereby violating the law of large numbers in probability theory (Tversky and Kahneman,
1971). For example, in a well-known experiment it was found that subjects thought it equally
likely that more than 60 percent of births on a given day would be boys in a small hospital as
in a large hospital. In general, people do not appear to realize how fast the variance of the
sample mean of a random variable decreases with sample size.

More precisely, according to the statistical laws of large numbers, the probability distribution
of the mean from a large sample of independent observations of a random variable is
concentrated at the expected value of the random variable, and the variance of the sample
mean goes to zero as the sample size increases.
8
According to the psychological law of small
numbers, by contrast, people believe that the mean value from a small sample also has a
distribution concentrated at the expected value of the random variable. This leads to over-
inference from short sequences of independent observations.

the real numbers.
8
According to the most basic version of the law of large numbers, the following claim is essentially true for all
>0 and for any infinite sequence of independent and identically distributed random variables with mean : the
probability that the sample mean will deviate more than from goes to zero as the sample size goes to infinity.
14
An example of the law of small numbers is when an investor observes a fund manager
performing above average two years in a row and concludes that the fund manager is much
better than average, while the true statistical implication is very weak. A related example is
the so-called gamblers fallacy: many individuals expect the second draw of a random
mechanism to be negatively correlated with the first, even if the draws are statistically
independent. If a few early tosses of a fair coin give disproportionately many heads, many
individuals believe that the next flip is more likely to be tails. Recent work, such as Rabin
(2002), describes the importance of the law of small numbers for economic decisions.

The law of small numbers is related to representativeness, a heuristic which Kahneman and
Tversky discovered to be an important ingredient in human judgment. Tversky and
Kahneman (1973, 1974, 1982) illustrated the function of this heuristic in several elegant
experiments. Subjects were asked to categorize persons, e.g., as a salesman or a member
of parliament, on the basis of given descriptions. Confronted with a description of an
individual randomly drawn from a given population as interested in politics, likes to
participate in debates, and is eager to appear in the media, most subjects would say that the
person is a member of parliament, even though the higher proportion of salespersons in the
population makes it more likely that the person is a salesman. This observed heuristic way of
thinking was examined further by Tversky and Kahneman (1973), who report an experiment
where some subjects received explicit information about the true proportions in the
population. One design stated that the person to be categorized was drawn from a pool of 30
percent engineers and 70 percent lawyers, while another design reversed these proportions.
The results revealed that this difference had virtually no effect on subjects judgment.

The same heuristic can also prompt people to believe that the joint probability of two events is
larger than the probability of one of the constituent events, in contradiction to a fundamental
principle of probability (the so-called conjunction rule). For instance, some subjects in an
experiment thought that if Bjrn Borg reached the Wimbledon final, he would be less likely to
lose the first set than to lose the first set and win the match.

In an overview of behavioral finance, Shleifer (2000) argues that the law of small numbers
and representativeness may explain certain anomalies in financial markets. For example, the
excess sensitivity of stock prices (Shiller, 1981) may be a result of investors overreacting to
short strings of good news.
15

Another bias common in probabilistic judgment is availability, whereby people judge
probabilities by the ease of conjuring up examples. The result is that disproportionately high
weight is assigned to salient or easily remembered information (Tversky and Kahneman,
1973). People thus overstate, say, the probability of violent crimes in a city if they personally
know someone who has been assaulted, even if they have access to more relevant aggregate
statistics. A general finding in cognitive psychology is that, compared to unfamiliar
information, familiar information is more easily accessible from memory and is believed to be
more real or relevant. Familiarity and availability may thus serve as cues for accuracy and
relevance. Therefore, mere repetition of certain information in the media, regardless of its
accuracy, makes it more easily available and therefore falsely perceived as more accurate.

Such evidence on human judgment demonstrates that peoples reasoning violates basic laws
of probability in a systematic way. By demonstrating this, Kahneman's research has seriously
questioned the empirical validity of one of the fundamentals of traditional economic theory.

2.3. Decision-making under uncertainty: prospect theory
Available evidence indicates that not only judgment, but also decision-making under
uncertainty departs in a systematic way from traditional economic theory. In particular, many
decisions under uncertainty diverge from the predictions of expected-utility theory.

Departures from the von Neumann-Morgenstern-Savage expected-utility theories of decisions
under uncertainty were first pointed out by the 1988 economics laureate Maurice Allais
(1953a), who established the so-called Allais paradox (see also Ellsberg, 1961, for a related
paradox). For example, many individuals prefer a certain gain of 3,000 dollars to a lottery
giving 4,000 dollars with 80% probability and 0 otherwise. However, some of these same
individuals also prefer winning 4,000 dollars with 20% probability to winning 3,000 dollars
with 25% probability, even though the probabilities for the gains were scaled down by the
same factor, 0.25, in both alternatives (from 80% to 20%, and from 100% to 25%). Such
preferences violate the so-called substitution axiom of expected-utility theory.
9
Kahneman
has provided extensive evidence of departures from the predictions of expected utility (see

9
By this axiom, if a decision-maker prefers lottery A to B, he should also prefer a probability mixture pA + (1-
p)C to the probability mixture pB + (1-p)C , for all lotteries C.
16
Kahneman and Tversky, 1979, Tversky and Kahneman, 1991, 1992, Kahneman and Lovallo,
1993, and Kahneman, Knetsch and Thaler, 1990).

One striking finding is that people are often much more sensitive to the way an outcome
differs from some non-constant reference level (such as the status quo) than to the outcome
measured in absolute terms. This focus on changes rather than levels may be related to well-
established psychophysical laws of cognition, whereby humans are more sensitive to changes
than to levels of outside conditions, such as temperature or light.

Moreover, people appear to be more adverse to losses, relative to their reference level, than
attracted by gains of the same size. Tversky and Kahneman (1992) estimated that the value
attached to a moderate loss is about twice the value attached to an equally large gain. That is,
peoples preferences seem to be characterized by (local) loss aversion. With small stakes,
they generally prefer the status quo to a fifty-fifty chance of winning, say, 12 dollars or losing
10 dollars. This renders counterfactual the implied preferences over large gains and losses,
according to conventional economic analysis; see Rabin (2000). The common finding of
apparently risk-loving behavior with respect to large losses is inconsistent with the traditional
assumption of risk aversion.
10
For example, Kahneman and Tversky (1979) found that seven
out of ten people prefer a 25% probability of losing 6,000 dollars, to a 50% probability of
losing either 4,000 or 2,000 dollars, with equal probability (25%) for each. Since the
expected monetary value of the two lotteries is the same, the first lottery is a mean-preserving
spread of the second, and should thus not be preferred under conventional risk aversion.

Kahneman and Tversky moved beyond criticism, however, and suggested an alternative
modeling framework in their seminal article, Prospect Theory: An Analysis of Decisions
under Risk (1979). While expected-utility theory is axiomatic, their prospect theory is
descriptive. It was thus developed in an inductive way from empirical observations, rather
than deductively from a set of logically appealing axioms. Later, Tversky and Kahneman
(1986) argued that two theories are in fact required: expected-utility theory to characterize
rational behavior and something like prospect theory to describe actual behavior. Although
expected-utility theory provides an accurate representation of actual choices in some

10
As explained in Section 1, a decision-maker is called risk neutral if she is indifferent between participating in
any actuarially fair lottery and obtaining the expected prize for sure. A decision-maker who prefers the expected
prize for sure is called risk averse, while she is called risk loving if she prefers the lottery.
17
transparent and simple decision problems, most real-life decision problems are complex and
call for behaviorally richer models.

What, then, are the differences between the two theories? In the case of monetary gains and
losses, the decision criterion in expected-utility theory, equation (1) above, presumes the
existence of a real-valued function u of wealth w, for the decision-maker in the current
situation. If action a induces probabilities p
i
over the different levels w
i
of wealth, and
action b induces probabilities q
i
, then the decision-maker (strictly) prefers a to b if and
only if

i
p
i
u(w
i
) >
i
q
i
u(w
i
) . (2)

By contrast, prospect theory postulates the existence of two functions, v and , such that the
decision-maker (strictly) prefers action a over action b if and only if

i
(p
i
)v(w
i
) >
i
(q
i
)v(w
i
) , (3)

where w
i
=w
i
-w
o
is the deviation in wealth from some reference level w
o
(which may be
initial or aspired wealth, see below).

There are three differences between the two models. First, in prospect theory, the decision-
maker is not concerned with final values of wealth per se, but with changes in wealth, w,
relative to some reference point. This reference point is often the decision-makers current
level of wealth, so that gains and losses are defined relative to the status quo. But the
reference level can also be some aspiration level: a wealth level the subject strives to acquire,
given his or her current wealth and expectations. Kahneman and Tversky argued that a
decision problem has two stages. It is edited, so as to establish an appropriate reference
point for the decision at hand. The outcome of such a choice is then coded as a gain when it
exceeds this point and as a loss when the outcome falls short of it. This editing stage is
followed by an evaluation stage, which is based on the criterion in (3).

The second difference relative to expected-utility theory concerns the value function v . In
addition to being defined over changes in wealth, this function is S-shaped. Thus it is concave
for gains and convex for losses, displaying diminishing sensitivity to change in both
directions. Furthermore, it has a kink at zero, being steeper for small losses than for small
18
gains. The function u in expected-utility theory, by contrast, is usually taken to be smooth
and concave everywhere. The form of the value function is illustrated in Figure 2 (Figure 3
in Kahneman and Tversky, 1979).

Third, the decision-weight function is a transformation of the objective probabilities p and
q . This function is monotonically increasing, with discontinuities at 0 and 1, such that it
systematically gives overweight to small probabilities and underweight to large probabilities.
Its typical shape is illustrated in Figure 3 (Figure 4 in Kahneman and Tversky, 1979).

Figure 2 Figure 3

These differences make prospect theory consistent with the experimental evidence mentioned
earlier in thissection. Since people evaluate risky prospects on the basis of changes in wealth
relative to some reference level, appropriate assumptions about the editing stage would make
the model consistent with the common observation that people choose differently depending
on how a problem is framed. The kink on the value function at the reference point making
the function much steeper for small losses than for small gains implies that choices are
consistent with loss aversion. As a consequence of the diminishing marginal sensitivity to
change in the v function, decision-makers become risk averse towards gains (they value large
gains less than proportionally) and risk loving towards losses (they value large losses less than
proportionally), in line with the evidence. Moreover, the fact that the decision-weight
function overweighs small probabilities and underweighs large probabilities can explain the
Allais paradox.

19
Already Allais (1953 a,b) outlined foundations for a psychologically based theory of
preferences over uncertain prospects with monetary outcomes. Unlike prospect theory, Allais
attached (cardinal) utilities to final wealth levels, but like prospect theory, he made a
distinction between objective probabilities and the decision-maker's perception of these.
Allais suggested that objective probabilities be transformed differently for gains and losses, in
such a way that the perceived probabilities sum to one.

Prospect theory may also capture several regularities that appear as anomalies from the
perspective of traditional economic theory: the propensity for people to take out expensive
small-scale insurance when buying appliances; their willingness to drive to a distant store to
save a few dollars on a small purchase, but reluctance to make the same trip for an equally
large discount on an expensive item; or their resistance to lowering consumption in response
to bad news about lifetime income.

In sum, the empirical work conducted by Kahneman and others indicates several regularities
in choice under uncertainty, and the ideas incorporated in prospect theory go a long way
towards explaining these regularities. Kahnemans results have provided researchers in
economics with new insights and have been instrumental in subsequent model building by
alerting decision analysts to the errors commonly committed by real-life decision-makers. A
further extension of prospect theory, known as cumulative prospect theory (Tversky and
Kahneman, 1992) addresses some weaknesses of the original version. In particular,
cumulative prospect theory allows for prospects with a large number of outcomes, and it is
consistent with stochastic dominance.
11

Prospect theory and its extensions have taken important steps towards a more accurate
description of individual behavior under risk than expected-utility theory. It now forms the
basis for much of the applied empirical work in this field.

11
Cumulative prospect theory combines prospect theory with a cumulative approach developed by Quiggin
(1982), Schmeidler (1989) and Luce and Fishburn (1991).
20

3. Summary

Daniel Kahneman has used insights from cognitive psychology regarding the mental
processes of answering questions, forming judgments, and making choices, to help us better
understand how people make economic decisions. Other psychologists have also made
important contributions along the same lines. But Kahnemans work with Tversky on
decision-making under uncertainty clearly stands out as the most influential. Kahneman also
made early contributions to other areas of behavioral economics. One example is his joint
work with Knetsch and Thaler (Kahneman, Knetsch and Thaler, 1986) on the importance of
fairness considerations. This has become a lively field of research, and many experimental
studies have subsequently been carried out by other researchers, showing that a variety of
market behaviors can be derived from considerations of fairness and reciprocity (see e.g. Fehr
and Falk, 2002 for a recent review). Through this and other work, Kahneman has been a
major source of inspiration behind the recent boom of research in behavioral economics and
finance. His research has also had a substantial impact in other fields. It is widely quoted in
other social sciences as well as within the natural sciences, the humanities and medicine.

