American Economic Association The American Economic Review

American Economic Association
An Econometric Analysis of U.K. Money Demand in Monetary Trends in the United States
and the United Kingdom by Milton Friedman and Anna J. Schwartz
Author(s): David F. Hendry and Neil R. Ericsson
Source: The American Economic Review, Vol. 81, No. 1 (Mar., 1991), pp. 8-38
Published by: American Economic Association
Stable URL: https://www.jstor.org/stable/2006786
Accessed: 22-02-2019 16:26 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
American Economic Association is collaborating with JSTOR to digitize, preserve and extend
access to The American Economic Review
This content downloaded from 45.5.165.20 on Fri, 22 Feb 2019 16:26:25 UTC
All use subject to https://about.jstor.org/terms
An Econometric Analysis of U.K. Money Demand in
Monetary Trends in the United States and
the United Kingdom
by Milton Friedman and Anna J. Schwartz
By DAVID F. HENDRY AND NEIL R. ERICSSON *
This paper evaluates an empirical model of UK money demand developed by

Milton Friedman and Anna J. Schwartz in Monetary Trends in the United
States and the United Kingdom. Testing reveals misspecification and hence the
potential for an improved modeL Using recursive procedures on their annual
data, we obtain a better-fitting, constant, dynamic error-correction (cointegra-
tion) modeL Results on exogeneity and encompassing imply that our money-
demand model is interpretable as a model of money but not of prices, since its
constancy holds only conditionally on contemporaneous prices. (JEL 210)
In their 1982 book, Monetary Trends in cal findings as support for a range of eco-
the United States and the United Kingdom: nomics hypotheses.1 To isolate the longer-
Their Relation to Income, Prices, and Inter- term relationships that are their primary
est Rates, 1867-1975, Milton Friedman and concern, Friedman and Schwartz estimate
Anna Schwartz present a wealth of empiri- their empirical models from data trans-
formed by averaging annual observations
over phases of NBER reference cycles. In
*Hendry is Professor of Economics at Nuffield Col- our analysis, we first replicate one of their
lege, Oxford, England OX1 1NF, and Visiting Re-
preferred models for U.K. money demand
search Professor at the Institute of Statistics and Deci-
sion Sciences, Duke University, Durham, NC 27706;
on the phase-average data and then evalu-
Ericsson is a staff economist in the International Fi- ate it econometrically to investigate aspects
nance Division, Federal Reserve Board, Washington, of model specification about which the data
DC 20551. This research was supported in part by U.K. are most informative. The outcome indi-
Economic and Social Research Council (E.S.R.C.)
cates the potential for an improved equa-
grants HR8789 and B00220012. We are grateful for the
financial assistance from the E.S.R.C. although the tion but does not entail the form of respeci-
views expressed in this paper are solely the responsibil- fication required. In seeking to construct an
ity of the authors and should not be interpreted as improved model, we formulate an equation
reflecting those of the E.S.R.C., the Board of Gover-
that integrates long-run properties with
nors of the Federal Reserve System, or other members
of their staffs. Helpful discussions with and comments
short-run dynamics, based on the recent
from Chris Allsopp, Julia Campos, Nicholas Dimsdale, merging of the theories of error-correction
Rob Engle, John Flemming, Milton Friedman, David and cointegration. Moreover, because
Germany, John Geweke, Dale Henderson, Kevin phase-averaging may entail a loss of infor-
Hoover, Jeroen Kremers, Bonnie Loopesko, Mary
mation about relevant parameters, we re-
Morgan, John Muellbauer, Adrian Neale, Charles Nel-
son, Jean-Francois Richard, Ken Rogoff, John Taylor, turn to analyzing the underlying annual ob-
Ken Wallis, Arnold Zellner, and several anonymous servations. Finally, the resulting model is
referees are gratefully acknowledged. We are indebted critically evaluated.
to Jeroen Kremers, Cathy Fitzgerald, and Aileen Liu
for their invaluable assistance with the calculations.
This is an extensively revised and shortened version of
papers presented to the Bank of England's Panel of 1For helpful reviews and summaries of Friedman
Academic Consultants (Hendry and Ericsson, 1983) and Schwartz (1982), see inter alia Charles Goodhart
and circulated by the Federal Reserve Board (Hendry (1982), Thomas Mayer (1982), Tim Congdon (1983),
and Ericsson, 1985). Basil Moore (1983), and Michael Artis (1984).
VOL. 81 NO. 1 HENDRYAND ERICSSON: MODELING MONEY DEMA ND 9
The sequence in the previous paragraph In Section I, we consider the data and
involves a natural progression from model data transformations used by Friedman and
discovery to model evaluation through repli- Schwartz (1982). In Section II, we discuss
cation and testing, and then via new conjec- the money-demand relationships they esti-
tures back to discovery, seeking models that mated from the phase-average data for the
account for previous findings and explain United Kingdom and evaluate many of their
additional phenomena. The paper's main empirical claims about the selected money-
objective is to achieve this last goal for U.K. demand equation. The statistical framework
money-demand models over the period and its associated concepts are briefly ex-
1878-1970. An additional objective is to posited in Section III to clarify the joint
exposit an econometric framework that destructive and constructive roles for tests.
makes precise the notion of an improved In Section IV, we constructively apply that
model, explains the construct of accounting approach to modeling money demand using
for previous findings (denoted encompass- the annual U.K. data series. Our conclu-
ing), and delineates the criteria for model sions are presented in Section V.
evaluation. This enables us to clarify the
concepts of exogeneity, invariance, encom- I. The Data Series and Transformations
passing, and cointegration in the context of
an important economics debate. The empir- In Friedman and Schwartz's (1982) and
ical model reported below improves upon our studies of U.K. money demand, the
the phase-average U.K. money-demand basic data series are annual values for the
equation from Friedman and Schwartz, our United Kingdom from 1871 to 1975 of the
first annual-data specification (proposed in broad money stock (M), real net national
1983), and those in the studies that the income (I), the price level (P), short-term
latter stimulated, thus illustrating a progres-and long-term nominal interest rates (RS
sive research methodology in action. and RL), population (N), and high-powered
0.9 18
vt (left axis)
0.8 --- RSt (right axis: percentage) 16
0.7 14
0.6 12
0.5 10
0.4 Nj 8
0.3 '4 6
0.2 ij I
0.1 ~ /r /V~12
0 0~~~~'
1870 1880 1890 1900 1910 1920 1930 1940 1 950 1960 1970 1980
Year
FIGURE 1. THE LOG OF THE VELOCITY OF MONEY (V') AND THE YIELD ON THREE-MONTH BILLS (RSZ)
(ANNUAL DATA)
10 THE AMERICAN ECONOMIC REVIEW MARCH 1991
(m-P)t
- -it-0.6
8.5
7.5
6.5 1 I I I I
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980
Year
FIGURE 2. THE LOG OF THE REAL MONEY STOCK ([m - P]) AND THE LOG OF REAL INCOME (it)
(ANNUAL DATA)
money (H); and the price level (P *) in the war period, velocity is lower and more
United States. All series are in Friedman volatile. It falls during the Second World
and Schwartz's tables 4.8 and 4.9; for details War, reaching its lowest observed level in
of construction and definition of the series, 1947, corresponding to the lowest level for
see both Friedman and Schwartz (1982 Ch. interest rates (of 0.5 percent per annum).
4, 5) and our Appendix, which note reserva- Next, velocity trends upwards, more than
tions on the measurement and interpreta- doubling by 1970, while interest rates fluc-
tion of those series. Unless otherwise indi- tuate increasingly around an upward trend.
cated, capital letters denote both the generic Finally, between 1971 and 1975 velocity falls
name and the level; logs of scalars are de- while RS rises.
noted by lowercase letters. Figures 1, 2, and Some insights into this behavior of veloc-
3 respectively show the annual time series ity can be gleaned from Figure 2, which
for velocity V (- PI/M) and RS, real plots two of its constituents, m - p and i.
money (M/P) and real income, and nomi- The interwar period begins with a sharp fall
nal money and prices.2 in i being matched by a large rise in m - p,
From Figure 1, three distinct episodes are which then remains systematically above i
discernible in the behavior of velocity, coin- with its greatest departure from i in 1947.
ciding with historical periods before the First Thereafter, m - p falls steadily until around
World War (1871-1914), interwar (1918- 1960, whereas i rises considerably, and then
1939), and after the Second World War m - p "levels off' until its sharp rise in
(1945-1975). Initially, velocity and RS cycle 1971. Thus, between 1947 and 1970 mea-
around constant levels. Then, in the inter- sured velocity more than doubles because
m - p drops while i rises. Lastly, Figure 3
shows nominal money and the price level.
2When there is no loss of clarity, we often refer to These series are highly correlated in levels
v (= p + i - m) as velocity. although, overall, money increases 5.6 times
VOL. 81 NO. 1 HENDRYAND ERICSSON: MODELING MONEY DEMA ND 11
11
mt
--- Pt + 7.8
10
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980
Year
FIGURE 3. THE LOG OF THE NoMINAL MONEY STOCK (Me) AND THE LOG OF THE IMPLICIT PRICE
DEFLATOR (ps) (ANNUAL DATA)
relative to p (real income increases by 6.7 rates of change. The latter are calculated
times), and during 1920-1922 a large fall in from least-squares slopes fitted to triplets of
p occurs alongside a small fall in m. To overlapping phases (weighted appropriately)
represent adequately the century of data, an and are arithmetically nearly the same as
econometric model of U.K. money demand differences from one expansion (or contrac-
must encompass such features. tion) to the next.4 The data to which these
Since Friedman and Schwartz aim to ex- moving-average filters are applied are the
amine the long-period movements in the logarithms of money, prices, incomes, and
data, they do not analyze these annual se- population, and the original values of inter-
ries directly but first transform them by est rates. We denote phase-average data by
averaging separately over contraction and an overbar (e.g., ij for the jth of J phase-
expansion phases of data-selected choices average observations of the logarithm of
of reference business cycles. From some money stock). These data are the basis for
hundred years of annual U.K. observations, the analysis in Section II, whereas in Sec-
Friedman and Schwartz identify 37 such tion IV we utilize the original annual time
phases with average lengths of 2.1 and 3.4 series. The relationship between the phase-
years respectively for contraction and ex- average and the annual series for velocity is
pansion phases. These phase averages, shown in Figure 6, discussed below.
weighted by a function of duration, are the
basic units of analysis (pp. 75-9 and table
5.8)3 and are analyzed both in levels and as 4There are typographical errors in the reported
values for "rates of change" in table 5.10 affecting
observations 18-21. We corrected those errors follow-
3Unreferenced page numbers and table numbers ing the procedures described by Friedman and Schwartz
when "table" is uncapitalized refer to Friedman and (1982 section 3.2): our amendments are reported in
Schwartz (1982). Hendry and Ericsson (1985 appendix F).
II. Testing the U.K. Money-Demand Model unit income elasticity (pp. 280-4). From
on the Phase-Average Data equation (d), I does not depend upon M,
whereas P does [eq. (c)] and, hence, so does
In this section, we summarize the central PI [eq. (b)] (cf. pp. 422, 351).
empirical results obtained by Friedman and Equations (a)-(d) represent types of mod-
Schwartz (1982) for U.K. money demand els (of money, nominal incomes, prices, and
using the phase-average data, replicate one output, respectively), but from our perspec-
of their preferred models, and econometri- tive their joint existence raises difficulties.
cally evaluate it. Table 1 reports estimates Because M is a regressor in equations
of four equations out of those that Fried- (b)-(d), M must be (weakly) exogenous if
man and Schwartz record for the United the coefficients in these regressions are to
Kingdom as establishing evidence concern- be interpreted as parameters of interest (cf.
ing the demand for money, and the influ- Robert Engle, Hendry, and Jean-Francois
ences of money on income and on prices. Richard, 1983). With M exogenous, equa-
"Levels" and "rates of change" equations tion (a) is determining P, not M; but so is
are analyzed by Friedman and Schwartz but, equation (c). Conversely, if M does not
as they see little to choose between them maintain its exogeneity status in equation
(e.g., see p. 286), we focus on the equations (a), P and M are jointly determined, and
in levels: full details can be found in Fried- hence equation (c) must be invalid. How-
man and Schwartz. We summarize their in- ever, since Friedman and Schwartz take (a)
terpretation of the estimates in Table 1 as as one of their "final equations" for U.K.
follows. Equation (a) is their preferred U.K. money demand (p. 281), we will evaluate it
money-demand equation and has a nearly on that basis.
TABLE 1 -EQUATIONS FOR U.K. MONEY DEMAND, NOMINAL INCOMES, PRICES, AND OUTPUT FROM FRIEDMAN
AND SCHWARTZ (1982) (PHASE-AVERAGE DATA)
Equation Source
(a) ( n-p-=n) = 0.16 + 0.88(i - n) - 11.16RN- 0.22G(p + i) + 1.4W + 215 p. 282