Vernon Smith is the most influential figure in launching experiments as an empirical
methodology in economics. Unlike Kahneman, he did not start out by challenging the
traditional economic theory of rational decision-making. Rather, he tested alternative
hypotheses regarding market performance, in particular the importance of different market
institutions. While Kahnemans surveys and experiments have mainly focused on decisions
by individual agents, Smith has focused his experiments on the interaction between agents in
specific market environments. He has also emphasized methodological issues, developing
practical experimental methods and establishing standards for what constitutes a good
experiment.
12
Other researchers have furthered this tradition. Charles Plott, in particular, has
written several important papers, further developed the experimental methodology and spear-
headed experimental research in new areas. But it is largely through Smith's achievement that
many economists have come to view laboratory experiments as an essential tool.

12
Since experimentation with human subjects had been a well-established method in psychology for almost a
century, it was more important for Smith than for Kahneman to develop experimental methodology.
21
A current wave of research draws on the combined traditions of psychology and experimental
economics. This new research is potentially significant for all areas of economics and
finance. Experimental evidence indicates that certain psychological phenomena such as
bounded rationality, limited self-interest, and imperfect self-control are important factors
behind a range of market outcomes. To the extent parsimonious behavioral theories,
consistent with this evidence, can be developed, they may eventually replace elements of
traditional economic theory. A challenging task in financial economics is to consider the
extent to which the effects of systematic irrationality on asset prices will be weeded out by
market arbitrage.

Although Kahneman's and Smith's research agendas differ in many respects, their combined
scientific contributions have already changed the direction of economic science. Economics
used to be limited to theorizing by way of a relatively simple rationalistic model of human
decision-making, homo oeconomicus, and to empirical work on field data. When they
appeared, Kahnemans and Smiths initial works were received with skepticism by the
scientific community in economics. It took considerable time and much further research
before their main ideas seriously began to penetrate the profession. It is their achievement
that many perhaps most economists today view psychological insights and experimental
methods as essential ingredients in modern economics.

4. Suggestions for further reading

Smith (1962) and Kahneman and Tversky (1979) are two classical articles by this years
laureates. For collections of papers we refer to Smith (2000) and Kahneman and Tversky,
eds. (2000). Overviews of the fields are given in Kagel and Roth, eds. (1995), and in Rabin
(1998).

22
References

Allais M. (1953a), Le comportement de lhomme rationnel devant le risque: Critique des
postulats et axiomes de lcole americaine, Econometrica 21, 503-546.

Allais M. (1953b), Fondements d'une Thorie Positive des Choix Comportant un Risque et
Critique des Postulats et Axiomes de lEcole Americaine, in Econometrie, Colloques
Internationaux du Centre de la Recherche Scientifique, Volume XL, 257-332, Paris.
(Translated into English in M. Allais and O. Hagen (eds.): Expected Utility Hypotheses
and the Allais Paradox, D. Reidel Publishing Company, Dortrecht, Holland, 1979.)

Bulfin R.L., S.J. Rasenti, and V.L. Smith (1982), A combinatorial auction mechanism for
airport time slot allocation, Bell Journal of Economics 13, 402-417.

Chamberlin E.H. (1948), An experimental imperfect market, Journal of Political Economy
56, 95-108.

Coppinger V.M., V.L Smith, and J.A. Titus (1980), Incentives and behavior in English,
Dutch, and sealed-bid auctions, Economic Inquiry 18, 1-22.

Coursey D.L. and V.L. Smith (1984), Experimental tests of an allocation method for public,
private, or externality goods, Scandinavian Journal of Economics 86, 468-484.

Cox J.L., B. Roberson, and V.L Smith (1982), Theory and behavior of single object
auctions, in V.L. Smith (ed.) Research in Experimental Economics, JAI Press,
Greenwich CT.

Edwards W. (1954), Behavioral decision theory, Annual Review of Psychology 12, 473-98.

Ellsberg D. (1961), Risk, ambiguity, and the Savage axioms, Quarterly Journal of

Fehr E. and A. Falk (2002), Psychological foundations of incentives. 2001 Schumpeter
Lecture, European Economic Review 46, 687-724.

Fiorina M. and C. Plott (1978), Committee decisions under majority rule: An experimental
study, American Political Science Review 72, 575-98.

Flood M. (1959), Some experimental games, Management Science 5, 5-26.

Fouraker L.E. and S.Siegel (1963), Bargaining Behavior, McGraw Hill, New York.

Kagel, J. H. and A. E. Roth, eds. (1995), Handbook of experimental economics, Princeton
University Press, Princeton.

Kahneman D., J. Knetsch, and R. Thaler (1986), Fairness and the assumptions of
economics, Journal of Business 59, S285-S300.
23

Kahneman D., J. Knetsch, and R. Thaler (1990), Experimental tests of the endowment effect
and the Coase theorem, Journal of Political Economy 98, 1325-1348.

Kahneman D. and D. Lovallo (1993), Timid choices and bold forecasts: A cognitive
perspective on risk taking, Management Science 39, 17-31.

Kahneman D. and A. Tversky (1972), Subjective probability: A judgment of
representativeness, Cognitive Psychology 3, 430-454.

Kahneman D. and A. Tversky (1973), On the psychology of prediction, Psychological
Review 80, 237-251.

Kahneman D. and A. Tversky (1979), Prospect theory: An analysis of decision under risk,

Kahneman, D, and A. Tversky, eds. (2000), Choices, values and frames, Cambridge
University Press, Cambridge.

Kalish G.K., J.W. Milnor, J. Nash, and E.D. Nehrig (1954), Some experimental n-person
games, in R.M. Thrall, C.H. Coombs, and R.L. Davis (eds.), Decision Processes,
Wiley, New York.

Krishna, V. (2002), Auction Theory, Academic Press.

Loewenstein G. (1999), Experimental economics from the vantage-point of behavioral
economics, Economic Journal 109, S25-S34.

Luce, R. D. and P. C. Fishburn (1991), "Rank-and sign-dependent linear utility models for
finite first-order gambles", Journal of Risk and Uncertainty 4, 29-59.

McCabe K.A., S.J. Rasenti, and V.L. Smith (1989), Designing 'smart' computer-assisted
markets, European Journal of Political Economy 5, 259-283.

von Neumann J. and O. Morgenstern (1944), Theory of Games and Economic Behavior (3
rd

ed.), Princeton University Press, Princeton.

Nilsson L-G. (1987), Motivated memory: Dissociations between performance data and
subjective reports, Psychological Research 46, 183-188.

Plott C. and V.L. Smith (1978), An experimental examination of two exchange institutions,

Quiggin, J. (1982), "A theory of anticipated utility", Journal of Economic Behavior and
Organization 3, 323-343.

Rabin, M. (1998), Psychology and Economics," Journal of Economic Literature 35, 11-46.

Rabin, M. (2000), Risk aversion and expected-utility theory: A calibration theorem,

24
Rabin M. (2002), Inference by believers in the law of small numbers, Quarterly Journal of
Economics 117, 775-816.

Rasenti S.J., V.L. Smith, and B.J. Wilson (2001), Controlling market power and price spikes
in electricity networks: Demand-side bidding, mimeo, George Mason University.

Samuelson, P. and W. Nordhaus (1985), Economics, McGraw-Hill.

Sauerman, H., and R. Selten (1959), Ein Oligopolexperiment, Zeitschrift fr die gesamte
Staatswissenschaft 115, 427-471.

Savage L. (1953), The Foundations of Statistics, John Wiley and Sons, New York.

Schmeidler, D. (1989), "Subjective probability and expected utility without additivity",

Shiller R. (1981), Do stock prices move too much to be justified by subsequent changes in
dividends?, American Economic Review 71, 421-436.

Shleifer A. (2000), Inefficient Markets An Introduction to Behavioral Finance. Clarendon
Lectures in Economics, Oxford University Press, Oxford.

Siegel S. and L.E. Fouraker (1960), Bargaining and Group Decision Making, McGraw Hill,
New York.

Simon H.A. (1956), Rational choice and the structure of the environment, Psychological
Review 63, 129-138.

Smith V.L. (1962), An experimental study of competitive market behavior, Journal of

Smith V.L. (1965), Experimental auction markets and the Walrasian hypothesis, Journal of

Smith V.L. (1973), Notes on some literature in experimental economics, Social Sciences
working paper No. 21, California Institute of Technology.

Smith V.L. (1976a), Experimental economics: Induced value theory, American Economic
Review, Papers and Proceedings, 274-279.

Smith V.L. (1976b), Bidding and auctioning institutions: Experimental results, in Y.
Amihud (ed.), Bidding and Auctioning for Procurement and Allocation, New York
University Press, New York.

Smith V.L. (1979a), An experimental comparison of three public good decision
mechanisms, Scandinavian Journal of Economics 81, 198-215.

Smith V.L. (1979b), Incentive compatible experimental processes for the provision of public
goods, in V.L. Smith (ed.), Research in Experimental Economics, JAI Press.
Greenwich, CT.

25
Smith V.L. (1979c), Indirect revelation of the demand for public goods: An overview and
critique, Scottish Journal of Political Economy 26, 183-189.

Smith V.L. (1980), Experiments with a decentralized mechanism for public good decisions,

Smith V.L. (1982), Microeconomic systems as an experimental science, American

Smith V.L. (1991a), Rational choice the contrast between economics and psychology,

Smith V.L. (1991b), Papers in Experimental Economics, Cambridge University Press,
Cambridge.

Smith, V.L. (2000), Bargaining and Market Behavior : Essays in Experimental Economics,
Cambridge University Press, Cambridge.

Tversky A. and D. Kahneman (1971), Belief in the law of small numbers, Psychological
Bulletin 76, 105-110.

Tversky A. and D. Kahneman (1973), Availability: A heursistic for judging frequency and
probability, Cognitive Psychology 5, 207-232.

Tversky, A. and D. Kahneman (1974), Judgment under uncertainty: Heuristics and biases,
Science 185, 1124-1131.

Tversky, A. and D. Kahneman (1982), Judgment of and by representativeness, in
Kahneman, D., P. Slovic, and A. Tversky (eds.). Judgment Under Uncertainty:
Heuristics and Biases, Cambridge University Press, Cambridge.

Tversky A. and D. Kahneman (1986), Rational choice and framing of decisions, Journal of
Business 59, S252-278.

Tversky A. and D. Kahneman (1991), Loss aversion in riskless choice: A reference-
dependent model, Quarterly Journal of Economics 106, 1039-1061.

Tversky A. and D. Kahneman (1992),Advances in prospect theory: Cumulative
representation under uncertainty, Journal of Risk and Uncertainty 5, 297-323.
Markets with Asymmetric Information
October 10, 2001
For more than two decades, research on incentives and market equilibrium in sit-
uations with asymmetric information has been a prolic part of economic theory. In
1996, the Bank of Sweden Prize in Economic Sciences in Memory of Alfred Nobel
was awarded to James Mirrlees and William Vickrey for their fundamental contri-
butions to the theory of incentives under asymmetric information, in particular its
applications to the design of optimal income taxation and resource allocation through
di!erent types of auctions. The theory of markets with asymmetric information rests
rmly on the work of three researchers: George Akerlof (University of California,
Berkeley), Michael Spence (Stanford University) and Joseph Stiglitz (Columbia Uni-
versity). Their pioneering contributions have given economists tools for analyzing a
broad spectrum of issues. Applications extend from traditional agricultural markets
to modern nancial markets.
1
Why are interest rates often so high on local lending markets in Third World
countries? Why do people looking for a good used car typically turn to a dealer rather
than a private seller? Why do rms pay dividends even if they are taxed more heavily
than capital gains? Why is it in the interest of insurance companies to o!er a menu of
policies with di!erent mixes of premiums, coverage and deductibles? Why do wealthy
landowners not bear the entire harvest risk in contracts with poor tenants? These
questions exemplify familiar but seemingly di!erent phenomena, each of which
posed a challenge to traditional economic theory. This years laureates showed that
these and many other phenomena can be understood by augmenting the theory
with the same realistic assumption: one side of the market has better information than
the other. The borrower knows more than the lender about his creditworthiness; the
seller knows more than the buyer about the quality of his car; the CEO and board of
a rm know more than the shareholders about the protability of the rm; insurance
clients know more than the insurance company about their accident risk; and tenants
know more than the landowner about harvesting conditions and their own work e!ort.
1
See Riley (2001) for a survey of developments in the economics of information over the last 25
years.
1
Markets with Asymmetric Information 2
More specically, the contributions of the prizewinners may be summarized as fol-
lows. Akerlof showed how informational asymmetries can give rise to adverse selection
in markets. When lenders or car buyers have imperfect information, borrowers with
weak repayment prospects or sellers of low-quality cars may thus crowd out everyone
else from their side of the market, stiing mutually advantageous transactions. Spence
demonstrated that informed economic agents in such markets may have incentives to
take observable and costly actions to credibly signal their private information to un-
informed agents, so as to improve their market outcome. The management of a rm
may thus incur the additional tax cost of dividends, so as to signal high protability.
Stiglitz showed that poorly informed agents can indirectly extract information from
those who are better informed, by o!ering a menu of alternative contracts for a spe-
cic transaction, so-called screening through self-selection. Insurance companies are
thus able to divide their clients into risk classes by o!ering di!erent policies where,
say, lower premiums can be exchanged for higher deductibles. Stiglitz also analyzed
a range of similar mechanisms in other markets.
Akerlof, Spence and Stiglitzs analyses form the core of modern information eco-
nomics. Their work transformed the way economists think about the functioning of
markets. The analytical methods they suggested have been applied to explain many
social and economic institutions, especially di!erent types of contracts. Other re-
searchers have used and extended their original models to analyze organizations and
institutions, as well as macroeconomic issues, such as monetary and employment
policy.
Sections 1 though 3 below give a brief account of the most fundamental contribu-
tions by the laureates. Section 4 describes some applications and empirical tests of
their models. Suggestions for further reading and a list of references are given at the
end.
1. George Akerlof
Akerlofs article, The Market for Lemons: Quality Uncertainty and the Market
Mechanism (Akerlof, 1970), is probably the single most important contribution to
the literature on economics of information. This paper has all the typical features
of a truly seminal piece. It introduces a simple but profound and universal idea,
o!ers numerous interesting implications and points to broad applications. Nowadays,
Akerlofs insights regarding adverse selection are routinely taught in microeconomics
courses at the undergraduate level.
2
His essay analyzes a market for a product where
2
More recently, the term private information or hidden information has become increasingly
common in describing such situations. Those terms say more about the causes of the phenomenon
whereas adverse selection emphasizes its consequences.
sellers are better informed than buyers about the quality of the good; one example
is the market for used cars. Since then, lemons (a colloquialism for defective cars)
has become a well-known metaphor in every economists vocabulary.
Akerlofs idea may be illustrated by a simple example. Assume that a good is
sold in indivisible units and is available in two qualities, low and high, in xed shares
! and 1 !!. Each buyer is potentially interested in purchasing one unit, but cannot
observe the di!erence between the two qualities at the time of the purchase. All
buyers have the same valuation of the two qualities: one unit of low quality is worth
w
L
dollars to the buyer, while one high-quality unit is worth w
H
> w
L
dollars. Each
seller knows the quality of the units he sells, and values low-quality units at v
L
< w
L
dollars and high-quality units at v
H
< w
H
dollars.
If there were separate markets for low and high quality, every price between v
L
and w
L
would induce benecial transactions for both parties in the market for low
quality, as would every price between v
H
and w
H
in the market for high quality. This
would amount to a socially e"cient outcome: all gains from trade would be realized.
But if the markets are not regulated and buyers cannot observe product quality,
unscrupulous sellers of low-quality products would choose to trade on the market for
high quality. In practice, the markets would merge into a single market with one and
the same price for all units. Suppose that this occurs and that the sellers valuation
of high quality exceeds the consumers average valuation. Algebraically, this case is
represented by the inequality v
H
> w, where w = !w
L
+ (1 ! !)w
H
. If trade took
place under such circumstances, the buyers (rational) expectation of quality would
be precisely w. In other words, the market price could not exceed w (assuming that
consumers are risk averse or risk neutral). Sellers with high-quality goods would
thus exit from the market, leaving only an adverse selection of low-quality goods, the
lemons.
3
In his paper, Akerlof not only explains how private information may lead to the
malfunctioning of markets. He also points to the frequency with which such informa-
tional asymmetries occur and their far-reaching consequences. Among his examples
are social segregation in labor markets and di"culties for elderly people in buying
individual medical insurance. Akerlof emphasizes applications to developing coun-
tries. One of his examples of adverse selection is drawn from credit markets in India
3
Classical economic analysis disregarding asymmetric information would misleadingly predict
that goods of both qualities would be sold on the market, at a price close to the consumers average
valuation.
A very early prototype of Akerlofs result is usually referred to as Greshams law: bad money
drives out good. (Thomas Gresham, 1519-1579, was an adviser to Queen Elisabeth I on currency
matters.) But as Akerlof (1970, p. 490) himself points out, the analogy is somewhat lame; in
Greshams law both sellers and buyers can presumably distinguish between good and bad money.
in the 1960s, where local moneylenders charged interest rates that were twice as high
as the rates in large cities. However, a middleman trying to arbitrage between these
markets without knowing the local borrowers creditworthiness, risks attracting those
with poor repayment prospects and becomes liable to heavy losses.
Another fundamental insight is that economic agents attempts to protect them-
selves from the adverse consequences of informational asymmetries may explain ex-
isting institutions. Guarantees made by professional dealers in the used-car market
is but one of many examples. In fact, Akerlof concludes his essay by suggesting
that this (adverse selection) may indeed explain many economic institutions. This
prophecy has come true; his approach has generated an entire literature analyzing
how economic institutions may mitigate the consequences of asymmetric information.
In a later article, The Economics of Caste and the Rat Race and Other Woeful
Tales (Akerlof, 1976), Akerlof enters into a more thorough discussion of the signif-
icance of informational asymmetries in widely di!ering contexts, such as the caste
system, factory working conditions and sharecropping. He uses illustrative exam-
ples to show how certain variables, called indicators, not only provide important
e"ciency-enhancing economic information, but may also cause the economy to be-
come trapped in an undesirable equilibrium. In the case of sharecropping, where
tenancy is repaid by a xed share of the harvest, a tenants volume of production
acts as an indicator of his work e!ort on the farm. On the assembly line in a fac-
tory, the speed of the conveyor belt acts as an indicator of the workers ability, and
can therefore be used as an instrument to distinguish between workers of di!erent
abilities.
Apart from his work on asymmetric information, Akerlof has been innovative
in enriching economic theory with insights from sociology and social anthropology.
Several of his papers on the labor market have examined how emotions such as reci-
procity towards an employer and fairness towards colleagues can contribute to
higher wages and thereby unemployment; see Akerlof (1980, 1982) and Akerlof and
Yellen (1990). This kind of emotionally motivated behavior has recently been con-
rmed experimentally, see e.g., Fehr and Schmidt (1999, 2000), and has also received
empirical support from interview surveys, see e.g., Bewley (1999).
2. Michael Spence
Spences most important work demonstrates how agents in a market can use signaling
to counteract the e!ects of adverse selection. In this context, signaling refers to
observable actions taken by economic agents to convince the opposite party of the
value or quality of their products. Spences main contributions were to develop and
formalize this idea and to demonstrate and analyze its implications.
4
A fundamental
insight is that signaling can succeed only if the signaling cost di!ers su"ciently among
the senders. Subsequent research contains many applications which extend the
theory of signaling and conrm its importance in di!erent markets.
Spences seminal paper Job Market Signaling (Spence, 1973) and book Market
Signaling (Spence, 1974) both deal with education as a signal in the labor market. If
an employer cannot distinguish between high- and low-productivity labor when hiring
new workers, the labor market might collapse into a market where only those with
low productivity are hired at a low wage this is analogous to the adverse-selection
outcome in Akerlofs market where only lemons remain.
Spences analysis of how signaling may provide a way out of this situation can
be illustrated by slightly extending Akerlofs simple example above. Assume rst
that job applicants (the sellers) can acquire education before entering the labor
market. The productivity of low-productivity workers, w
L
, is below that of high-
productivity workers, w
H
and the population shares of the two groups are ! and 1!!,
respectively. Although employers (the buyers) cannot directly observe the workers
productivity, they can observe the workers educational level. Education is measured
on a continuous scale, and the necessary cost in terms of e!ort, expenses or time
to reach each level is lower for high-productivity individuals. To focus on the signaling
aspect, Spence assumes that education does not a!ect a workers productivity, and
that education has no consumption value for the individual. Other things being
equal, the job applicant thus chooses as little education as possible. Despite this,
under some conditions, high-productivity workers will acquire education.
5
Assume next that employers expect all job applicants with at least a certain educa-
tional level s
H
> 0 to have high productivity, but all others to have low productivity.
Can these expectations be self-fullling in equilibrium? Under perfect competition
and constant returns to scale, all applicants with educational level s
H
or higher are
o!ered a wage equal to their expected productivity, w
H
, whereas those with a lower
educational level are o!ered the wage w
L
. Such wage setting is illustrated by the step-
wise schedule in Figure 1. Given this wage schedule, each job applicant will choose
either the lowest possible education s
L
= 0 obtaining the low wage w
L
, or the higher
educational level s
H
and the higher wage w
H
. An education between these levels does
not yield a wage higher than w
L
, but costs more; similarly, an education above s
H
does not yield a wage higher than w
H
, but costs more.
In Figure 1 job applicants preferences are represented by two indi!erence curves,
4
Informal versions of this idea can be traced to the sociological literature; see Berg (1970).
5
Obviously, job applicants incentives to acquire education will be strengthened under the more
realistic assumption that education enhances productivity.
w
H
w
w
L
B
s
^
s
H
s
A C
Figure 1.
Indifference curve for low-productivity job applicants (steep).
Indifference curve for high-productivity job applicants (at).
0
which are drawn to capture the assumption that education is less costly for high-
productivity individuals. The atter curve through point A thus represents those
education-wage combinations (s, w) that high-productivity individuals nd equally
good as their expected education-wage pair (s
H
, w
H
). All points northwest of this
curve as regarded as better than this alternative, while all points to the southeast are
regarded as worse. Likewise, the steeper curve through B indicates education-wage
combinations that low-productivity individuals nd equally good as the minimum
education s
L
= 0 and wage w
L
.
6
With these preferences, high-productivity individuals choose educational level s
H
,
neither more nor less, and receive the higher wage, as alternative B gives them a
worse outcome than alternative A. Conversely, low-productivity individuals optimally
choose the minimum educational level at B, since they are worse o! with alternative A
the higher wage does not compensate for their high cost of education. Employers
expectations that workers with di!erent productivity choose di!erent educational