(0.08)(18.13) (3.42) (0.74) (2.38) (7.56)
(a = 5.54, R2 = 0.970, dw = 1.51, flJ[18, 12] = 6.3)
(b) (p + i) = 0.38 + 1.01m + 14.17RN+ 0.53G(p + i) - 1.1W- 19S p. 349
(a = 5.99, R2 = 0.9977, dw = 1.33, dJ[l8, 12] = 5.7)
(c) p = -5.99 - 0.015i + 1.02m + 0.94RS- 1.10G(p + i) p. 420

(31.1) (7.4) (17.4) (0.9) (2.8)
(a = 6.0, dw = 1.01, lJ[l,5 10] = 3.1)
(d) i = 6.50 + 0.017- - 0.05m + 2.34R3+ 2.20G(p + i) p. 420

(34.2) (8.3) (0.9) (2.4) (5.7)
(v = 5.9, dw = 0.97, lJ[l,5 10] = 2.1)
Notes: Notation is as in Sections I and II, but here parentheses enclose t values, as reported in the original text. No t
ratios or standard errors appear in Friedman and Schwartz (1982) for the equation labeled (b) above. RN =
(RS)-H/M; G(-) denotes a rate of change; and W and S are the dummies for postwar adjustment and demand
shift, rescaled by 1/100 (see Appendix). Phase averages 12-14 and 24-26 are omitted in (c) and (d). Here and in
Table 3, values of a with logarithmic regressands are quoted as approximate percentages relative to the level of the
regressand in its original units [e.g., if log(Y) is the regressand, a is a percentage of Y]. Values for dw and qI are
our calculations and are not reported in Friedman and Schwartz. Our equations replicating (c) and (d) had
somewhat smaller values of a than those in Friedman and Schwartz.
VOL. 81 NO. 1 HENDRYAND ERICSSON: MODELING MONEYDEMAND 13
We could closely replicate equation (a): conditional on treating it as provisionally

valid, many empirical phenomena are ex-
cluded. Consequently, a model is testable
(1) (m - p - i)j against the potential occurrence of such
phenomena, using tests that would be valid
= 0.012 + 0.885(i -n
given the assumptions of that model. The
(0.19) (0.049)
resulting tests have central distributions with
approximately 5-percent critical levels un-
- 11.21RN - 0.22G(,D + 0j der the null hypothesis of correct specifica-
(3.3) (0.29)
tion and noncentral distributions under their
respective alternatives. Additionally, they
+ 1.37Wj + 20.6Sj may have power to reject alternatives other
(0.58) (2.7)' than those against which they were de-
signed or have larger implicit than explicit
(J= 36, = 5.66 percent, dw = 1.51, R2= null hypotheses (cf. Grayham Mizon and
0.97) where RN is the differential between Richard, 1986).
RS and the "own-yield" on money, G(p + i) Constancy is a major issue for money-
is the growth rate of nominal income (inter- demand equations (see John Judd and John
preted by Friedman and Schwartz as proxy- Scadding, 1982), and Friedman and
ing for the rate of return on physical assets), Schwartz regard their own U.K. money-
W is a dummy for "postwar adjustment" demand equation as being constant: "[A]
(p. 228), and S is a data-based dummy for more sophisticated analysis [than the simple
"[a]n upward demand shift, produced by quantity theory] reveals the existence of a
economic depression and war..." (p. 281) stable demand function for money cover-
during 1921-1955. S captures a 21-percent ing the whole of the period we examine"
shift in (1) and so accounts for much of the (p. 624; see also pp. 7,14,64,283). However,
variation in real per capita money, condi- they do not formally test for constancy, and
tional on real per capita income. Figure 4 many investigators would regard the need
shows the fitted and actual values for (mn -for the data-based shift dummy S spanning
p5 - ni) derived from (1). Throughout, esti- one-third of the sample as prima facie evi-
mated standard errors are shown in paren- dence against the model's constancy. Refit-
theses (except in Table 1), a^ is the standard ting (1) to the first half of the sample and
deviation of the residuals [e.g., in (1), as a predicting over the second half (doing each
percentage of M/(PN)], adjusted for de- with both S and W), we tested for constancy
grees of freedom, and R2 is the squared using Gregory Chow's (1960 pp. 594-5)
multiple correlation coefficient for the cor- statistic: that yielded 71[18, 121= 6.3, which
responding unweighted regression.5 exceeds the 1-percent point of the F distri-
Having replicated equation (a) by (1), we bution. Although the one-step-ahead 95-
evaluate (1) against a range of alternative percent confidence interval based on Sf is
models. A summary of our approach is pro- larger than +11 percent, parameter non-
vided in Section III; here, we focus on the constancy can be detected, revealing an as-
usual concerns of parameter constancy, pect about which the present data set is
price homogeneity, omitted variables, ho- informative. The values of Sr for the corre-
moscedasticity, and normality. Once an em- sponding subperiods are markedly different:
pirical model has been formalized, then, 2.8 percent and 6.0 percent, respectively.
Even if these differences were due solely to
5Like those in Friedman and Schwartz, all our esti- heteroscedasticity, the inferences that
mates using phase-average data are based on weighted Friedman and Schwartz draw from their
least squares, correcting for the different phase lengths. regressions would be invalid because the t
However, parameter estimates are not very different
whether ordinary or weighted least squares are used.
ratios are biased. Moreover, nonconstancy
All empirical calculations are from PC-GIVE (see is not restricted to the residual variance.
Hendry, 1989). Figure 5 records the recursive estimates of
4.6
4.4 - - / 1
4.2
3.8
3.6
3.4 L l l 1 1 l
0 5 10 15 20 25 30 35
Phase observation
FIGURE 4. EQUATION (1): ACTUAL AND FITTED VALUES FOR (th - phj (THE TRANSFORMED
REGRESSAND)
the interest rate coefficient from phases is exogenous. Treating M as endogenous