levels are indeed self-fullling in this signaling equilibrium. Instead of a market failure,
where high-productivity individuals remain outside of the market (e.g., by moving
away or setting up their own business), these workers participate in the labor market
and acquire a costly education solely to distinguish themselves from low-productivity
job applicants.
Absent further conditions, there is a whole continuum of educational levels s
H
with
corresponding signaling equilibria. However, incentive compatibility requires that the
expected level of education not be so high that high-productivity individuals prefer to
refrain from education, or so low that low-productivity applicants prefer to educate
themselves up to that level. Geometrically, these conditions imply that point B lies
below the indi!erence curve of high-productivity individuals through any equilibrium
point corresponding to A, and points like A lie below the indi!erence curve of low-
productivity individuals through point B.
Spence (1973) indicates that a certain signaling equilibrium is the socially most
e"cient. In this equilibrium, high-productivity individuals opt for (and are ex-
pected to do so by the employers) the minimum education to distinguish themselves
from those with low productivity. In other words, high-productivity workers choose
the combination given by point C in Figure 1. Low-productivity workers are then
indi!erent between the education-wage combination ( s, w
H
) at point C and the com-
bination (0, w
L
) at their chosen point B. High-productivity individuals, conversely,
prefer point C to B. Riley (1975) showed that this is the only signaling equilibrium
6
The crucial assumption that more productive applicants nd it su!ciently less costly to acquire
an education the atter indi"erence curve in Figure 1 is closely related to Mirrlees (1971)
so-called single-crossing condition. A similar condition is found in numerous contexts in modern
microeconomic theory and is often referred to as the Mirrlees-Spence condition.
which is robust to wage experimentation by employers. Spences signalling model
also spurred a urry of game-theoretic research. In particular, various renements of
the Nash equilibrium concept have been developed to discriminate between the many
signaling equilibria in Spences model. Many of these renements select the socially
most e"cient signaling equilibrium. An inuential paper in this genre is Cho and
Kreps (1987).
Spence (1973, 1974) also demonstrates the existence of other equilibria, e.g., one
where no applicant acquires education. Assume that employers do not expect educa-
tion to be a productivity signal, i.e., they expect all job applicants, regardless of educa-
tion, to have the average productivity on the market: w = !w
L
+(1!!)w
H
. Employ-
ers then o!er this wage to all job applicants, and their expectations are self-fullling,
as it is optimal for all applicants to choose the minimum level of education s
L
= 0.
Spence also notes the possibility of equilibria where, say, high-productivity men are
expected to acquire another level of education than equally productive women. In
such an equilibrium, the returns to education di!er between men and women, as do
their investments in education.
Apart from his work on signaling, Spence has made distinguished contributions to
the eld of industrial organization. During the period 1975-1985, he was one of the
pioneers in the wave of game-theory inspired work within the so-called new industrial
organization theory. His most important studies in this area deal with monopolistic
competition (1976) and market entry (1977). Spences models of market equilibrium
under monopolistic competition have also been inuential in other elds, such as
growth theory and international trade.
3. Joseph Stiglitz
Stiglitzs classical article with Rothschild on adverse selection, Equilibrium in Com-
petitive Insurance Markets: An Essay on the Economics of Imperfect Information
(Rothschild and Stiglitz, 1976), is a natural complement to the analyses in Akerlof
(1970) and Spence (1973, 1974).
7
Rothschild and Stiglitz ask what uninformed agents
can do to improve their outcome in a market with asymmetric information. More
specically, they consider an insurance market where companies do not have infor-
mation on individual customers risk situation. The (uninformed) companies o!er
their (informed) customers di!erent combinations of premiums and deductibles and,
under certain conditions, customers choose the policy preferred by the companies.
Such screening through self-selection is closely related to Vickrey (1945) and Mir-
rlees (1971) analyses of optimal income taxation, where a tax authority (unaware
7
Salop and Salop (1976) similarly analyze how rms can use self-selection when employing workers
with private information about their propensity to quit.
of private productivities and preferences) gives wage earners incentives to choose the
right amount of work e!ort.
8
Rothschild and Stiglitzs model may be illustrated by means of a simple example.
Assume that all individuals on an insurance market are identical, except for the
probability of injury of a given magnitude. Initially, all individuals have the same
income y. A high-risk individual incurs a loss of income d < y with probability p
H
and a low-risk individual su!ers the same loss of income with the lower probability p
L
,
with 0 < p
L
< p
H
< 1. In analogy with Akerlofs buyer and Spences employer, who
do not know the sellers quality or the job applicants productivity, the insurance
companies cannot observe the individual policyholders risk. From the perspective
of an insurance company, policyholders with a high probability p
H
of injury are of
low quality, while policyholders with a low probability p
L
are of high quality.
In analogy with the previous examples, there is perfect competition in the insurance
market.
9
Insurance companies are risk neutral (cf. the earlier implicit assumption
of constant returns to scale), i.e., they maximize their expected prot. An insurance
contract (a, b) species a premium a and an amount of compensation b in the case of
income loss d. (The deductible is thus the di!erence d !b.)
Rothschild and Stiglitz establish that equilibria may be divided into two main
types: pooling and separating. In a pooling equilibrium, all individuals buy the same
insurance, while in a separating equilibrium they purchase di!erent contracts. Roth-
schild and Stiglitz show that their model has no pooling equilibrium. The reason is
that in such an equilibrium an insurance company could protably cream-skim the
market by instead o!ering a contract that is better for low-risk individuals but worse
for high-risk individuals. Whereas in Akerlofs model the price became too low for
high-quality sellers, here the equilibrium premium would be too high for low-risk in-
dividuals. The only possible equilibrium is a unique separating equilibrium, where
two distinct insurance contracts are sold in the market. One contract (a
H
, b
H
) is
purchased by all high-risk individuals, the other (a
L
, b
L
) by all low-risk individuals.
The rst contract provides full coverage at a relatively high premium: a
H
> a
L
and
b
H
= d, while the second combines the lower premium with only partial coverage:
b
L
< d. Consequently, each customer chooses between one contract without any
deductible, and another contract with a lower premium and a deductible. In equilib-
rium, the deductible barely scares away the high-risk individuals, who are tempted by
the lower premium but choose the higher premium in order to avoid the deductible.
This unique possible separating equilibrium corresponds to the socially most e"cient
8
Stiglitz (1975) actually used the word screening, but addressed what is today known as sig-
naling. Stiglitz refers to Arrow (1973) and Spence (1973), while discussing and extending their
ideas.
9
Stiglitz (1977) provides an analysis of the monopoly case.
signaling equilibrium, point C of Figure 1 in the simple illustration of Spences
model above.
10
Rothschild and Stiglitz also identify conditions under which no (pure
strategy) equilibrium exists.
11
The uniqueness of equilibrium is typical of screening models, as is the correspon-
dence between the screening equilibrium and the socially most e"cient signaling equi-
librium. Rothschild and Stiglitzs article has been very inuential. In particular, their
classication of equilibria has become a paradigm; pooling and separating equilibria
are now standard concepts in microeconomic theory in general and in information
economics in particular.
Stiglitz has made many other contributions regarding markets with asymmetric in-
formation. He is probably the most cited researcher within the information economics
literature perhaps also within a wider domain of microeconomics. In his large pro-
duction, often with coauthors, Stiglitz has time and again pointed out that economic
models may be quite misleading if they disregard informational asymmetries. The
message has been that in the perspective of asymmetric information, many markets
take on a di!erent guise, as do the conclusions regarding the appropriate forms of
public-sector regulation. Several of his essays have become important stepping stones
for further research.
Two papers coauthored by Stiglitz and Weiss (1981, 1983) analyze credit markets
with asymmetric information.
12
Stiglitz and Weiss show that to reduce losses from
bad loans, it may be optimal for banks to ration the volume of loans instead of
raising the lending rate, as would be predicted by classical economic analysis. Since
credit rationing is so common, these insights were important steps towards a more
realistic theory of credit markets. They have had a substantial impact in the elds
of corporate nance, monetary theory and macroeconomics.
Stiglitzs work with Grossman (Grossman and Stiglitz, 1980) analyzes the hypoth-
esis of e"ciency on nancial markets. It introduces the so-called Grossman-Stiglitz
paradox: if a market were informationally e"cient i.e., all relevant information is
10
Rileys (1975) robustness test, with respect to experimenting employers, led to the same equi-
librium in Spences model. In fact, Rileys idea is not wholly unlike that of Rothschild and Stiglitz
(1976). However, Rothschild and Stiglitz made ... a more radical departure from Spences analysis
by proposing that the model should be viewed as a non-cooperative game between the consumers.
(Riley 2001, p. 438).
11
The non-existence problem has spurred some theoretical research. Wilson (1977), for example,
suggests a less stringent denition of equilibrium, based on the idea that unprotable contracts can
be withdrawn. This renders certain otherwise protable deviations unprotable and makes existence
more likely.
12
Stiglitz and Weiss also study moral hazard, a concept already used by Arrow (1963) to refer to
situations where an economic agent cannot observe some relevant action of another agent after a
contract has been signed.
reected in market prices no agent would have an incentive to acquire the infor-
mation on which prices are based. But if everyone is uninformed, then it pays some
agent to become informed. Thus, an informationally e"cient equilibrium does not
exist. This work has exerted considerable inuence in nancial economics.
Stiglitz has proposed an information-based explanation of involuntary unemploy-
ment. In a widely cited article, Shapiro and Stiglitz (1984) develop a labor-market
model with so-called e"ciency wages.
13
By denition, an e"ciency wage exceeds
a workers reservation wage (the wage level which makes him indi!erent between
remaining on the job or quitting) and thus gives workers incentives to perform well
(more e"ciently) to keep their jobs. In Shapiro and Stiglitzs model, an employer is as-
sumed to carry out random surveys among his employees to observe their work e!ort.
A worker caught shirking is red and ends up with his reservation wage (by looking
for another job or setting up his own business), a level lower than if he had refrained
from shirking and instead kept his job at the prevailing wage. Optimal behavior of
both employers and employees results in equilibrium unemployment. Shapiro and
Stiglitzs model is an important ingredient in modern labor and macroeconomics.
Stiglitz is also one of the founders of modern development economics. He has
shown that economic incentives under asymmetric information are not merely aca-
demic abstractions, but highly concrete phenomena with far-reaching explanatory
value in the analysis of institutions and market conditions in developing economies.
One of his rst studies of informational asymmetries (Stiglitz, 1974a) deals with
sharecropping, an ancient but still common form of contracting. As the term implies,
the contract regulates how the harvest should be divided between a landowner and
his tenants. The size of a harvest generally depends on external circumstances such
as weather and on the tenants work e!ort. Under the conventional assumption that
absolute risk aversion is decreasing in wealth, the optimal outcome would be to let
the richer party (here, the landowner) bear the entire risk. In practice, however, the
harvest is divided up between the parties according to xed shares, usually half each.
Stiglitz (1974a) and Akerlof (1976) both attempted to explain this relation, in terms
of asymmetric information between the two parties. Since the landowner usually can-
not observe tenants work e!ort, an optimal contract strikes a balance between risk
sharing and incentives, letting the tenants assume some share of the risk.
In addition to his work on the economics of information, Stiglitz has made sig-
nicant contributions to public economics, especially the theory of optimal taxation
(see e.g., Stiglitz and Dasgupta, 1971), industrial organization (see e.g., Dixit and
Stiglitz, 1977), and the economics of natural resources (see e.g., Stiglitz 1974b and
Dasgupta and Stiglitz, 1980).
13
Concurrent research with similar ideas is reported in Bowles and Boyer (1988).
4. Applications and Evidence
Akerlof, Spence and Stiglitzs analyses of markets and information asymmetries are
fundamental to modern microeconomic theory. This research has furthered our un-
derstanding of phenomena in real markets which could not be fully captured by
traditional neoclassical theory. Moreover, their models have been used to explain the
emergence of many social institutions that counteract the negative e!ects of informa-
tional asymmetries. The range of application is remarkable: from nancial markets,
through industrial organization, all the way to issues in economic development. This
section o!ers a selection of such applications from recent research and a brief discus-
sion of some empirical tests of the models.
In nancial economics e.g., Myers and Majluf (1984) have shown how shareholders
can become victims of adverse selection among rms. In a new sector (such as todays
IT) most rms may appear identical in the eyes of an uninformed investor, while some
insiders may have better information about the future protability of such rms.
Firms with less than average protability will therefore be overvalued by the stock
market where, of course, uninitiated investors also trade. Such rms will therefore
prefer to nance new projects by issuing new shares (as opposed to debt). Firms with
higher than average protability, on the other hand, will be undervalued and nd it
costly to expand by share issue. Under asymmetric information, the low-quality
rms (with low future protability) thus tend to grow more rapidly than high-
quality rms, implying that the market will gradually be dominated by lemons.
When uninitiated investors ultimately discover this, share prices fall (the IT bubble
bursts).
Another puzzle in nancial economics is why some rms choose to distribute
dividends to their shareholders, even if dividends are more heavily taxed (due to
double taxation) than capital gains, as is the case in many countries. A cheaper
alternative would be to retain the prots within the rm and favor shareholders by
way of capital gains through a higher share price. John and Williams (1985) show
that, under asymmetric information, dividends can act as a credible signal for a high-
protability rm on the stock market. Firms with positive insider information pay
dividends to their shareholders, but this signal is too costly for rms with inferior
insider information. The stock market thus interprets costly dividends as a credible
signal for favorable prospects and therefore pays a high price for the stock. Under
certain conditions, the share price rises enough to compensate shareholders for the
extra tax they have to pay on dividends a separating equilibrium is achieved.
In the sphere of industrial organization, numerous applications have shown how
consumers may interpret price setting and advertising as signals for good quality. As
in Spences (1973, 1974) model of a labor market with high- and low-productivity
workers, equilibria can arise when it is protable for rms with high-quality products
to engage in costly advertising, whereas rms which produce low-quality goods refrain
(see e.g., Nelson, 1974 and Milgrom and Roberts, 1986). Tirole (1988) provides
an extensive overview of other applications of information economics in the eld of
industrial organization.
In labor economics, Waldman (1984) examines a situation where rms competing
for labor use the job assignment of a competitors employee as a signal his ability.
Since an employer does not want to signal the true capacity of a good employee
to potential competitors, employees might not necessarily be assigned tasks which
maximize their contribution to the rms prots. Such allocation of labor within
rms might be optimal for an individual rm in a labor-competitive situation, but
results in social ine"ciency,
Bernhardt (1995) develops these arguments into an analysis of promotions, ex-
plaining why low-education employees promoted to high positions are usually ex-
traordinarily capable. An employer who wants to hide his private information about
employees from a competing employer has an incentive not to promote competent
workers. For a promotion to be protable, a low-educated worker therefore has to
be su"ciently capable to compensate for the higher wage the rm is forced to pay
to retain a worker whose competence is revealed to potential competitors. Similar
mechanisms can also explain wage discrimination. Milgrom and Oster (1987) point
out that such discrimination leads to social ine"ciency when workers are assigned to
the wrong jobs or are not given su"cient incentives to become better educated.
Riley (1979) makes an early attempt to empirically test Spences signaling model.
Rileys idea is that signaling should be most important in those sectors of the economy
where worker productivity is di"cult to measure. In such sectors, wages and educa-
tion are thus expected to be strongly correlated at the outset of a workers career,
whereas the correlation should be weaker in sectors where productivity is more easily
observed. Over time, as rms learn more about the productivity of their employees,
the correlation between wages and education should become weaker, particularly in
sectors where productivity is hard to measure. Riley was able to conrm these e!ects
empirically. More recent tests of Spences signaling model were carried out by Lang
and Kropp (1986) and Bedard (2001). Both studies show that data on high-school
enrollments and dropout rates are consistent with a signaling model and inconsistent
with a pure human-capital model.
In their empirical analysis of ring on a labor market with asymmetric information,
Gibbons and Katz (1991) test the relevance of adverse selection and signaling. If rms
can freely decide which employees should be red, other agents on the labor market
will conclude that the ability of red workers is below average (they are lemons).
Workers who are alike in all other (measurable) respects, but who had to leave their
jobs because the rm closed down, should thus nd it easier to get a new job and
receive a higher wage. Based on a large sample of redundant workers, Gibbons and
Katz nd empirical support for these predictions.
Farber and Gibbons (1996) developed Spences signaling model by allowing em-
ployers to obtain information on worker productivity by observing their careers. The
model predicts that the wage e!ect of education is independent of the length of time
a worker has been on the labor market, whereas the wage e!ect of constant, unob-
servable characteristics, which are positively correlated with worker ability, increases
with the time a worker has been employed. Both predictions are consistent with data
regarding young people on the US labor market.
Acemoglu and Pischke (1998) show that asymmetric information about worker
ability can explain on-the-job training in rms. The mechanism resembles that in
Waldman (1984) and Gibbons and Katz (1991). Informational asymmetries concern-
ing a trained workers productivity generate a monopsony (a buyer monopoly) on the
local labor market, implying that the rm can successively pay for the training by
a wage which falls short of the competitive wage. The predictions are empirically
supported when confronted with data from the German apprentice system.
Other attempts to test for the predicted e!ects of asymmetric information have
produced ambiguous results. One di"culty with such tests is to distinguish, in prac-
tice, between adverse selection and moral hazard; another is that screening and sig-
naling partially eliminate the e!ects of informational asymmetries.
14
In recent years, many insights from the economics of information have been in-
corporated into development economics. It is perhaps not so surprising that models
suggested by Akerlof and Stiglitz have had a large inuence in this eld, as their early
studies were largely inspired by issues in development economics. Prime examples
are Akerlofs lemons model and Stiglitzs sharecropping model. Extensions of the
latter e.g., have been used to explain institutional relationships between landowners
and tenants, such as why landowners often grant credit to tenants (it has positive
incentive e!ects on work e!ort). Arguments based on asymmetric information have
also been used to clarify the dichotomy between modern and traditional sectors in
developing economies. Basu (1997) is an example of a modern advanced textbook in
development economics that builds heavily on the economics of information.
14
A direct test carried out by Bond (1982) on data from a market for second-hand small trucks
does not lend support to the asymmetric information hypothesis. Dahlby (1983, 1992) nds some
support for adverse selection using aggregate data on Canadian car insurance. In a study of data
from a car-insurance company, Puetz and Snow (1994) nd support for both adverse selection and
signaling. Chiappori and Salanie (2000) examine whether individuals with a higher risk of having an
accident systematically choose car insurance with better coverage. They are unable to nd statistical
support for such a correlation.
5. Suggested Reading
The laureates own original works remain highly recommended reading: see e.g.,
Akerlof (1970, 1976), Spence (1973, 1974), Rothschild and Stiglitz (1976), Stiglitz and
Weiss (1981) and Shapiro and Stiglitz (1984). Riley (2001) gives a detailed survey
of economic analyses of markets with asymmetric information. Gibbons (1992) o!ers
an accessible introduction to game-theoretic modeling of asymmetric information.
A more advanced introduction to adverse selection, signaling and screening can be
found in Chapter 13 of Mas-Colell, Whinston and Green (1995).
References
[1] Acemoglu D. and S. Pischke (1998): Why do Firms Train? Theory and Evi-
dence, Quarterly Journal of Economics 113, 79-119.
[2] Akerlof G. (1970): The Market for Lemons: Quality Uncertainty and the Mar-
ket Mechanism, Quarterly Journal of Economics 84, 485-500.
[3] Akerlof G. (1976): The Economics of Caste and of the Rat Race and other
Woeful Tales, Quarterly Journal of Economics 90, 599-617.
[4] Akerlof G. (1980): A Theory of Social Customs of which Unemployment May
be One Consequence, Quarterly Journal of Economics 94, 749-775.
[5] Akerlof G. (1982): Labor Contracts as Partial Gift Exchange, Quarterly Jour-
nal of Economics 97, 543-569.
[6] Akerlof G. and J. Yellen (1990): The Fair Wage Hypothesis and Unemploy-
ment, Quarterly Journal of Economics 105, 255-283.
[7] Arrow K. (1963): Uncertainty and the Welfare Economics of Medical Care,
[8] Arrow K. (1972): The Value of and Demand for Information, in McGuire, C.
and R. Radner (eds.), Decision and Organization, North-Holland (London).
[9] Arrow K. (1973): Higher Education as a Filter, Journal of Public Economics
2, 193-216.
[10] Banerjee A. and A. Newman (1998): Information, the Dual Economy, and
Development, Review of Economic Studies 65, 631-653.
[11] Basu K. (1997): Analytical Development Economics, MIT Press (Cambridge,
MA).
[12] Bedard K. (2001): Human Capital versus Signaling Models: University Access
and High School Dropouts, Journal of Political Economy 109, 749-775.
[13] Berg I. (1970): Education and Jobs: The Great Training Robbery, Praeger (New
York, NY).
[14] Bernhardt D. (1995): Strategic Promotion and Compensation, Review of Eco-
nomic Studies 62, 315-339.
[15] Bewley T. (1999): Why Wages Dont Fall During a Recession, Harvard Univer-
sity Press (Cambridge, MA).
[16] Bond E. (1982): A Direct Test of the Lemons Model: The Market for Used
Pickup Trucks, American Economic Review 72, 836-840.
[17] Bowles S. and R. Boyer (1988): Labor Discipline and Aggregate Demand,
[18] Braverman A. and J. Stiglitz (1982): Sharecropping and the Interlinking of
Labor Markets, American Economic Review 72, 695-715.
[19] Cho I.-K. and D. Kreps (1987): Signaling Games and Stable Equilibria, Quar-
terly Journal of Economics 102, 179-221.
[20] Chiappori P. and B. Salanie (2000): Testing for Asymmetric Information in
Insurance Markets, Journal of Political Economy 107, 56-78.
[21] Dahlby B. (1983): Adverse Selection and Statistical Discrimination: An Analy-
sis of Canadian Automobile Insurance, Journal of Public Economics 20,121-130.
[22] Dahlby B. (1992): Testing for Asymmetric Information in Canadian Automobile
Insurance, in G. Dionne (ed.) Contributions to Insurance Economics, Kluwer
(Boston).
[23] Dasgupta P. and J. Stiglitz (1980): Uncertainty, Industrial Structure, and the
Speed of R&D, Bell Journal of Economics 11, 1-28.
[24] Dixit A. and J. Stiglitz (1977): Monopolistic Competition and Optimal Product
Diversity, American Economic Review 67, 297-308.
[25] Farber H. and R. Gibbons (1996): Learning and Wage Dynamics, Quarterly
[26] Fehr E. och K. Schmidt (1999): A Theory of Fairness, Competition and Co-
operation, Quarterly Journal of Economics 114, 817-868.
[27] Fehr E. and K. Schmidt (2000): Theories of Fairness and Reciprocity: Evidence
and Economic Applications, forthcoming in M. Dewatripont et al. (eds.), Ad-
vances in Economic Theory, Eigth World Congress of the Econometric Society,
Cambridge University Press (Cambridge, UK).
[28] Gibbons R. (1992): A Primer in Game Theory, Harverster-Wheatsheat, (New
York, NY).
[29] Gibbons R. and L. Katz (1991): Layo!s and Lemons, Journal of Labor Eco-
nomics 9, 351-380.
[30] Grossman S. and J. Stiglitz (1980): On the Impossibility of Informationally
E"cient Markets, American Economic Review 70, 393-408.
[31] John K. and J. Williams (1985): Dividends, Dilution and Taxes: A Signalling
Equilibrium, Journal of Finance 40, 1053-69.
[32] Lang K. and D. Kropp (1986), Human Capital versus Sorting: The E!ects of
Compulsory Attendance Laws, Quarterly Journal of Economics 101, 609-624.
[33] Marschak J. (1959): Remarks on the Economics of Information, Contribu-
tions to Scientic Research in Management, Western Data Processing Center,
University of California, Los Angeles, 79-98.
[34] Mas-Colell A., M. Whinston and J. Green (1995): Microeconomic Theory, Ox-
ford University Press (Oxford).
[35] Milgrom P. and J. Roberts (1986): Price and Advertising Signals of Product
Quality, Journal of Political Economy 94, 795-81.
[36] Milgrom P. and S. Oster (1987): Job Discrimination, Market Forces, and the
Invisibility Hypothesis, Quarterly Journal of Economics 102, 453-476.
[37] Mirrlees J. (1971): An Exploration in the Theory of Optimum Income Taxa-
tion, Review of Economic Studies 38, 175-208.
[38] Myers S. and N. Majluf (1984): Corporate Financing and Investment Decisions
when Firms Have Information that Investors Do Not Have, Journal of Financial
[39] Nelson P. (1974): Advertising as Information, Journal of Political Economy
82, 729-54.
[40] Puetz R. and A. Snow (1994): Evidence on Adverse Selection: Equilibrium
Signalling and Cross-Subsidization in the Insurance Market, Journal of Political
Economy 102, 236-257.
[41] Riley J. (1975): Competitive Signalling, Journal of Economic Theory 10, 174-
186.
[42] Riley J. (1979): Testing the Educational Screening Hypothesis, Journal of
[43] Riley J. (2001): Silver Signals: Twenty-Five Years of Screening and Signaling,
Journal of Economic Literature 39, 432-478.
[44] Rothschild M. and J. Stiglitz (1976): Equilibrium in Competitive Insurance
Markets: An Essay on the Economics of Imperfect Information, Quarterly Jour-
[45] Shapiro C. and J. Stiglitz (1984): Equilibrium Unemployment as a Worker
Discipline Device, American Economic Review 74, 433-444.
[46] Spence M. (1973): Job Market Signalling, Quarterly Journal of Economics 87,
355-374.
[47] Spence M. (1974): Market Signaling, Harvard University Press (Cambridge,
MA).
[48] Spence M. (1976): Product Selection, Fixed Costs and Monopolistic Competi-
tion, Review of Economic Studies 43, 217-235.
[49] Spence M. (1977): Entry, Capacity, Investment and Oligopolistic Pricing, Bell
[50] Stiglitz J. and P. Dasgupta (1971): Di!erential Taxation, Public Goods and
Economic E"ciency, Review of Economic Studies 38, 151-174.
[51] Stiglitz J. (1974a): Incentives and Risk Sharing in Sharecropping, Review of
[52] Stiglitz J. (1974b): Growth with Exhaustible Natural Resources: E"cient and
Optimal Growth Paths, Review of Economic Studies 41, 123-38.
[53] Stiglitz J. (1975): The Theory of Screening, Education, and the Distribution
of Income, American Economic Review 65, 283-300.
[54] Stiglitz J. (1977): Monopoly, Non-Linear Pricing, and Imperfect Information:
The Insurance Market, Review of Economic Studies 44, 407-430.
[55] Stiglitz J. and A. Weiss (1981): Credit Rationing in Markets with Imperfect
Information, American Economic Review 71, 393-410.
[56] Stiglitz J. and A. Weiss (1983): Incentive E!ects of Terminations: Applications
to the Credit and Labor Markets, American Economic Review 73, 912-927.
[57] Tirole J. (1988): The Theory of Industrial Organization, MIT. Press (Cambridge,
MA).
[58] Vickrey W. (1945): Measuring Marginal Utility by Reactions to Risk, Econo-
metrica 13, 319-333.
[59] Waldman M. (1984): Job Assignments, Signalling, and E"ciency, Rand Jour-
[60] Wilson C. (1977): A Model of Insurance Markets with Incomplete Information,