16-36, together with a confidence region and relaxing unit price homogeneity in (1),
based upon plus-or-minus twice its esti- we obtain
mated standard error at each sample size.6
As the sample size increases, the estimated
coefficient changes substantially relative to (2) (mn- i- F)j
its estimated uncertainty, with the final esti-
mate lying outside the initial confidence in- -1.01 + 0.68(i- R)j + 0.128p
terval. Such evidence refutes constancy in (0.38) (0.08) (0.043)
Friedman and Schwartz's reported model.
Since the dependent variable in (1) is
-18.4RNj- 0.04G(p+i)j
m - p- - n-, price homogeneity can be tested (3.8) (0.26)
under two distinct maintained hypotheses
about exogeneity, namely, whether M or P
+ 1.05W. + 16.3S
(0.53) (2.8)
6In textbook notation, Figure 5 plots the ith coef-
ficient and its estimated standard error: ,l +
(J = 36, 6f= 5.03 percent, dw = 1.85). The
2ESE(.8jj), j=16,...,36, where f3 = (XXj)-11X4Y,
Xj = (x1 .1 xi)', Yj = (y1 ... yj)', and ESE( 11) coefficient on pj is significant at the 99-per-
cent level. If, instead, the exogeneity of M
= +/ diag, [2 (XX1) X. Under the null hypothesis
is maintained so that (1) purports to explain
of correct specification, ,aq - ,Bi and ESE( 3ij) - 0 as
i -j oo (see Andrew Harvey [1981 pp. 54-9] on the then adding mn, and R, to (1) to relax the
calculation of recursive least squares and associated homogeneity assumption yields F[2,28s =
test statistics and Kerry Patterson [19861 for an applica- 20.56, again rejecting the specification
tion). For annual data, the index j becomes t. of (1).
VOL. 81 NO. 1 HENDRYAND ERICSSON: MODELING MONEY DEMAND 15
10
coefficient estimate
~_ ---coefficient estimate ?2 standard errors
o k \ /~~ss\;;-;-;
10
1 0 5
15 _- \ X s_o-
20 I I I I I I I i
16 18 20 22 24 26 28 30 32 34 36
Phase observation
FIGURE 5. EQUATION (1): RECURSIVE ESTIMATION OF THE COEFFICIENT ON RNj AND ITS ESTIMATED
STANDARD ERROR
Next, to test for the absence of a trend, residual skewness and excess kurtosis is
we estimate (3): 65[2] = 6.9 in (i).7 Testing the normality of
the disturbances is of interest for two rea-
sons. First, Friedman and Schwartz often
base their statistical inferences on critical
values of the F distribution (e.g., see pp.
= - 13.92 + 0.14(i - n)
232, 236), and given the small sample size
(2.99) (0.16)
(J = 36), those inferences may be affected
substantively by nonnormal disturbances.
- 16.ORN1 + 0.23G(5 + i).
Second, the Jarque-Bera statistic also has
(2.7) (0.24)
power against other alternatives, as sug-
gested by it being insignificant when unit
+1.80W,+ 6.1S + O.0090i. homogeneity of money with respect to prices
(0.46) (3.7) (0.0019)' is not imposed.
Taking this evidence together, equation
(J = 36, r = 4.35 percent, dw = 1.99). The (1) is not an adequate characterization of
coefficient on i1 is statistically significant. By
the data and is not consistent with the hy-
way of comparison, when Friedman and
pothesis of a constant money-demand equa-
Schwartz tested for trends and price homo-
geneity in more restrictive models, they did
not obtain rejections on this data set 7Given that the number of regressors fitted is often
(pp. 253-9). large relative to the sample size, we modify Jarque
and Bera's (1980 p. 257) statistic to be 65[2] =
Although Friedman and Schwartz did not .[2 +-2
[(J - k)/6HSK +EK /4], where SK and EK are
test for it, normality of the disturbances is
skewness and excess kurtosis, respectively. 55[2I is
also rejected: Carlos Jarque and Anil Bera's asymptotically distributed as X2 under the null hy-
(1980) X2] statistic for testing normality via pothesis of normality, in which case SK = EK = 0.
tion, homogeneous of degree one in prices outcome because of a highly nonnormal er-
over the last century, even though Friedman ror distribution. These problems frequently
and Schwartz state that "[t]his parallel- arise in approaches that proceed by general-
isml8] is a manifestation of the stable izing de- simple models (see Hendry, 1979).
mand curve for money plus the excellence Thus, we disagree with the methodology
of the simple quantity theory approxima- adopted by Friedman and Schwartz of bas-
tion" (p. 7). The equations for nominal ining many of their inferences on the analysis
come and prices in Friedman and Schwartz of simple empirical models:
were not further investigated because, as
explained in Hendry and Ericsson (1985 ap- Our ultimate objective is an expla-
pendix D), they are approximately renor- nation of the behavior of velocity,
malizations of their money-demand regres- which is to say, of the quantity of
sion and so do not provide additional infer- money demanded, that takes account
ences (cf. pp. 344, 417). Additionally, any simultaneously of all the variables af-
inferences based on (b)-(d) are hazardous fecting velocity. Nonetheless, we be-
because the split-sample Chow statistics for lieve that this ultimate objective is
(b) and (c) are highly significant, and the better approached indirectly, by exam-
ining variables one or two at a time,
Durbin-Watson statistics in (c) and (d) are
than by what has become the prevail-
less than the 5-percent lower bound.
ing fashion in econometric work, the
When interpreting the outcomes of the
immediate computation of multiple
above tests, four points should be borne in regressions including all variables that
mind. First, the rejections are not mutually can reasonably be regarded as rele-
independent sources of information, since vant. We believe that the indirect ap-
the implicit assumptions of any given test proach yields insights that cannot be
are shown to be invalid by the outcomes on obtained from the more sweeping ap-
other tests. Secondly, none of the relevant proach-that multiple correlations
hypotheses could have been tested by Fried- with many variables are almost impos-
sible to interpret correctly unless they
man and Schwartz for this equation without
are backed by more intensive investi-
their having obtained a rejection. This shows
gations of smaller sets of variables.
the power of statistical evaluation to reveal Accordingly, we shall proceed in the
the potential for model improvement and following sections to consider vari-
the ability of the data to discriminate be- ables one or two at a time, and reserve
tween empirical models. Thirdly, discover- to section 6.7 estimating their simulta-
ing that their empirical model has a variety neous effect. [footnote:] The indirect
of specification problems has no implica- approach played a critical role in the
tions for the existence (or otherwise) of a formulation of the multiple regres-
correctly specified empirical model of money sions that we calculate in section
6.7.... (p. 215)
demand being homogeneous in prices and
having constant parameters. Finally, rejec-
tion of any null hypothesis does not imply This quote raises four methodological is-
that the alternative is correct. For example, sues. First, the claim that simple models
constancy may have been rejected because yield insights is frequently mistaken, since
of dynamic misspecification, and not be- the associated measures of relationship and
cause the underlying money-demand model of uncertainty are misleading unless such
has nonconstant parameters; or, in small models are coherent with the data. Sec-
samples, an F test could yield a misleading ondly, it is incorrect to test hypotheses in
one model and infer that the outcome will
hold in generalizations of that model unless
all the additional influences are orthogonal
80f nominal income and the nominal quantity of
to those already included and remain so
money, and of the rate of change of nominal income
and the rate of change of the nominal quantity of throughout the sample. Thirdly, although
money. we concur that it is difficult to interpret
multiple regressions, this does not justify being those which Friedman and Schwartz
fitting misspecified submodels. Finally, the chose to analyze.
penultimate sentence of the above quote
conflates the issues of estimation and mod- In order to isolate the longer-term
eling by seeming to imply that the only step relations that are our primary concern,
remaining after examining the smaller sets we have converted our basic data,
is to estimate their joint effect: that would which are annual, into a form de-
be true only if there were no interactions. signed to be free from the shorter-term
movements that are called business cy-
The crucial issue is that a reject outcome at
cles..... The device we have used to
any stage of modeling invalidates all previ-
free the data from cyclical fluctuations
ous inferences, so that decision-taking dur- is to take as our basic observation an
ing model generalizations is ill-founded. In average of annual observations over a
Section III, we briefly describe an alterna- cycle phase.... (p. 13)
tive methodology which resolves many of
these issues and also reveals that there are This device is taken from the NBER ap-
other approaches to understanding general proach to business-cycle analysis, as docu-
models. Moreover, a constructive empirical mented by Arthur Burns and Wesley
task remains to be undertaken, and we turn Mitchell (1946). Friedman and Schwartz ar-
to that in Section IV, using the model de- gue that phase-averaging reduces serial cor-
veloped by Friedman and Schwartz as a relation arising from the business cycle
benchmark against which to compare other (p. 78) and attenuates measurement errors
equations. (p. 86). Doing so is important to sustain the
Before proceeding, we address the ques- validity of their statistical analyses, since,
tion of whether to use the annual observa- for example, no explicit account is made for
tions or the phase-average data, the latter biases in coefficient estimates and estimated
1.0 Annual data (vt)

- -- Phase-average data (Vj)
0.8 /
//
0.6 t/
0.4
0.2
1880 1900 1920 1940 1960 1980

Year
FIGURE 6. THE LOG OF THE VELOCITY OF MONEY: ANNUAL AND PHASE-AVERAGE DATA
standard errors resulting from residual au- ships that are interpretable in light of eco-
tocorrelation. Friedman and Schwartz claim nomic knowledge, remain reasonably con-
that phase-averaging successfully isolates stant over time, and account for the findings
longer-run behavior: "All in all, we con- of preexisting models. For general exposi-
clude that our phase bases do eliminate the tions of the associated methodology, see
bulk of the systematic cyclical fluctuation" Hendry and Richard (1982, 1983), Hendry
(p. 81). (1983, 1987), Hendry and Kenneth Wallis
The use of phase-average data raises three (1984), Ericsson and Hendry (1985),
distinct issues: the theoretical statistical ef- Christopher Gilbert (1986), and Aris Spanos
fects of phase-averaging (qua aggregation), (1986).
the observed effects of phase-averaging on The formal methodology is based on the
Friedman and Schwartz's data, and the im- statistical theory of data reduction. Empiri-
pact of selecting the intervals over which to cal models arise from transformations and
average by prior analysis of an interrelated reductions of the data generation process
data set. Julia Campos, Ericsson, and (DGP, a shorthand for the actual mecha-
Hendry (1990) address these issues and show nism that generated the observed data),
that (i) the aggregation from annual data to which is characterized by the joint density
phase averages entails a loss of information of the observable variables. The main steps
by retaining only a subset of a previously producing models from the DGP are aggre-
averaged annual data set, (ii) phase-averag- gation, algebraic transformation of data,
ing does not materially reduce the serial marginalization, sQquential conditioning,
correlation in Friedman and Schwartz's contemporaneous conditioning, truncation
data, and (iii) phase-averaging involves a of lag length, and linearization. Each step
data-based set of filters but ignores the sta- entails a corresponding reduction and/or
tistical consequences of such filters on later transformation of the parameters of the
inferences. Figure 6 illustrates points (i) and DGP to produce the reduced reparameteri-
(ii) for the time-series on velocity by jointly zation in the econometric model. As a con-
plotting vt and -j. Neither the variance nor sequence, all empirical models are derived
the autocorrelation of -j is notably smaller entities. Additionally, given the observed
than that of vt, whereas period-to-period data and a formal model specification, the
changes in -j are much larger than those in model's error process must contain every-
vt due to the observations being a phase thing in the data not explicitly allowed for
rather than a year apart. by the model, and hence it also is derived
Thus, because of these unsatisfactory rather than autonomous. This implies that
characteristics of phase-averaging, we ana- models can be designed to satisfy preas-
lyze Friedman and Schwartz's original an- signed criteria, as when residual autocorre-
nual observations in an attempt to develop lation is removed by a Cochrane-Orcutt
a money-demand equation that encom- transformation. While this example may be
passes (1) and accounts for its failures. We optimal in some circumstances, in others it
next summarize the statistical framework entails invalid restrictions (cf. Hendry and
and the associated theory of testing on which Mizon, 1978).
both the above analysis and the later mod- The applied methodology distinguishes
els are based. between the constructive and destructive
roles of econometrics. The former corre-
III. Econometric Modeling and sponds to John Herschel's (1830) "context
Economic Theory of discovery" and concerns issues of model
design such as research efficiency. The lat-
This section discusses the nature of em- ter is Herschel's "context of justification"
pirical econometric models and serves as a and, for model evaluation, was illustrated in
guide to our uses of test statistics. Modeling Section II. Criteria relevant to both design
is seen as an attempt to characterize data and evaluation are entailed by the condi-
properties in simple parametric relation- tions required to validate the (sometimes
implicit) sequence of reductions and trans- exogeneity, D) parameter constancy, E) data

formations from which empirical models admissibility, and F) encompassing. A model
arise. Clearly, the validity of one-to-one data is said to be congruent if it satisfies all of
transformations is not an issue. However, A-F and hence captures the salient fea-
reductions may entail a loss of information, tures of the data and delivers reliable infer-
and to judge that loss we use a taxonomy ences on economic issues. Many of the cri-
that partitions the universe of available in- teria from A-F are standard and include
formation into disjoint sets: a) economic the statistical and economic interpretation
theory, the b) past, c) present, and d) future of estimated coefficients and the validity of
data of one's own model, e) the measure- a priori restrictions, goodness-of-fit and the
ment system of the data, and f) the data of absence of both residual autocorrelation and
alternative models. These information sets heteroscedasticity, valid exogeneity, predic-
generate the following criteria for design (in tive ability and parameter constancy, appro-
the context of discovery) and for evaluation priate functional form, and the ability of a
(in the context of justification): A) theory model to account for the properties of alter-
consistency, B) innovation errors, C) weak native models.
TABLE 2-CRITERIA FOR EVALUATING AND DESIGNING ECONOMETRIC MODELS
Null
hypothesis Alternative hypothesis Statistic Sources
B first-order residual autocorrelation dw James Durbin and Geoffrey Watson

(1950, 1951)
B qth-order residual autocorrelation 62[q]; George Box and David Pierce (1970);
?)2[q, T - k - q] Leslie Godfrey (1978), Andrew Harvey
(1981 p. 173)
B q invalid parameter restrictions ?73[q, T - k - q] J. Johnston (1963 p. 126)
B qth-order ARCH 64[q]; ?74[q, T - k - 2q] Engle (1982)
B skewness and excess kurtosis 65[2] Jarque and Bera (1980)
B heteroscedasticity quadratic in 716[q, T - k - q - 1] White (1980), Desmond Nicholls and

regressors (q quadratic terms) Adrian Pagan (1983)
B qth-order RESET ?77[q, T - k - q] J. B. Ramsey (1969)
C q instrumental variables not 8[q- k]; J. Denis Sargan (1958, 1964);

independent of errors 78[q- k, T - q] Sargan (1980a p. 1136)
D parameters not constant over Y)9[k, T - 2k] R. A. Fisher (1922), Chow (1960 pp.
subsamples 595-602)
D predictive failure over a subset Y71[q, T - k - q] Chow (1960 pp. 594-5)

of q observations
Notes: Null hypotheses correspond to criteria for design, as designated in the text. There are T observations and k
regressors in the model under each null hypothesis. The value of q may differ across statistics, as may those of k
and T across models and samples. ei[q] and r17[q, r] denote statistics that have central X 1] and F[q,r] distributions,
respectively, under a common null and against the ith alternative. Thus, 62[q] and ?72[q, T - k - q] both test fo
qth-order residual autocorrelation. We have labeled the Chow statistic Y71[q, T - k - q] both to highlight the
preeminence of the issue of constancy in the substantive debate on monetary behavior and because of its crucial
role as an indirect test of weak exogeneity through testing the conjunction of hypotheses embodied in super
exogeneity. The covariance test statistic Y79[k, T - 2k] is often (and confusingly) referred to as the "Chow statistic"
although Chow (1960 p. 592) was well aware of its presence in the literature.
In the context of discovery, satisfying a long-run relationship between variables with