Phone: +46 8 673 95 00, Fax: +46 8 15 56 70, E-mail: info@kva.se, Web site: www.kva.se
__________________________________________________________________________

Advanced information on
the Bank of Sweden Prize in Economic Sciences
in Memory of Alfred Nobel, 2000

The Scientific Contributions of
James Heckman and Daniel McFadden
Microeconometrics and microdata
Microeconometric research is concerned with empirical analysis of the economic
behavior of individuals and households, such as decisions on labor supply, con-
sumption, migration or occupational choice. Microeconometric methods are
equally relevant in studies of individual rms, for example their production and
employment decisions. Over the last several decades, signicant breakthroughs
in empirical microeconomic research have been triggered by innovations in mi-
croeconometric methods and by greater availability of new types of data. The
raw material in microeconometric research is microdata, where the units of ob-
servation are individuals, households or rms. Microdata appear as cross-section
data and, to an increasing degree, as longitudinal (panel) data.
While oering new means of testing economic hypotheses and estimating
economic models, the analysis of microdata has also raised new econometric
problems. This, in turn, has inspired methodological research in microecono-
metrics, which can be loosely dened as a collection of econometric methods for
handling problems of model specication, estimating and testing that arise in
the analysis of microdata. A hallmark of recent microeconometric research is the
close interplay of applied work on substantive economic issues and theoretical
work on methodological problems.
New data bases have been crucial in this development. Until the late 1960s,
the availability of data sources for empirical studies of individual economic be-
havior was very limited. Nowadays, there are a number of longitudinal data
sets covering individuals and households, in the United States as well as in most
European countries. A pathnding early eort to create an infrastructure for
microeconometric research was the Panel Study of Income Dynamics (PSID)
set up by James Morgan and others at the University of Michigan in the late
1960s. The PSID has been used intensively in applied research and has served
as a model for the construction of longitudinal data sets in other countries.
It is only recently that this remarkable growth of microdata on individuals
and households has been matched by similar information on individual rms. As
a result, microeconometric applications have largely been dominated by studies
of individual and household behavior.
Microeconometric applications cover a wide range of elds in economics. La-
bor economists have used microeconometric techniques to study labor supply
decisions, individual earnings, educational choice, worker mobility, and the du-
ration of spells of employment and unemployment. Microeconometric methods
are essential for studies in empirical public nance, e.g., the eects of taxes
and welfare policies on labor supply; in consumer research, e.g., the choice of
dierent brands; and in urban and transportation economics, e.g., the choice of
residence or mode of transportation. Applied research in microeconomics and
industrial organization relies on microeconometrics in studies of rms produc-
tion and factor demand decisions. Similar methods are also used by researchers
in other social sciences.
James Heckman and Daniel McFadden have made fundamental contributions
to microeconometrics. Heckmans most inuential work deals with problems
1
that arise when data are generated by a non-random selection process, a common
phenomenon in microeconometric studies. McFaddens foremost contributions
concern theory and methods for discrete choice analysis, such as the choice of
occupation or mode of transportation.
James J. Heckman
James Heckman was born in Chicago, IL in 1944. After undergraduate studies
at Colorado College, majoring in mathematics, he went on to study economics
at Princeton University where he received his Ph.D. in 1971. Heckman has
taught at Columbia University, Yale University and the University of Chicago.
Since 1995 he is Henry Schultz Distinguished Service Professor of Economics at
the University of Chicago.
Heckmans numerous contributions to microeconometric theory have been
developed in conjunction with applied empirical research, especially in labor eco-
nomics. His applied research covers labor supply, labor earnings, unemployment
duration, evaluation of labor-market programs, fertility, and discrimination.
Heckmans analysis of selection bias in microeconometric research has pro-
foundly changed applied research, in economics as well as in other social sciences.
Selection bias may arise when a sample under study does not randomly repre-
sent the underlying population. The problem facing the analyst is to obtain
estimates of relevant population parameters even in the wake of a selective sam-
ple. Non-random sample selection may result from individual decisions by the
agents under study (self-selection), but may also reect administrative rules, or
decisions on the part of sampling statisticians.
Selection bias and self-selection
Selection problems are pervasive in applied microeconometric research. Working
hours and wages are observed only for those individuals who have chosen to
work; earnings of migrants are observed only for those who have chosen to move;
earnings of university graduates are observed only for those who have completed
a university education, and so on. The selection problem can be viewed as a
problem of missing observations. Wages and hours cannot be observed among
non-working individuals, had they chosen to work; likewise, the earnings of
non-migrants, had they chosen to migrate, are unobservable to the analyst;
analogously, there is a lack of information on the earnings of workers with a
high-school education, had they pursued a university education.
Heckmans approach to the selection problem is closely linked to economic
theory. His key insight is that observations are often missing because of con-
scious (self-selection) choices made by economic agents (e.g., the decision to
work, to migrate or to pursue higher education). The relation between the rea-
sons for missing observations and the nature of non-missing observations thus
takes on an intriguing theoretical structure. Heckmans proposed solutions to
2
selection problems can be appreciated not only statistically, but also in terms
of microeconomic theory.
Heckmans contributions to the econometrics of selective samples emerged
concurrently with his studies of labor supply in the mid-1970s. These studies
pioneered the second generation models of labor supply, which are distin-
guished by estimating equations derived explicitly from utility maximization
with stochastic error terms as an integral part of the model, rather than added
as an afterthought. They enabled a unied analysis of the factors determining
work hours and labor-force participation.
An important early example of this strand of research, the contribution in
Heckman (1974) is remarkable for its treatment of the selectivity problem inher-
ent in all studies of labor supply.
1
Standard economic theory views labor-force
participation as a result of utility maximization, where the participants are indi-
viduals whose market wages exceed their reservation wages. To obtain unbiased
estimates of basic structural parameters, the estimation procedure has to recog-
nize the sample of labor-force participants is not the result of random selection,
but the result of individual self-selection implied by utility maximization.
Heckman (1974) presented a model of married womens labor supply based
on the utility maximization hypothesis. The sample of working women is self-
selected in the sense that hours of work are only observed for women with market
wages higher than their reservation wages. Heckman derived a likelihood func-
tion for this problem, estimated equations for market wages, the probability for
working, and hours of work, and then used the estimated structural parameters
to predict the probability of working, hours of work, reservation wages and mar-
ket wages. This paper is an excellent example of how microeconomic theory can
be combined with microeconometric methods to clarify an important economic
issue.
Heckmans subsequent work oered computationally simpler methods for
handling selection bias (Heckman 1976, 1979). The well-known Heckman cor-
rection also called the two-stage method, Heckmans lambda or the Heckit
method
2
has become part of the standard toolbox in applied microecono-
metric work. The method may be described by means of the following two
equations:
w
i
= x
1i
1
+"
1i
, (1)
e
i
= x
2i
2
+"
2i
: (2)
Equation (1) determines the individuals market wage, whereas (2) is a partici-
pation equation describing the individuals propensity to work. Thus, w
i
is the
observed market wage for individual i if she works and e
i
a latent variable that
1
See also Gronau (1974) and Lewis (1974) for important early discussions of self-selection
in the context of data on wages and labor supply.
2
The label Heckit was presumably invented to acknowledge similarities with the famous
Tobit estimator due to 1981 economics laureate James Tobin (1958).
3
captures the propensity to work; x
1i
and x
2i
are vectors of observed explana-
tory variables, such as age and education; "
1i
and "
2i
, nally, are mean-zero
stochastic errors representing the inuence of unobserved variables aecting w
i
and e
i
. The parameters (vectors) of interest are
1
and
2
.
Although the latent variable e
i
is unobserved, we can dene a dummy vari-
able e
i
= 1 if e
i
0 and e
i
= 0 otherwise; we thus observe the market wage
only if e
i
= 1, i.e., if the individual works. It is likely that the unobserved terms
"
1i
and "
2i
are positively correlated; individuals with higher wages, given x
1i
and x
2i
, are presumably also more likely to work. If so, the sample of individu-
als observed as working will not accurately represent the underlying population,
even in a large sample. Failure to recognize this selectivity generally produces
inconsistent estimates of the parameters in the wage equation.
Heckman suggested a simple method to deal with this selection problem.
Note that the conditional mean of "
1i
can be written as:
E("
1i
j e
i
0) = E("
1i
j "
2i
x
2i
2
) , (3)
and hence
E(w
i
j x
1i
; e
i
= 1) = x
1i
1
+E("
1i
j "
2i
x
2i
2
) . (4)
Thus, the regression equation on the selected sample depends on both x
1i
and
x
2i
. Omitting the conditional mean of "
1i
biases the estimates of
1
(unless
"
1i
and "
2i
are uncorrelated, in which case the conditional mean of "
1i
is zero).
Selection bias can thus be regarded as a standard problem of omitted-variable
bias. The problem is to nd an empirical representation of the conditional mean
of "
1i
and include this variable in the wage equation.
Under the assumption that "
1i
and "
2i
are drawn from a bivariate normal
distribution, we can derive the regression equation:
E(w
i
j x
1i
; e
i
= 1) = x
1i
1
+
1
i
. (5)
In (5) is the correlation coecient between "
1i
and "
2i
,
1
is the standard
deviation of "
1i
, and
i
the inverse of Mills ratio is given by
i
=
(x
2i
2
=
2
)
(x
2i
2
=
2
)
, (6)
where and are the density and distribution functions of the standard normal
distribution and
2
is the standard deviation of "
2i
.
Heckman showed how to estimate (5) in a two-step procedure. The rst step
involves estimating the parameters in (2) by the probit method, using the entire
sample. These estimates can then be used to compute
i
for each individual
in the sample. Once
i
is computed, we can estimate (5) over the sample of
working individuals by ordinary least squares regression, treating
1
as the
regression coecient for
i
.
The sign of the selection bias depends on the correlation between the errors
in the wage and participation equations () and the correlation between
i
and
4
the variables in the wage equation (x
1i
). Since
i
is a decreasing function of the
probability of sample selection, it follows that the -coecient on variables in
x
1i
that are likely to raise both wages and participation, such as education, will
be biased downwards if the Heckit technique is not applied (provided > 0).
Heckmans seminal work in this area has generated many empirical applica-
tions in economics as well as in other social sciences and applied statistics dealing
with non-randomly missing data. One early example of empirical applications is
the paper by Lee (1978), who examined relative wage eects of union member-
ship in the United States, recognizing that membership is not random but the
outcome of individual self-selection. Another is Willis and Rosen (1979), who
investigated the wage premiums associated with higher education, recognizing
endogenous educational choice as depending on the perceived gains to educa-
tion. Later on, Heckman and Guilherme Sedlacek (1985) presented an empirical
model of the sectoral allocation of workers in the U.S. labor market along the
lines of the Roy (1951) model of income distribution: utility-maximizing individ-
uals can work in several sectors, but only in one sector at a time. The analysis
includes an assessment of how self-selection impacts on wage inequality.
Heckmans work has also generated a sizable literature on econometric method.
The original model has been extended in a number of ways by Hackman and
others.
3
These eorts, typically aimed at eliminating the restrictive assump-
tion of bivariate normality, have involved the use of semi-parametric methods;
see, for example, Heckman and Robb (1985b), Heckman (1990), Manski (1989),
Newey, Powell and Walker (1990) and Lee (1994a,b).
Evaluation of active labor-market programs
Active labor-market programs, such as training, job-search assistance and em-
ployment subsidies, have become increasingly widespread in most OECD coun-
tries. Such programs are generally targeted at individuals who are unemployed
or have low skills or earnings.
The classical problem of a program evaluation is to determine how partic-
ipation in a specic program aects individual outcomes, such as earnings or
employment, compared to non-participation. The paramount diculty is to
characterize the counterfactual situation, i.e., to answer the question: what
would have happened if the individual had not participated in the program?
Since it is impossible to observe an individual as both participant and non-
participant, it is necessary to use information on non-participants outcomes for
this purpose. Given that the allocation of individuals to programs is seldom
purely random, the group of participants becomes a selected sample with ob-
served and unobserved characteristics that may dier from those of the overall
population.
James Heckman is the worlds foremost researcher on econometric policy
evaluation. In this area, as in his labor-supply research, Heckman relies on a
structural approach based on microeconomic theory to guide the model speci-
cation and to interpret the empirical results. The main ingredients of policy
3
See Vella (1998) for a survey of econometric literature following Heckmans seminal work.
5
evaluation are twofold: (i) a model of participation in programs, and (ii) a model
of program outcomes. Heckmans research on program evaluation can be seen as
a natural continuation of his earlier work on selection models. In joint work with
others, Heckman has presented a number of new results concerning identication
and estimation of the eects of social programs (Heckman and Robb, 1985 a,b)
and, more recently, improved our understanding of the pros and cons of using
experimental rather than non-experimental data in program evaluation (Heck-
man and Smith, 1995; Heckman, Ichimura and Todd, 1997; Heckman, Smith
and Clements, 1997). The latter set of studies has also spurred theoretical ad-
vances in the use of matching methods as econometric evaluation estimators.
On balance, Heckman does not emerge as a strong supporter of the experimental
approach, arguing that social experiments are valid only under special statisti-
cal and behavioral assumptions. A main lesson seems to be that there are no
universally correct methods for evaluating programs. What method works best
depends on the issue in question and on the economic models that determine
participation and outcomes. Heckman has also presented substantive empirical
results on the eects of various labor-market programs. Conclusions regarding
the eects are often somewhat pessimistic. A survey primarily of U.S. studies
(Heckman, LaLonde and Smith, 1999) concludes that programs frequently have
very small (sometimes negative) eects for the participants and do not appear
to pass conventional cost-benet tests. On the other hand, there does seem to
be substantial heterogeneity in program outcomes across participants and types
of programs.
Duration models
The analysis of duration data has a long tradition in engineering and biomedical
research.
4
More recently, it has also entered into economic and social science
research, where duration models have been applied to a variety of problems.
Duration models are now standard tools when studying the length of unem-
ployment spells, demographic events (marriage, fertility, mortality, and migra-
tion), political events (e.g., how the occurrence of government crises depends
on the time elapsed since the last election), and some industrial relations (e.g.,
the length of strikes). The models are also used in consumer research, to study
the timing of purchases of products, as well as in macroeconomic research, to
examine issues such as the duration of business cycles.
In his work on econometric duration analysis, Heckman has been particularly
preoccupied with the eects of unobserved heterogeneity, i.e., individual dier-
ences in unobserved variables that may inuence the duration of unemployment
or employment. As unobserved heterogeneity in the context of duration data
introduces specic selection problems, Heckmans work in this area ts well with
his overall research agenda on sample selection.
This may be exemplied by studies of how the exit rate from unemployment
to employment evolves over a spell of unemployment. One problem here is that
4
So-called failure time analysis (e.g., analysis of the durability of electrical equipment) has
been common in engineering for decades. Survival analysis likewise has a long tradition in
biomedical research (e.g., studies of survival after surgery).
6
individuals with relatively weak employment prospects seem to survive as
unemployed to a higher degree than individuals with more favorable character-
istics. Thus, the quality of the stock of unemployed at each point in time
is the result of a selection process partly driven by factors unobserved by the
analyst. Overrepresentation of more unemployment-prone individuals at long
durations can easily lead to the conclusion of negative duration dependence,
i.e., that the exit rate to employment declines over a spell of unemployment.
However, what appears to be negative duration dependence may simply be a
sorting or selection eect.
In joint work with Burton Singer, Heckman addressed the problem of treat-
ing unobserved heterogeneity without imposing restrictive assumptions regard-
ing the distribution of unobserved variables. Heckman and Singer (1984a) pro-
posed a non-parametric estimator that has become widely used in applied work
in economics and demography.
Other noteworthy contributions by Heckman to the duration literature in-
clude identication results for a class of duration models (Heckman and Singer,
1984b and Heckman and Honor, 1989) and further treatment of identication
issues (Heckman, 1991 and Heckman and Taber, 1994). Heckman has also writ-
ten applied empirical papers on unemployment duration and fertility.
Daniel L. McFadden
Daniel McFadden was born in Raleigh, NC in 1937. He received his under-
graduate degree from the University of Minnesota, with a major in physics.
McFadden switched to economics in the late 1950s and received a Ph.D. from
the University of Minnesota in 1962. His academic appointments include profes-
sorships at the University of Pittsburg, Yale University, Massachusetts Institute
of Technology and the University of California at Berkeley. Since 1990 he is E.
Morris Cox Professor of Economics at Berkeley.
McFaddens most important and inuential contribution is his development
of the economic theory and econometric methodology for discrete choice, i.e.,
choice among a nite set of alternatives. He has also made signicant contribu-
tions in other elds of economics, including production theory and environmen-
tal economics. McFaddens research may best be characterized by his ability
to combine the development of theory and methodology with applications to
substantive empirical problems.
Discrete choice analysis
Discrete choice problems appear frequently in economics as well as in other social
sciences. Consider, for example, the modeling of phenomena such as individual