given criterion implies no loss of informa- a statistical model of those variables.9 For
tion from the associated reduction. In the convenience, we adopt the time-series nota-
context of evaluation, the criteria are inter- tion that a nonstationary variable integrated
pretable as null hypotheses, and Table 2 of order d [i.e., I(d)] requires differencing d
summarizes the associated statistics. Since times to make it stationary [i.e., 1(0)]. For
satisfying these criteria is a necessary condi- I(1) variables y, and zt, arbitrary linear
tion for an empirical model to be congru- combinations y, + 8z, = u, are also I(1). If a
ent, failing any one of them is a sufficient value of 8 exists such that ut is I(O), then yt
condition for rejection. Conversely, since the and zt are "cointegrated" and so do not
route chosen for discovery cannot affect the drift too far apart (e.g., because of agent
intrinsic validity (or otherwise) of the finally behavior). Proportionality between Y and Z
selected model, issues of model design only implies 8 = - 1. Cointegration of yt and zt
concern the efficiency of alternative model- can be tested by testing for the absence of a
building strategies. Thus, we emphasize both unit root in ut, that is, whether ut i-s I(O)
explicitly designing empirical economic rather than I(1), with I(O) being necessary
models to be congruent and rigorously test- for theory consistency.
ing the congruency of given model specifi-
cations. Since the latter takes models as B. Innovation Errors
formulated by their proprietors, the evalua-
tive role of econometrics is not impugned Dynamic specification influences model
by any of the current methodological design because of the statistical and eco-
debates (cf. Christopher Sims, 1980; nomic importance of (white-noise) innova-
Edward Leamer, 1983; Michael McAleer, tion errors. As a rule, dynamic misspecifi-
Adrian Pagan, and Paul Volker, 1985; cation invalidates inference, so dynamics
Hendry and Mizon, 1990). The failure to cannot be safely ignored. There are many
emphasize evaluation and testing is a major possible dynamic formulations for econom-
lacuna in Leamer's (1983) analysis, and our ic models (e.g., see Hendry, Pagan, and J.
paper is a counterexample to the view that Denis Sargan, 1984); but since cointegration
data evidence is not able to discriminate implies and is implied by the existence of a
between alternative hypotheses. Rigorous (dynamic) error-correction representation of
evaluation of empirical claims seems a nec- the relevant variables, we consider the
essary first step toward taking the con out of properties of and intuition behind error-
economics, and indeed, Halbert White correction models (cf. Sargan, 1964; James
(1988) has shown that a sufficiently thor- Davidson et al., 1978 pp. 679-82; Mark
ough testing procedure will arrive at a well- Salmon, 1982). For expositional simplicity,
specified characterization of the DGP with suppose that only the current values of y
confidence approaching certainty as the and z and their one-period lags matter, that
sample size grows without bound. The re- y and z are cointegrated, and that the rela-
mainder of this section discusses the above tionship is linear, in which case
criteria, their interrelationships, and their
role in progressive research strategies. (4) yt a + alyt-I + 80zt + lzt_1 + Pt
A. Theory Consistency Pt IN(O, o,)
Economic theory often suggests long-run where IN means "is distributed indepen-
relationships between economic variables,
such as between two variables Y and Z
(e.g., money and income) of the form Y =
9See Clive Granger (1981), Hendry (1986), Engle
KZ where K is a constant. In logs, that
and Granger (1987), Jeff Hallman (1987), Peter Phillips
theory becomes y = K + Z with K= ln(K). (1987), and James Stock (1987). Engle (1987) and Juan
Cointegration links the economic notion of a Dolado and Tim Jenkinson (1987) survey the literature.
dently normally." In (4), long-run homo- malization when the economy exhibits struc-
geneity between y and z requires that a1 + tural change, data can preclude some
.80 + /83 = 1. Rewriting the autoregressive- choices. Thus, normalization and condition-
distributed lag relationship in (4) by impos- ing both lead to the next two issues: exo-
ing that restriction gives geneity and parameter constancy.
(5) Ayt=a+f3Azt+y(yt-i-Zt-l)+vt C. Weak Exogeneity
where a, f3, and y are the corresponding The four distinct concepts of exogeneity
unrestricted parameters.10 Intuitively, the (weak, strong, super, and strict), discussed
term f3Azt reflects the immediate effectby ofEngle et al. (1983), correspond to differ-
a change in zt on yt. The term y(yt-1- ent notions of being "determined outside
zt-1) (with y negative for dynamic stability the model under consideration" according
and, therefore, cointegration) is statistically to the purposes of the inferences being con-
equivalent to having y(Yt-j - K - zt-1) in- ducted (i.e., conditional inference, predic-
stead, and hence reflects the effect on Ayt tion, policy analysis, and forecasting, re-
of having yt-1 out of line with K+Zt-1. spectively).12 In no case is it legitimate to
Such discrepancies could arise from errors "make variables exogenous" simply by not
in agents' past decisions, with the presence modeling them. Weak exogeneity can occur
of y(yt- - zt1) reflecting their attempts when agents act contingently on available
to correct such errors; so, (5) belongs to the information. If agents use that information
class of error-correction models (ECM's).11 efficiently, innovation errors are implied, re-
For a steady-state growth rate of Zt equal lating back to the issue of dynamic specifi-
to g (i.e., g = Azt = Ayt) and vt = 0, then cation. Weak exogeneity is testable, often as
solving (5), we have an implication of super exogeneity (and so
of models having constant parameters). Sec-
(6) Yt=Ztexp{[-a+g(1-f3)]/y} tion IV further discusses exogeneity in the
context of the empirical model.
which reproduces the nonstochastic steady-
state theory of proportionality between Yt D. Parameter Constancy
and Zt. This example is readily extended to
include further lags, nonproportionality, and Parameter constancy is at the heart of
nonlinearity, and so is representative of a model design from both statistical and eco-
large class of models that satisfy steady-state nomic perspectives. Since economic systems
economic-theoretic restrictions and allow for are far from being constant and the coeffi-
general dynamic responses. cients of derived ("nonstructural" or "re-
The choice of normalization of the coin- duced form") equations may change when
tegrating vector is an unresolved issue, but any of the underlying parameters or data
both economics and the data can help. The- correlations change, it is important to iden-
ory may suggest which variables agents aim tify empirical models that have reasonably
to control and on which ones they condition constant parameters, which remain inter-
their plans; and as parameter constancy in pretable when change occurs. As seen in
an empirical model is not invariant to nor- Section II above and Section IV below, re-
cursive estimation provides an incisive tool
for investigating parameter constancy, both
10With the lag operator L defined as Lxt = xt _ through
1, the sequence of estimated coeffi-
we let the difference operator A be (1 - L); hence,
Axt = xt- X- 1. More generally, Aqxt = (1- L)qXt. If
q (or r) is undefined, it is taken to be unity.
"1Alternatively, Stephen Nickell (1985) justifies er- 12This formulation is also discussed in Jean-Pierre
ror-correction mechanisms as arising from the optimal Florens and Michel Mouchart (1985a, b) and builds on
response of economic agents in some dynamic environ- Tjalling Koopmans (1950) and Ole Barndorff-Nielsen
ments. (1978).
cient values and via the associated Chow Chalmers, 1976), since an encompassing
statistics for constancy.'3 The Chow statis- model is a "sufficient representative" of
tics also play crucial roles for testing weak previous empirical findings.15
exogeneity indirectly through testing the These six criteria not only characterize
conjunction of hypotheses embodied in su- the conditions required to sustain reduc-
per exogeneity and for testing feedback ver- tions, they also correspond to concepts cen-
sus feedforward empirical models (cf. Engle tral to econometric analysis. Taking the
et al., 1983; Hendry, 1988). reductions in turn, a set of current-dated
variables can be eliminated (marginalized)
E. Data Admissibility without loss of information if those retained
correspond to sufficient statistics, and lagged
Many economic variables are inherently variables can be marginalized if they do not
positive, a model property ensured by the Granger-cause the remaining variables. Se-
use of logarithmic transformations of the quential conditioning generates innovation
data. However, if the resulting model does errors, and contemporaneous conditioning is
not correspond to the DGP, the cost of valid if the variables so treated are weakly
enforcing data admissibility may be a loss of exogenous for the parameters of interest.
parameter constancy. The concept of encompassing introduced
above determines the limits to model reduc-
F. Encompassing tion (i.e., the degree of parsimony feasible).
As shown in Hendry and Richard (1989),
This concept can be understood intu- parsimonious encompassing (in which the
itively as follows. Suppose model lpredicts 6 model accounts for the results of a
smaller
as the value for the parameter 0 in model 2, larger model within which it is nested) is
while model 2 actually has the estimate t. transitive, antisymmetric, and reflexive, thus
Model 1 encompasses model 2 if 6 is "stat- defining a partial ordering over models and
istically close" to 6, so that model 1 explains
thereby sustaining a progressive research
why model 2 obtains the results it does. To strategy.
the extent that model 1 accurately mimics
the DGP, it will encompass model 2. For IV. Econometric Modeling of Money
single equations estimated by least squares, Demand Using the Annual Data
a necessary condition for encompassing is
variance dominance, where one equation This section develops a conditional
variance-dominates another if the former econometric model of money demand on
has a smaller error variance.14 Thus, en- the annual data series for 1878-1970 only
compassing is more demanding than select- (T = 93), since the data from 1971 to 1975
ing models purely on the basis of their appear to have a different stochastic struc-
goodness of fit. It is also consistent with the ture [see (11) below]. We follow the proce-
concept of a progressive research strategy dures outlined in Section III, representing
(e.g., see Imre Lakatos, 1970; Alan the joint density of (mt,pt,it,RSt,RLt) in
terms of an autoregressive-distributed lag
model for mt conditional on (Pt,it,RSt,
13These tests of constancy are intimately related to
RLt) and a marginal model for (pt,
tests of forecast accuracy (cf. R. Brown, Durbin, and it,RSt,RLt). The conditional model is sim-
J. M. Evans, 1975; Hendry, 1979; Jan Kiviet, 1987). plified to an ECM and evaluated in light of
Jean-Marie Dufour (1982) elegantly summarizes recur-
sive techniques and their implications.
14Formally, variance dominance refers to the under-
lying (and unknown) error variances. Without loss of 15For comprehensive accounts of tests for encom-
clarity, we often will say a model variance-dominates passing and of related nonnested hypothesis tests,
another if the estimated residual variance of the former see Mizon and Richard (1986), Mizon (1984), James
is smaller than that of the latter. MacKinnon (1983), and Hashem Pesaran (1982).
the model design criteria.16 Phillips (1988) nants of money demand (reflecting poten-
and Phillips and Bruce Hansen (1990) tially different costs of adjustment and of
demonstrate that this error-correction ap- disequilibrium), yet via the error-correction
proach produces nearly optimal inferences mechanism ensures that long-run targets are
in cointegrated processes. Such a methodol- achieved (e.g., velocity and interest rates are
ogy of "learning from the data" while being cointegrated) (cf. Hendry et al., 1984). Eco-
guided by economic theory in the interpre- nomically, (8) is related to the theory of
tation of results contrasts with the approach money adjustment in Merton Miller and
adopted by Friedman and Schwartz of using Daniel Orr (1966) and Ross Milbourne
regression results to corroborate their eco- (1983), in which the short-run factors deter-
nomic theory. We conclude this section by mine money movements given the desired
considering two important issues for eco- bands, and the longer-run factors influence
nomic policy: the constancy of the money- the levels of the bands (e.g., see Gregor
demand function and the exogeneity of Smith, 1986). To be interpretable as a de-
money. mand equation, A1(1) < 0, A2(1) ? 0, A3(1) <
The economic framework of our empiri- 0, and A4(1) < 0; and for cointegration, A5 <
cal model is that of a log-linear long-run 0. However, the choice of parameterization
money-demand function: (e.g., in terms of lagged first differences or
higher-order differences) is arbitrary within
(7) (m-P) *=8o+ 8i - 2RS lag polynomials, so no sign restrictions on
individual polynomial coefficients can be
- 83(l +1p)
imposed a priori. Moreover, since monetary
where an asterisk denotes the long-run tar- theory does not yet specify how quickly
get value and p is the rate of inflation (cf. agents react to changes in real time, the
Friedman, 1956). Equation (7) parallels the orders of the Ai(L) must be data-based.
condition y = K + Z in Section III-A. Dy- Statistically, (8) is an ECM reparameteri-
namic adjustment is characterized by a con- zation of an autoregressive-distributed lag
tingent planning model of the form model of the variables in levels, as is (5) of
(4). A model such as (8) is of interest only if
(8) A(m - p), the regressors are weakly exogenous for the
resulting parameters, which in turn must be
= AO(L) A(m -P)- constant and invariant to historical changes
in the processes determining the marginal
+ A1(L)Apt + A2(L)Ait distributions. The error in (8) will be an
innovation if the plan efficiently incorpo-
+ A3(L)Arst + A4(L)Arlt rates available information and the model is
correctly specified. All of these issues are
+ A5[(m -P)t - -(m -P)t - d examined below in order to discriminate
between contingent planning and expecta-
+ et
tions interpretations of the empirical model.
where Ai(L) (i = 0,...,4) denotes a finite In Hendry and Ericsson (1983), we pre-
polynomial in the lag operator L and Et is sented an autoregressive-distributed lag
the deviation of the outcome from the plan. representation of money conditional on
This approach generalizes the conventional prices, incomes, and interest rates to estab-
partial-adjustment model, allowing separate lish the innovation error variance. That for-
reaction speeds to the different determi- mulation was simplified to an ECM like (8)
based on extant money-demand models for
the United Kingdom, with a static-equi-
16See Hendry and Mizon (1978) and McAleer et al. librium solution of the form v = - 80 +
(1985) on modeling from general to simple; see Leamer 82RS (thus taking v, and RS, to be cointe-
(1978) for an analysis of specification searches. grated) and an equation standard error of
TABLE 3-A GENERAL AUTOREGRESSIVE-DISTRIBU

CONDITIONAL ON INCOMES, PRICES, AND INTEREST RATES
lag i (or index)
Variable 0 1 2 3 4 5 i=0
Mt-l -1.0 1.316 -0.621 0.295 -0.091 0.00047 -0.100

(0.0) (0.157) (0.226) (0.242) (0.220) (0.111) (0.084)
Pt-i 0.447 -0.410 0.201 -0.176 0.083 - 0.050 0.095

(0.077) (0.120) (0.119) (0.116) (0.110) (0.074) (0.093)
1t-i 0.087 0.031 0.047 - 0.024 0.034 - 0.065 0.110

(0.072) (0.121) (0.090) (0.088) (0.091) (0.071) (0.078)
rst-i -0.019 0.014 - 0.00417 0.00510 - 0.00449 0.00407 - 0.00403

(0.00908) (0.016) (0.010) (0.011) (0.011) (0.00889) (0.013)
rl,-i -0.069 - 0.075 0.147 - 0.084 0.071 -0.00664 -0.016

(0.045) (0.070) (0.072) (0.075) (0.073) (0.052) (0.041)
Ut-_ 10.077 0.485 -1.821

(0.092) (0.204) (1.285)
Di 3.993 0.607 3.624

(1.220) (1.751) (1.024)
Constant -0.201
(0.137)
Notes: Values in parentheses are estimated standard errors; T = 1878-1970, R2 = 0.99987, Cr = 1.5535, dw = 1.94,
e2[11] = 9.62, 712[3,54] = 0.46, 714[4,49] = 1.39, e5[2] = 1.01, 716[39,17] = 0.23, -17[2,55] = 1.28. For readability, coeffi-
cients and estimated standard errors on D1, D2, and D3 are multiplied by 100.
1.71 percent of M.17 Those results stimu- neither of their models could encompass
lated further studies, including an improved certain features of the other, indicating that
specification on the same information set an improved specification might be possible.
in Andrew Longbottom and Sean Holly Continuing in a progressive research
(1985), a nonlinear reformulation in Alvaro strategy, we utilize their (and our previous)
Escribano (1985), and an extended informa- evidence, beginning with the static cointe-
tion set for 1875-1913 in Jan Klovland gration regression for vt and RSt:
(1987).18 Longbottom and Holly's and Es-
cribano's models were significant improve- (9) (m-i-p)t =-0.309-7.OORSt
ments on our 1983 model (i.e., each of their
models encompassed ours, but ours could (T = 1873-1970, R2 = 0.56, 5' = 10.86 per-
not encompass either of theirs). However, cent, dw = 0.33, ADF(1) = -2.77). (Adding
RL makes little difference, unsurprisingly if
RS and RL are cointegrated.) Exact signifi-
17Hendry and Ericsson (1985) investigate the hy-
cance levels are not available for David
pothesis that v, is a random walk and conclude that
there is little contrary evidence. Likewise, the data on Dickey and Wayne Fuller's (1979, 1981)
m, i, p, m - p, rs, rl, and p* all behave like I(1) series, augmented statistic ADF(1) or for Sargan
whereas their corresponding first differences behave and Alok Bhargava's (1983) Durbin-Wat-
like I(O) series. Nevertheless, equations (9) and (10) son-based statistic when testing for a unit
show that a more general cointegrating relationship
can be established.
root in the residuals, but the 10-percent
18Klovland (1987) constructs a new measure of the critical values in Engle and Granger (1987),
own interest rate on M. based on simulation, are respectively - 2.84
and 0.32 for 100 observations, leading to an ble and insignificant effects:
inconclusive outcome. Engle and Granger
(1987) show that these tests have low power
against highly dynamic stationary alterna-
(10) I\(m - P),
tives, so we cautiously proceed under the
assumption of cointegration. Economically,
= 0.45 A(m - p),
the solution to (9) is within the framework [0.06]
of (7) and implies that a one-percentage-
point increase in the short-term interest rate -0.10 A2(m - p)t-2
(e.g., from 5 percent to 6 percent) reduces [0.04]
M relative to PI by 7 percent in the long
run. -0.60 Apt + 0.39 Apt
Cointegration implies an error-correction [0.04] [0.05]
representation, so to model the behavior of
-0.021 A rst - 0.062 A 2rlt
money we estimate an unrestricted fifth-
[0.006] [0.021]
order autoregressive-distributed lag equa-
tion related to (4); it is shown in Table 3
-2.55( Ut l0.2)u-
together with relevant test statistics.19 The
[0.59]
model in Table 3 generalizes on (4) by inter
alia allowing the speed of adjustment to + 0.005 + 3.7(D1 + D3)A
vary with the extent of disequilibrium via [0.002] [0.6]
nonlinear error-correction terms. Following
Escribano (1985), Table 3 includes the
lagged level, square, and cube of Ut, the (T= 1878-1970, R2 =0.87, 06 =1.424 per-
residual from (9). Three zero-one dummies cent, -72[6, 78] = 1.56, -73[27, 57] = 0.50,
(D1, D2, and D3), which are unity for 74[4, 76] = 1.24, 65[2] = 1.6, 7}6[15,68] = 0.87,
1914-1918, 1921-1955, and 1939-1945, re- 77[2,82]= 0.21; see Table 2 for definitions
spectively, also appear. This regression sat- of the statistics). Values in brackets are
isfies all of the diagnostic checks reported heteroscedasticity-consistent estimated
and provides generally sensible estimates, standard errors (see White, 1980; Desmond
despite very high intercorrelations between Nicholls and Pagan, 1983; MacKinnon and
regressors. Most coefficients of lags beyond White, 1985).
three are negligible as well as insignificant, Concerning its economic interpretation,
and deleting those lags lowers cr. The en- (10) is similar in form and in numerical
tailed static solution is consistent with long- parameter values to several successful
run price and income elasticities of around money-demand models for the United
unity. Kingdom (cf. Hendry and Mizon, 1978;
The representation in Table 3 was simpli- Hendry, 1979; John Trundle, 1982; David-
fied to the error-correction model (10) using son, 1987; Keith Cuthbertson, 1988). Its co-
the approach described in Hendry (1983) of efficients satisfy the sign restrictions on the
first transforming the model to an inter- Ai(1) in (8) to be interpretable as a money-
pretable and near orthogonal specification demand function. The coefficients' sizes im-
paralleling (8) and then eliminating negligi- ply large immediate responses to changes in
inflation and interest rates but slow adjust-
ment subsequently to remaining disequilib-
ria, via the error-correction term. Inflation
enters as A pt + A2pt (approximately), which
19We have chosen five lags on the grounds that this
is a predictor of next period's inflation, opti-
implies 36 parameters estimated in Table 3 (vs. 93
observations total), in line with guidelines in Sargan
mal if prices vary quadratically. Thus, (10)
(1980b p. 880). Also, five lags implies a maximum lag of has a forward-looking interpretation, albeit
approximately one cycle. one based on data functions rather than on
0.16 A(m-p)t
-A(m-P)t
0.12 --- A(m-p)t
A
0.08
0.04 l / l
+ I
0--------- ------4-
0.04
0.08
0.121
1880 1890 1900 1910 1920 1930 1940 1950 1960 1970
Year
FIGURE 7. EQUATION (10): ACTUAL AND FITTED VALUES FOR A(M - P)t
models of the right-hand-side variables.20 significant (if regarded as test statistics) and
The coefficients correspond to nearly or- indicate design of a model congruent with
thogonal decision variables (with only two the information available. Even so, the range
of the 28 regressor intercorrelations exceed- of alternatives considered is sufficiently large
ing 0.5), consistent with (10) representing a to endow (10) with some credibility. From
contingent plan of agents who partition 2, the residuals are white noise and, from
available information into conceptually sep- 3, are also an innovation process against
arate entities. the information set in Table 3. There is no
The empirical parameterization in (10) ARCH, RESET, or heteroscedastic evi-
exhibits multiple equilibria, with two corre- dence of misspecification; the residuals are
sponding to the long-run solution (9) and a approximately normally distributed; and (10)
third being that solution shifted by 20 per- encompasses our earlier models and those
cent. In that sense, the results are consis- of Escribano and of Longbottom and Holly
tent with the use of an adjustment factor of (but not conversely).21 Figure 7 shows the
about that order by Friedman and Schwartz.
However, the dating of the disequilibria is
determined by the values of ut -1 and oper-21By appropriately (statistically) reducing the den-
ates over the entire sample, not just for the sity for our model, one could derive an estimate of the
period 1921-1955. parameters in Friedman and Schwartz's model. From
that constructed estimate and the estimate they actu-
Concerning the statistical attributes of
ally obtain, one could test whether our model (using
(10), the various diagnostic checks are in- annual data) encompasses theirs (using phase-average
data). Note also that it is easy to construct examples
for which the parameters in their model would not be
constant over time but those in ours would be [e.g., the
20Campos and Ericsson (1988) propose this inter- parameters in (10) are constant, p, is Granger-caused
pretation for similar inflation terms entering an equa- by m,, and the coefficients in that price equation
tion for consumers' expenditure in Venezuela. change over time]. Existence of the converse would
actual and fitted values for the rate of and 10 show the numerical values of two
change in real money over the sample pe- central coefficients, namely, those for Ap,
riod. Visual comparison of Figures 4 and 7 and the error-correction (- 0.2)u
(noting the logarithmic scale in both) high- together with plus-or-minus twice their se-
lights the better fit of the annual model: the quentially estimated standard errors, which
error variance of (10) is less than one-tenth provide an approximate 95-percent confi-
of that in (1). Although (10) has a rate of dence interval.23 Other than a minor fluc-
change as the dependent variable, it is an tuation directly following World War I, the
equation in log-levels because of the error- former varies by only a tiny fraction of its
correction term, as shown in (4)-(5) above, ex ante standard error; the latter is highly
so direct comparison of the two graphs is significant for the entire sample and also
valid. varies little relative to the estimated uncer-
The two remaining issues are constancy tainty. In both cases, the accrual of informa-
and exogeneity. Any claim to the constancy tion is apparent from the reduction in the
of a model for money demand would need width of the confidence interval over time.
both constant parameters and a similar Figure 11 graphs the one-step residuals for
goodness of fit over each of the epochs 1892-1913, using only 15 observations to
described above for Figure 1; in Section II, initialize estimation; not only is c constant,
we demonstrated that (1) has neither.22 To but its value throughout is very close to that
investigate the constancy of our model (10), for the full sample. In brief, our conditional
we adopt the recursive estimator, since the reformulation both fits well and is constant
one-step innovations allow the construction over the century to 1970, even though the
of sequences of constancy tests. Given the time-aggregated phase-average data reveal
world wars dummy (D1 + D3), 1915 is the the nonconstancy of (1) despite the atten-
earliest date for a continuous sequence un- dant loss of information.
til 1970, although a separate exercise is pos- Thus, we reach the issue of the exogene-
sible for the subsample 1878-1913 (see Fig. ity of money. Our analysis has taken every
11, below). Graphical presentation is effi- contemporaneous variable other than m as
cient for reporting the large volume of eval- if it were weakly exogenous (i.e., that it is
uation output: Figure 8 records the one-step valid to condition upon them for purposes
residuals and the corresponding calculated of statistical inference), interpreting the co-
equation standard errors (i.e., Y, - _Ix efficients of the resulting model as those of
and
0.0 + 2 in a standard notation). It is visu- a money-demand equation; by construction,
ally apparent that 6f has varied little over the (10) is invariant to whether Amt or A(m -
56-year test period. Further, none of the p)t is the regressand. The constancy of (10)
Chow statistics for the sequence 1915, reinforces our interpretation, which is also
1915-1916, 1915-1917, ... ., 1915-1970 or consistent with the institutional structure of
the sequence 1915-1970, 1916-1970, U.K. money markets in which the money
1917-1970, ..., 1969-1970, 1970 is signifi- stock appears to be endogenously deter-
cant at even the 5-percent level. Figures 9 mined by the decisions of the private sector
since the Bank of England in effect acts as a
lender of the first resort by standing ready
immediately refute any claim to encompass their find- to rediscount first-class bills at the going
ings. Bank Rate or Minimum Lending Rate (see
22A possible objection might be that a money- e.g., R. Hawtrey, 1938; Congdon, 1983;
demand equation's "constancy" need not be precise
Goodhart, 1984). Its constancy suggests that
but only relatively better than (say) the consumption
function's constancy. Since Friedman and Schwartz the conditioning variables may be super ex-
assert that their model is indeed constant, such an
objection is not germane. However, in response to
Mayer (1982 p. 1534), we note that the U.K. consump-
tion function was investigated in Hendry (1983) and 23Note, however, that the coefficients within u__1
was shown to be remarkably constant both before and are full-period estimates, as justified by the distribu-
after World War II. tional results in Engle and Granger (1987).
0.04 --residual
0_ 2 t
0.03 I
0.02
t -- -- - --- ---- -- - ---