labor-force participation, occupational or locational decisions, or travel mode
choice. Here the observations to be explained are discrete (or qualitative) and
cannot be represented by continuous variables. Standard demand theory as well
7
as traditional econometric methods, aimed at explaining variations in continuous
variables, are generally inappropriate to analyze discrete choice behavior.
Problems of qualitative choice were initially dealt with in work in psycho-
metrics, biometrics and to some extent also econometrics. Early contributions
of particular importance are Thurstone (1927) and Luce (1959), who formu-
lated models of discrete probabilistic choice. According to the psychological
interpretation, individual choice behavior is intrinsically probabilistic.
By contrast, the economic and econometric approach developed by McFad-
den treats individual choice as deterministic. It focuses instead on the lack of
information on the part of the analyst, such as imperfect information about the
characteristics of alternatives and individuals under study. Whereas psycholo-
gists have generally been concerned with individual choices per se, economists
have generally been more interested in aggregate outcomes, such as the fraction
of a population that selects a certain alternative.
5
The conditional logit model
McFaddens most fundamental contribution is the integration of economic the-
ory and econometric methodology for discrete choice analysis. His seminal
paper, entitled Conditional Logit Analysis of Qualitative Choice Behavior
(McFadden, 1974a), and contemporaneous empirical case studies fundamen-
tally changed researchers thinking about the econometric analysis of individual
behavior. Discrete choice analysis rapidly developed into one of the main elds
of modern econometrics.
McFaddens approach may be sketched as follows. Suppose that each indi-
vidual in a population faces a nite set of alternatives and chooses an alternative
that maximizes his or her utility. The data available to the analyst are assumed
to be generated by the repeated drawing of an individual at random from the
population and recording a vector a of the individuals attributes (such as age,
gender, income, etc.), the set I of alternatives available to the individual (e.g.,
travelling by car, bus, subway, etc.), and the individuals actual choice i from
the set I. Assume that the individuals utility from choosing i is of the addi-
tive form u(i; a) = v(i; a) +e(i; a; !), where v(i; a) is common to all individuals
with observed attributes a, while e(i; a; !) is particular to the drawn individ-
ual !. Both utility terms are deterministic, the rst reecting representative
tastes in the population, and the second reecting idiosyncratic taste variations.
Treating the unobserved utility terms e(i; a; !) as realizations of random vari-
ables "(i; a), and letting P(i j a; I) denote the conditional choice probability that
the randomly drawn individual will choose alternative i 2 I, given his or her
observed attributes a and the set of alternatives I; we obtain:
P(ija; I) = Pr[v(i; a) +"(i; a) v(j; a) +"(j; a) 8j 2 I] : (7)
This is called the additive random utility model (ARUM) of discrete choice. The
right-hand side of (7) is the probability that an individual drawn at random from
5
See Ben-Akiva and Lerman (1985) and Anderson, de Palma and Thisse (1992) for surveys
and discussions of discrete choice models.
8
the population has a utility function that makes i the utility-maximizing choice,
given the individuals attributes a and choice set I.
If the random vector h"(i; a)i
i2I
has a joint cumulative distribution function
F, and we write I = f1; :::; Ig, the right-hand side of equation (7) can be written
as an integral in the ith partial derivative, F
i
, of F:
P(ija; I) =
Z
+1
1
F
i
[x +v(i; a) v(1; a); :::; x +v(i; a) v(I; a)]dx : (8)
In particular, if the random variables "(i; a) are independently distributed with
cumulative distribution function exp[e
xi
] for > 0 (the Gumbel, or type-I,
extreme-value distribution), their joint distribution F becomes:
F(x
1
; :::; x
I
) = exp
2
4
X
j2I
e
x
j
3
5
; (9)
and the choice probabilities in equation (7) reduce to the analytically convenient
logit form:
P(i j a; I) =
expv(i; a)
P
j2I
expv(j; a)
: (10)
The parameter > 0 is inversely proportional to the standard deviation of the
random utility terms "(i; a). In the limit as ! 1, the choice probabilities
P(i j a; I) in (10) assign all probability mass to the alternatives with maximum
representative utility v(i; a) and we obtain the traditional microeconomic
model of fully deterministic utility maximization.
In order to make the resulting logit model tractable for predictive purposes,
it is usually assumed that the representative utility terms v(i; a) depend on
known characteristics of the alternatives and the population in some analytically
tractable way. For example, in the case of travel mode choice, these character-
istics could be travel time, travel costs, etc. The associated parameter vectors
can then be estimated by the maximum likelihood method.
McFadden called his innovation the conditional logit model. Although multi-
nomial logit models had been around for some time (Theil, 1969; Quandt, 1970),
McFaddens derivation of the model based on an economic theory of population
choice behavior was entirely new. His contribution was immediately recognized
as a paradigmatic breakthrough, and paved the way for statistical estimation
and applications.
Subsequent development of discrete choice analysis
The attractiveness of the multinomial logit model lies in its combination of
solid microeconomic foundations and computational simplicity. This simplicity
follows from the assumption of statistical independence of the random util-
ity terms, an assumption which implies independence of irrelevant alternatives
(IIA). The ratio of the probabilities of choosing any two alternatives is indepen-
dent of the properties of all other alternatives, as can be noted by using (10)
9
and taking the ratio of any two alternatives. For example, an expanded choice
set does not aect the odds ratio for two choices. Under the IIA assumption,
parameters pertaining to two alternatives can be consistently estimated by using
data only on those individuals who have chosen these two alternatives. Thus,
the IIA property allows estimation of the multinomial logit model on choice-
based samples, which are much more easily obtained than population samples,
with due compensation for the arising sample bias; see Manski and Lerman
(1977) and Manski and McFadden (1981).
However, as McFadden has pointed out, the IIA assumptions is restrictive
in many applications. For example, it is unlikely that the odds ratio of any two
choices would be invariant to the introduction of a new alternative that is a
close substitute for an existing alternative.
6
This problem exemplies the more
general diculty of estimating population parameters from self-selected samples.
Hausman and McFadden (1984) devised a procedure for testing the validity of
the IIA assumption based on the idea of comparing estimates from a self-selected
subset with estimates from the full choice set; the two estimates should not
change systematically if the IIA assumption is valid. Further specication tests
were developed in McFadden (1987).
McFadden has also showed how to relax the IIA assumption through his
development of nested multinomial logit and generalized extreme value (GEV)
models. The nested logit model, introduced by Ben-Akiva (1973) and McFadden
(1978), relaxes the IIA assumption by permitting certain statistical dependence
between the choices. In this model individuals decisions can be interpreted as
having a hierarchical structure. Consider, for instance, the joint choice of desti-
nation and travel mode. One possible nested logit formulation of this decision
problem would be to assume that for each destination the individual selects
the preferred mode of transportation and, taking this into account, chooses his
destination.
The GEV model, developed by McFadden (1978, 1981), is more general and
analytically elegant. To derive it, note that if we generalize equation (9) to
F(x
1
; :::; x
I
) = exp
G(e
x1
; :::; e
xI
)
; (11)
for some linearly homogeneous function G, equation (7) becomes
P(i j a; I) =
G
i
(e
v(1;a)
; :::; e
v(I;a)
) expv(i; a)
G(e
v(1;a)
; :::; e
v(I;a)
)
; (12)
where G
i
is the partial derivative of G with respect to its ith argument. Using
Eulers formula, the denominator can be written as a sum over Gs partial
derivatives, and we obtain:
P(i j a; I) =
expv(i; a) + lnG
i
(e
v(1;a)
; :::; e
v(I;a)
)
P
j
exp[v(j; a) + lnG
j
(e
v(1;a)
; :::; e
v(I;a)
)]
: (13)
6
The 1983 economics laureate Gerard Debreu (1960) was the rst to point this out, in the
context of Luces (1959) probabilistic choice model.
10
In other words, the choice probabilities still have a logit form. Now, however,
the choice probability for an alternative i depends not only on its own observed
attributes, via v(i; a), but also on the observed attributes of other alternatives,
such that IIA need no longer hold. The usual logit model is obtained as a special
case when the function G is the sum of its arguments (then all partial derivatives
are equal to unity, and the second term in the numerator and denominator of
(13) vanishes), and the nested logit model is obtained as a special case when G
is a CES function.
Another useful generalization of the multinomial logit model is the so-called
mixed logit model. This generalization is obtained by aggregating choice behav-
iors across diverse subpopulations, where all individuals in each subpopulation
have the same observed attributes, and each subpopulations choice behavior
is modelled as outlined above. As shown by McFadden and Train (1998), any
well-behaved random utility model of discrete choice can be approximated to
any degree of accuracy by such a mixed logit model. Applications of this model
usually require Monte Carlo simulation methods.
Yet another approach is the multinomial probit model with correlated errors.
However, as with the mixed logit model, computational diculties arise when
this model is applied to problems with more than a few alternatives, mainly
because the calculation of choice probabilities involves evaluating multiple inte-
grals. Lerman and Manski (1981) introduced the idea of computing the choice
probabilities by means of Monte Carlo simulation methods where repeated ran-
dom draws are taken from a multivariate normal distribution. McFadden (1989)
further developed this idea by proposing an estimation approach known as the
method of simulated moments. McFaddens article resolved the basic statistical
properties of this method and a sizeable literature has since emerged in this
area.
A natural step in the evolution of individual choice analysis is the develop-
ment of models and methods that explain both discrete and continuous choices.
The article by Dubin and McFadden (1984) is noteworthy as a ne example of
how to integrate a general methodological contribution with an empirical study
of practical usefulness (the households discrete choice among electric appliances
and its continuous choice of energy consumption).
As already emphasized, most of McFaddens work is indeed characterized by
a close relation between economic theory, econometric methodology and applied
empirical studies. Early empirical applications to urban travel demand are thus
reported in McFadden (1974b) and Domencich and McFadden (1975). During
the 1980s, and 1990s, McFadden was engaged in empirical work on residen-
tial energy demand (Cowing and McFadden, 1984), the demand for telephone
services (McFadden, Train and Ben-Akiva, 1987) and the demand for housing
among the elderly (McFadden, 1994a).
Other contributions
McFadden has made important contributions in several other elds. In the 1960s
and early 1970s, he worked intensively on the theoretical and econometric anal-
ysis of production. Most of these contributions remained unpublished until they
11
appeared in the two-volume collection of papers edited by Fuss and McFadden
(1978). McFaddens work in this area became highly inuential and established
the principle of duality between cost or prot functions and production functions
as a principal tool in the empirical analysis of production.
In other important work, in collaboration with Peter Diamond, McFadden
exploited the duality between the expenditure function and utility maximizing
demand functions to explore such problems as the deadweight burden of taxation
and optimal commodity taxes (Diamond and McFadden, 1974). This paper
established duality as an indispensable tool in modern public economics.
In the 1990s, McFadden has contributed to environmental economics, study-
ing the willingness-to-pay for natural resources. McFadden (1994b) examined
in detail the properties of the contingent valuation method for estimating the
so-called existence value of natural resources and developed new econometric
techniques. With Jerry Hausman and Gregory Leonard, he developed an empir-
ical discrete choice model to assess the welfare losses caused by natural resource
damage (Hausman, Leonard and McFadden, 1995). The model was applied
to recreational demand in Alaska and, in particular, to estimate the welfare
losses to Alaskans caused by the oil spill from the tanker Exxon Valdez in 1989.
McFaddens work in this eld is yet another example of his masterly skills in in-
tegrating economic theory, econometric methodology and substantive empirical
applications.
References
Anderson S., A. de Palma and J-F. Thisse (1992), Discrete Choice Theory of
Product Dierentiation, MIT Press.
Ben-Akiva M. (1973), Structure of Travel Passenger Demand Models, Ph. D.
thesis, MIT.
Ben-Akiva M. and S. Lerman (1985), Discrete Choice Analysis: Theory and
Applications to Predict Travel Demand, MIT Press.
Cowing T. and D. McFadden (1984), Microeconomic Modeling and Policy Anal-
ysis: Studies in Residential Energy Demand, Academic Press.
Debreu G. (1960), Review of D. Luce, Individual Choice Behavior: A Theoret-
ical Analysis, American Economic Review 50, 186188.
Diamond P. and D. McFadden (1974), Some Uses of the Expenditure Function
in Public Finance, Journal of Public Economics 3, 321.
Domencich T. and D. McFadden (1975), Urban Travel Demand: A Behavioral
Analysis, North-Holland.
Dubin J. and D. McFadden (1984), An Econometric Analysis of Residential
Electric Appliance Holdings and Consumption, Econometrica 52, 345362.
12
Fuss M. and D. McFadden (eds.) (1978), Production Economics: A Dual
Approach to Theory and Applications, vols I & II, North-Holland.
Gronau R. (1974), Wage Comparisons - a Selectivity Bias, Journal of Political
Economy 82, 11191143.
Hausman J. and D. McFadden (1984), Specication Tests for the Multinomial
Logit Model, Econometrica 52, 12191240.
Hausman J., G. Leonard and D. McFadden (1995), A Utility-Consistent, Com-
bined Discrete Choice and Count Data model: Assessing Recreational Use
Losses due to Natural Resource Damage, Journal of Public Economics 56,
130.
Heckman J. J. (1974), ShadowWages, Market Wages and Labor Supply, Econo-
metrica 42, 679693.
Heckman J. J. (1976), The Common Structure of Statistical Models of Trun-
cation, Sample Selection and Limited Dependent Variables and a Simple
Estimator for Such Models, Annals of Economic and Social Measurement
5, 475492.
Heckman J. J. (1979), Sample Selection Bias as a Specication Error, Econo-
metrica 47, 153161.
Heckman J. J. (1990), Varieties of Selection Bias, American Economic Review
80, 313318.
Heckman J. J. (1991), Identifying the Hand of the Past: Distinguishing State
Dependence from Heterogeneity, American Economic Review 81, 7579.
Heckman J. J. and B. Honor (1989), The Identiability of the Competing
Risks Model, Biometrica 89, 325330.
Heckman J. J., H. Ichimura and P. Todd (1997), Matching as an Economet-
ric Evaluation Estimator: Evidence from Evaluating a Job Training Pro-
gramme, Review of Economic Studies 64, 605654.
Heckman J. J., R. LaLonde and J. Smith (1999), The Economics and Econo-
metrics of Active Labor Market Programs, Handbook of Labor Economics,
vol 3A, North-Holland.
Heckman J. J. and R. Robb (1985a), Alternative Methods for Evaluating the
Impact of Interventions: An Overview, Journal of Econometrics 30, 239
267.
Heckman J. J. and R. Robb (1985b), Alternative Methods for Evaluating the
Impact of Interventions, in J. J. Heckman and B. Singer (eds.), Longitu-
dinal Analysis of Labor Market Data, Econometric Society Monographs
Series, Cambridge University Press.
13
Heckman J. J. and G. Sedlacek (1985), Heterogeneity, Aggregation, and Mar-
ket Wage Functions: An Empirical Model of Self-Selection in the Labor
Market, Journal of Political Economy 93, 10771125.
Heckman J. J. and B. Singer (1984a), A Method of Minimizing the Impact of
Distributional Assumptions for Duration Data, Econometrica 52, 271320.
Heckman J. J. and B. Singer (1984b), The Identiability of the Proportional
Hazard Model, Review of Economic Studies, 231241.
Heckman J. J. and J. Smith (1995), Assessing the Case for Social Experiments,
Journal of Economic Perspectives 9, 85110.
Heckman J. J., J. Smith and N. Clements (1997), Making the Most Out of
Programme Evaluations and Social Experiments: Accounting for Hetero-
geneity in Programme Impacts, Review of Economic Studies 64, 487535.
Heckman, J. J. and C. Taber (1994), Econometric Mixture Models and More
General Models for Unobservables in Duration Analysis, Statistical Meth-
ods in Medical Research 3, 279302.
Lee L-F. (1978), Unionism and Wage Rates: Simultaneous Equations Models
with Qualitative and Limited Dependent Variables, International Eco-
nomic Review 19, 415433.
Lee L-F. (1994a), Semiparametric Instrumental Variables Estimation of Simul-
taneous Equation Sample Selection Models, Journal of Econometrics 63,
341388.
Lee L-F. (1994b), Semiparametric Two-Stage Estimation of Sample Selection
Models Subject to Tobit-Type Selection Rules, Journal of Econometrics
61, 305344.
Lerman S. and C. Manski (1981), On the Use of Simulated Frequencies to
Approximate Choice Probabilities, in C. Manski and D. McFadden (eds.),
Structural Analysis of Discrete Data with Econometric Applications, MIT
Press.
Lewis H.G. (1974), Comments on Selectivity Biases in Wage Comparisons,
Luce R. D. (1959), Individual Choice Behavior: A Theoretical Analysis, Wiley.
Manski C. (1989), Anatomy of the Selection Problem, Journal of Human Re-
sources 24, 343360.
Manski C. and S. Lerman (1977), The Estimation of Choice Probabilities from
Choice Based Samples, Econometrica 45, 19771988.
14
Manski C. and D. McFadden (1981), Alternative Estimators and Sample De-
signs for Discrete Choice Analysis, in C. Manski and D. McFadden (eds.),
Structural Analysis and Discrete Data with Econometric Applications,
MIT Press.
McFadden D. (1974a), Conditional Logit Analysis of Qualitative Choice Be-
havior, in P. Zarembka (ed.), Frontiers of Econometrics, Academic Press.
McFadden D. (1974b), The Measurement of Urban Travel Demand, Journal of
Public Economics 3, 303328.
McFadden D. (1978), Modelling the Choice of Residential Location, in A. Kar-
lqvist, L. Lundqvist, F. Snickars and J. Weibull (eds.), Spatial Interaction
Theory and Planning Models, North-Holland.
McFadden D. (1981), Econometric Models for Probabilistic Choice, in C. Man-
ski and D. McFadden (eds.), Structural Analysis of Discrete Data with
Econometric Applications, Harvard University Press.
McFadden D. (1987), Regression Based Specication Tests for the Multinomial
Logit Model, Journal of Econometrics 34, 6382.
McFadden D. (1989), A Method for Simulated Moments for Estimation of
Discrete Response Models without Numerical Integration, Econometrica
57, 9951026.
McFadden D. (1994a), Demographics, the Housing Market, and the Welfare of
the Elderly, in D. Wise (ed.), Studies in the Economics of Aging, University
of Chicago Press.
McFadden D. (1994b), Contingent Valuation and Social Choice, American
Journal of Agricultural Economics 74, 689708.
McFadden D. and K. Train (1998), Mixed MNL Models for Discrete Response,
Journal of Applied Econometrics, forthcoming.
McFadden D., K. Train and M. Ben-Akiva (1987), The Demand for Local
Telephone Service: A Fully Discrete Model of Residential Calling Patterns
and Service Choices, Rand Journal of Economics 18, 109123.
Newey W., J. Powell and J. Walker (1990), Semiparametric Estimation of
Selection Models, American Economic Review 80, 324328.
Quandt R. (1970), The Demand for Travel: Theory and Measurement, Heath.
Roy A. (1951), Some Thoughts on the Distribution of Earnings, Oxford Eco-
nomic Papers 3, 135146.
Theil H. (1969), A Multinomial Extension of the Linear Logit Model, Interna-
tional Economic Review 10, 251259.
15
Thurstone L. (1927), A Law of Comparative Judgement, Psychological Review
34, 273286.
Tobin J. (1958), Estimation of Relationships for Limited Dependent Variables,
Vella F. (1998), Estimating Models with Sample Selection Bias: A Survey,
Journal of Human Resources 33, 127169.
Willis R. and S. Rosen (1979), Education and Self-Selection, Journal of Polit-
ical Economy 87, S1S36.
16