0.01
0.01
0.02
0.03 _
0.04 I l l
1910 1920 1930 1940 1950 1960 1970
Year
FIGURE 8. EQUATION (10): ONE-STEP RESIDUALS AND THE CORRESPONDING CALCULATED EQUATION
STANDARD ERRORS
-0.3 coefficient estimate
--- coefficient estimate ? 2 standard errors

-0.4
-0.5 -
-0.6
-0.7
-0.8
-0.9
-1
1 V
-1.1 I l l
1910 1920 1930 1940 1950 1960 1970
Year
FIGURE 9. EQUATION (10): RECURSIVE ESTIMATION OF THE COEFFICIENT ON AP, AND ITS ESTIMATED
STANDARD ERROR
0 ______ coefficient estimate

- - - coefficient estimate ? 2 standard errors
-1_
-2
-3
-4 /\.-
-5
_ I
-6 1 1
1910 1920 1930 1940 1950 1960 1970
Year
FIGURE 10. EQUATION (10): RECURSIVE ESTIMATION OF THE ERROR-CORRECTION COEFFICIENT AND ITS
ESTIMATED STANDARD ERROR
0.04 ______ residual
0.03
0.02
0.01
oo t- ------- ;j :--------------- - - ----1 --- --------
0.01
0.02 -
0.03 -
0.04 I l l
1890 1895 1900 1905 1910 1915
Year
FIGURE 11. EQUATION (10) WITHOUT (D1 + D3)1: ONE-STEP RESIDUALS AND THE CORRESPONDING
CALCULATED EQUATION STANDARD ERRORS
ogenous for the demand parameters over over, the demand equation is constant de-
the sample (i.e., that the parameters of the spite the manifest nonconstancy of the in-
conditional model remain constant even strumenting equation (cf. Fig. 12), revealing
though the DGP of the conditioning vari- how informative this data set is and preclud-
ables changes over the sample). The exo- ing a forward-looking expectations interpre-
geneity of prices and endogeneity of money tation of (10) (cf. Hendry, 1988).25
are substantive issues, noting that it is com- Second, super exogeneity is not invariant
monplace in macroeconomics to determine to alternative factorizations of the joint den-
prices via the money-demand equation, tak- sity of money and prices when parameters
ing income, interest rates, and an observed in that density are not constant over time.
money supply as given and equating supply That is, if the model (10) were inverted to
to demand (cf. Robert Barro, 1987 pp. make Ap, the regressand, conditional on
128-31, 195-210). In particular, the empiri- Amt as an exogenous variable, then the
cal super exogeneity of prices (shown be- resulting equation should be nonconstant.
low) invalidates "inverting" the money-de- This is the case, as seen in Figure 13, which
mand equation to obtain prices. Engle and records the behavior of the estimated coef-
Hendry (1989) show that several implica- ficient of Amt in the inverted equation.
tions of super exogeneity are testable. Further, the equation standard error be-
First, direct tests of super exogeneity can comes 2.6 percent and, unlike (10), the
be constructed, but they require additional inverted equation cannot encompass the
observables, the relevance of which to (10) simple model that Ap* determines Apt
is that those observables should not have through a purchasing power parity condi-
been used in model design [e.g., none of the tion: ?73[1, 83] = 14.9.
variables dropped in simplifying from Table This evidence is inconsistent with the hy-
3 to (10) would give the tests power]. For pothesis that, over the period 1878-1970,
these data, the crucial issue is the weak exogenous money determined prices in the
exogeneity of Ap, since, if that is accepted,United Kingdom via a stable money-
(10) cannot sustain the interpretation of de- demand function, precisely because we
termining prices with money exogenous. The have established a constant money-demand
additional instruments we have tried are a model conditional on prices. In Hendry and
trend and current and lagged U.S. inflation Ericsson (1986), we propose an alternative
mechanism for money causing inflation:
(Ap* and Ap* 1), using a recursive instru-
mental variables estimator to investigate namely, the long-run effects of deviations
constancy and exogeneity conjointly (see from purchasing-power parity.
Hendry and Adrian Neale, 1987). On endo- The final test of super exogeneity is that
genizing Apt, 6^ remains virtually un- the parameters are invariant to regime
changed at 1.44 percent, and Sargan's (1958) changes: here, we exploit the joint introduc-
statistic for testing the validity of the instru- tion of floating exchange rates and Compe-
ments yields 08[2] = 3.62, justifying the
choice of instruments and being consistent
with the weak exogeneity of Apt.24 More-
Ape, Ars, and A2rl, by recursive multivariate least
squares reveals nonconstancy in each. If there were
simultaneity bias in (10), the parameter estimates in
24We also investigated the assumed joint weak exo- (10) would change as those in the reduced-form equa-
geneity of prices and short- and long-term interest tions did; thus, the constancy of (10) in spite of non-
rates in (10). (The weak exogeneity of income is not at constant reduced-form equations implies the weak exo-
issue because it appears only at a lag, via the error-cor-
geneity of prices and interest rates. See Hendry, Neale,
rection term.) When estimated with A/ P*, A p,* I1, rs, tand
1, Srba (1988) on recursive multivariate least squares,
rst2, rl,t_1, rlt_2, and a trend as instruments and and see Jean Bronfenbrenner (1953) on the relation-
treating A(m-p)t, Apt, Arst, and A2rlt as endoge- ship between simultaneity bias and reduced-form pa-
nous, 6' increases only slightly to 1.56 percent, the rameters.
coefficients remain virtually unchanged from (10), and 25See Engle (1984) for a survey of exogeneity tests
Sargan's (1958) statistic is 68[4]=4.96 (insignificant).and Kiviet (1985, 1987) for finite-sample evidence on
Further, estimation of the reduced-form equations for their performance.
0.1 residual
0.08
0.06 -
0.04
0.02
0.02 006 L-----------

0.04
NEN
0.06 -
1910 1920 1930 1940 1950 1960 1970

Year
FIGURE 12. ONE-STEP RESIDUALS FROM THE INSTRUMENTING EQUATION FOR AP, USING AP*
2.5 coefficient estimate
--coefficient estimate ? 2 standard errors
J -- ----_.._
1.5I
1 __
0.5
L 1 -----------------------------
-I/
o'-
+ _/
0- \7 ----------- ^- - .
0.5 I I I
1910 1920 1930 1940 1950 1960 1970
Year
FIGURE 13. RECURSIVE ESTIMATION OF THE COEFFICIENT ON AMt IN THE INVERTED MODEL FOR Apt
USING (10)
tition and Credit Control regulations in and zero otherwise, and where ut,1 is still
1971. Narrow money demand (U.K. M1 calculated from (9). Now cr and most of the
measure) apparently was not perturbed by coefficients other than those involving D4
these switches, but most investigators have are virtually unaltered, consistent with Lu-
found major parameter changes in mod- brano et al.'s results.
els of broad money measures (e.g., U.K. A potential explanation for this finding,
fM3; see Hendry and Mizon [1978] and consistent with the earlier evidence and
Michel Lubrano, Richard Pierse, and economic analysis, is presaged by Klovland's
Richard [1986] for overviews). A Chow test (1987) result for pre-1914 data that the own
of (10) using predictions over 1971-1975 interest rate on broad money is an impor-
yields q7l[5, 84]= 19.5, and cr rises totant
2.03omitted variable from the present in-
percent, rejecting constancy and super exo- formation set.26 Over much of the sample,
geneity. The growth rate of nominal money U.K. commercial banks acted like a cartel
over this test period substantially exceeds with administered (and generally low) de-
that of any previous episode (including both posit interest rates; this situation changed
world wars) and could reflect a disequilib- after 1970 due to the competition regula-
rium adjustment to the removal of the tions. Thus, own interest rates rose rapidly,
chronic cartelization of the U.K. banking altering the historical differentials and in-
system (the motivation for the change in ducing predictive failure in models that ex-
banking regulations). Indeed, Lubrano et al. cluded that variable. We plan to extend
(1986) find that the long-run relationship of Klovland's data set to test this conjecture
money demand to prices and incomes holds and to continue the progressive research
after 1970 but that the short-run relation- strategy by encompassing previous models.
ship is severely perturbed, with money de-
mand increasing as competitive interest V. Conclusions
rates increase. With that structure in mind,
we obtained This paper focuses on the evaluation of
Friedman and Schwartz's (1982) empirical
model for U.K. money demand and the
(11) A\(m - P)t design of an improved specification using
- 0.47A(m - P)t their annual data. At the heart of model
[0.06] evaluation are the issues of model credibil-
ity and validity and the role of corroborat-
-0.11 A2(m - P)t2 - 0.59'Ap, ing evidence. The failure by Friedman and
[0.04] [0.04]
Schwartz to present statistical evidence per-
tinent to their main
+ 0.41 Ap-1 - 0.017A rs, claims about the United
[0.04] [0.006] Kingdom leaves those claims lacking in
credibility. The presence of substantial mis-
- 0.078 A 2rl,
specification invalidates many of their infer-
[0.019]
- 1.15(uGt_ -0.2)ut2_ 1 + 0.007