Economics Nobel Prize 2000 To 2013

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Economics Nobel Prize 2000 To 2013

Încărcat de

Drepturi de autor:

Formate disponibile

The Economics Nobel Prize 2000 to 2013

Compiled by Rajnish Kumar

. If each asset can be sold short, i.e., held in negative

. Because this money pump could

captures differences in expected

. If an event contains relevant

is the risk-free rate and

is the expected market return. This is a sufficient but

. See Sharpe (1964).

from 12 trading days before until 12 trading

denote the realization of the fundamental

], so that the forecast error,

, must be uncorrelated with any

. Basic pricing theory implies that this ratio should reflect

= , the so-called Gordon formula. In general, however,

) in order to specify explicitly the information

is labor income at time t. The key equation of CCAPM is the

), which is the marginal rate of substitution between consumption today and

is low, the marginal utility

) is high, and thus the ratio of marginal

) is correspondingly low (conversely, the discount factor should be

) can then be calculated from consumption data for any

is a vector stochastic process (sequence of random variables) and is a parameter

(for all assets ), and

. This element has expectation zero and can be

, which is allowed to be any weakly

are independent over time, e.g., as in the case of repeated independent

is a stationary and ergodic process.

, ). As already indicated, the proof allows rather general stochastic temporal

, takes the form

of a given financial asset i is given by

is the risk-free rate,

is the expected return on the market portfolio (i.e., a

is the key measure of systematic risk which

. Assuming that market expectations are rational so that

, the risk-free rate, and

is an exogenous (often called cost-push) shock, i.e.,

= 0 holds as a linearization around a zero-ination

> 0 means that the central bank raises the

will inuence the time-series properties

), i.e., the number of non-searchers

. Flow equilibrium implies:

and rejected for c c

is the expected lifetime utility of a searching agent and \

. The ow utility of a non-trader reads

and :. The equations can be depicted as two positively sloped relationships

, :) space. In general, multiple equilibria are possible and equilibria

is an equilibrium of the original mechanism. Then a direct mechanism can be

. Thus, the resource constraint is c(t)+k(t+1)=Ak(t)

, Country 2 will still mobilize for sure, so

its expected payo

)c + (1 0)/ is at least as large as the payo c from

that consists of the innite repetition of any given game G. Essentially,

. Moreover, he demonstrated that the result holds

is a decision rule that

is thus innite and contains very complex

coincides with the set of

by means of a certain limit of time averages of payos in G.

that constitute threats against deviations from strategies in G

( j), (1 c) is (weakly) exceeds

is the equilibrium (natural) rate of unemployment,

must hold, the policymaker will choose

, 0) is illustrated in Figure 1 as point O.