[0.19] [0.0021
26In Hendry and Ericsson (1983 pp. 77-8), we ex-
perimented with RS replaced by RN, Friedman and
+ 3.4(D1 + D3) + 0.071(D4)t
Schwartz's (1982 p. 270) measure of the marginal cost
[0.6] [0.010]
of money [RN RS (H/M)]. The resulting estimates
+ 0.090(D4)t * Arst are very similar to those found using RS and reveal no
improvement in the constancy of the interest-rate co-
[0.020]
efficient over 1971-1975. Following suggestions by
Chris Pissarides and by Ross Starr (1982), we also
(T -1878-1975, R2 =0.88, a = 1.478 per- experimented with adding variables that measured in-
terest-rate volatility (e.g., using 0.2E =0[RSt_- RS+]
cent, -q2[6,81] = 1.19, -q4[2,83] = 1.31, 65[2] =
where RS+ = 0.2E4= RSt_,). The effect was largest for
2.9, -q6[18,68] = 0.43), where D4 is a dummy the most recent period but did not produce constant
which is unity over the period 1971-1975 parameters or a constant fit over 1971-1975.
ences from equations based on the phase- posed above is not the end of the story
average data (cf. Table 1). In particular, since, for example, parameter nonconstancy
their final money-demand equation is not is evident during 1971-1975. That it is not
constant, contrary to their claim; and, on perfect is less than surprising, as the data
testing assumptions such as price homo- span a century during which financial insti-
geneity and the absence of trends, rejection tutions altered dramatically: witness the
results. Such negative findings are consis- growth of building societies and, after 1970,
tent with those reported by Meghnad Desai the introduction of Competition and Credit
(1981 [especially Ch. 4]). The procedure of Control regulations and of floating ex-
averaging data over business-cycle phases change rates. Even so, the evidence suggests
did not notably reduce the serial correlation that substantial benefits are available in
in the data series but did lose information, practice from a progressive research strat-
leading to rather badly fitting equations. As egy exploiting tests in both model design
an alternative, we recommend analyzing the and model evaluation.
annual data and modeling "trend" and
"cycle" jointly. DATA APPENDIX
Corroborating a subset of the implica-
tions of a theory is not by itself an adequate Legend
justification for deeming the theory useful
(see e.g., Friedman, 1953 pp. 8-9; Karl Pop- D, A dummy variable for World War I (= 1
for 1914-1918 inclusive, 0 elsewhere)
per, 1959 section 82; Lawrence Boland, 1982
D2 A dummy variable for 1921-1955, parallel-
Ch. 1). That is illustrated by the contrast ing S for phase-average data (= 1 for
between Friedman and Schwartz's claims to 1921-1955 inclusive, 0 elsewhere)
have empirically corroborated various as- D3 A dummy variable for World War II (= 1
for 1939-1945 inclusive, 0 elsewhere)
pects of their theories and our evidence that
D4 A dummy variable for Competition and
those claims are actually refutable from the Credit Control regulations (= 1 for
same data. Only well-tested theories that 1971-1975, 0 elsewhere)
have successfully weathered tests outside G(jp + i) Growth rate of phase-average nominal in-
come (fraction)
the control of their proponents and that
H High-powered money (million ?)
encompass the gestalt of existing empirical I Real net national product (million 1929 ?)
evidence seem likely to provide a useful M Money stock (million ?)
basis for applied economic analysis and pol- N Population (millions)
icy. P Deflator of I (1929 = 1.00)
P* Deflator of net national product in the
The tests used to evaluate Friedman and
United States (1929 = 1.00)
Schwartz's models indicate that scope exists RL Long-term interest rate (fraction)
for model development but do not indicate RN RS (H/M)
what changes are required. We attempted RS Short-term interest rate (fraction)
S A dummy variable for phase observations
an improved specification and presented an
16-28 (1921-1955; = 1 for observations
econometric model of the demand for 16-28 inclusive, 0 elsewhere)
money in the United Kingdom during W A dummy variable for phase observations
1878-1970 which incorporates economic 13-15 and 26-28 (1918-1921 and
theory, satisfies a wide range of statistical 1946-1955; = -4, -3, -2, 8, 5, and 3
for phase observations 13, 14, 15, 26, 27,
criteria, and encompasses existing models
and 28, respectively; 0 elsewhere)
based on the same data set. Those features
are essential for a model to characterize the
Coefficients and estimated standard errors of the dum-
underlying data generation process ade- mies D1, D2, D3, (but not D4), S, and W are reported
quately. On a substantive level, the empiri- times 100 for readability.
cal constancy of our model is consistent The data are as in Friedman and Schwartz (1982
with a structure in which nominal money is tables 4.8, 4.9), but relevant series are rescaled propor-
tionately from 1871 to 1920 to remove the break in
endogenously determined by demand fac- 1920 when southern Ireland ceased to be part of the
tors, conditional on prices, incomes, and United Kingdom. Also, P, P*, G(p + i), RS, and RL
interest rates. Undoubtedly, the model pro- have been divided by 100 so that the values of P and
P* in 1929 equal 1.00 (rather than 100) and the Boland, Lawrence A., The Foundations of Eco-
interest rates and G(p + i) are expressed as fractions nomic Method, London: Allen and Un-
(rather than as percentages).
win, 1982.
Box, George E. P. and Pierce, David A., "Distri-
Data Measurement
bution of Residual Autocorrelations in
We record several important caveats about Fried- Autoregressive-Integrated Moving Aver-
man and Schwartz's data, over'apping those which they age Time Series Models," Journal of the
carefully document. American Statistical Association, Decem-
ber 1970, 65, 1509-26.
(i) The choice of monetary measure seems too broad
to represent transactions demand, yet too narrow Bronfenbrenner, Jean, "Sources and Size of
for an overall index of "liquidity." M (above) is Least-Squares Bias in a Two-Equation
based on the U.K. monetary measure M2 (pp. Model," in William C. Hood and Tjalling
111-4) and hence excludes interest-bearing liabili-
C. Koopmans, eds., Studies in Economet-
ties of building societies but includes interest-
ric Method, New Haven: Yale University
bearing bank deposits. However, the building soci-
ety movement has grown rapidly over the last Press, 1953, 221-35.
century to become roughly equal in size to the Brown, R. L., Durbin, James and Evans, J. M.,
whole commercial banking sector, and building "Techniques for Testing the Constancy of
society liabilities are among the most liquid assets
Regression Relationships over Time,"
available to the personal sector. Note that money-
stock figures are centered on mid-years by averag- Journal of the Royal Statistical Society,
ing successive end-of-year values. Series B, 1975, 37 (2), 149-92.
(ii) The measurement of the price series (P) is ad- Burns, Arthur F. and Mitchell, Wesley C., Mea-
justed for rationing and controls (pp. 115-20), suring Business Cycles, New York: Na-
with real income then derived by deflating nomi-
nal income. However, it is unreasonable to hold
tional Bureau of Economic Research,
measured nominal income constant when mea- 1946.
sured prices are believed to be incorrect. Further- Campos, Julia and Ericsson, Neil R., "Econo-
more, over a century, one must be concerned metric Modeling of Consumers' Expendi-
about the effects on the measurement of P of the
ture in Venezuela," International Fi-
many dramatic changes that have occurred in
quality-adjusted relative prices. nance Discussion Paper No. 325, Board
(iii) Friedman and Schwartz emphasize an "errors- of Governors of the Federal Reserve Sys-
in-variables" paradigm and claim that the impor- tem, Washington, DC, June 1988.
tance of errors in variables is reduced by phase- ____ ____ and Hendry, David F., "An
averaging but enhanced by differencing (p. 86). All
Analogue Model of Phase-Averaging Pro-
the variables undoubtedly contain substantial mea-
surement errors, especially when interpreted as cedures," Journal of Econometrics, March
correspondences to economically meaningful la- 1990, 43, 275-92.
tent constructs, but time series of over a century Chalmers, Alan F., What Is This Thing Called
will also contain systematic errors. Differencing
Science? An Assessment of the Nature and
would remove most effects of such errors, whereas
averaging is of little help if errors are persistent Status of Science and Its Methods, St. Lu-
(e.g., highly autoregressive). cia, Queensland: University of Queens-
land Press, 1976.
Resolution of these difficulties is outside the scope of
Chow, Gregory C., "Tests of Equality between
this paper, but they must influence the interpretation
of the empirical evidence.
Sets of Coefficients in Two Linear Re-
gressions," Econometrica, July 1960, 28,
591-605.
REFERENCES Congdon, Tim, "Has Friedman Got It
Wrong?" The Banker, July 1983, 117-25.
Artis, Michael J., "Book Review," Econom- Cuthbertson, Keith, "The Demand for M1: A
ica, May 1984, 51, 205-7. Forward Looking Buffer Stock Model,"
Barndorff-Nielsen, Ole, Information and Ex- Oxford Economic Papers, March 1988, 40,
ponential Families in Statistical Theory, 110-31.
Chichester: Wiley, 1978. Davidson, James E. H., "Disequilibrium Mon-
Barro, Robert J., Macroeconomics, 2nd ed., ey: Some Further Results with a Mone-
New York: Wiley, 1987. tary Model of the UK," in C. A. E. Good-
hart, D. Currie, and D. T. Llewellyn, eds., University of California, San Diego, Au-
The Operation and Regulation of Financial gust 1987.
Markets, London: Macmillan, 1987, and Granger, Clive W. J., "Co-Integra-
125-49. tion and Error Correction: Representa-
, Hendry, David F., Srba, Frank and Yeo, tion, Estimation, and Testing," Econo-
Stephen, "Econometric Modelling of the metrica, March 1987, 55, 251-76.
Aggregate Time-Series Relationship be- _ and Hendry, David F., "Testing Super
tween Consumers' Expenditure and In- Exogeneity and Invariance," Discussion
come in the United Kingdom," Economic Paper No. 89-51, University of California,
Journal, December 1978, 88, 661-92. San Diego, November 1989.
Desai, Meghnad, Testing Monetarism, Lon- _ , and Richard, Jean-Fransois,
don: Frances Pinter, 1981. "Exogeneity," Econometrica, March 1983,
Dickey, David A. and Fuller, Wayne A., "Distri- 51, 277-304.
bution of the Estimators for Autoregres- Ericsson, Neil R. and Hendry, David F., "Con-
sive Time Series with a Unit Root," Jour- ditional Econometric Modeling: An Ap-
nal of the American Statistical Association, plication to New House Prices in the
June 1979, 74, 427-31. United Kingdom," in A. C. Atkinson and
and , "Likelihood Ratio S. E. Fienberg, eds., A Celebration of
Statistics for Autoregressive Time Series Statistics: The ISI Centenary Volume, New
with a Unit Root," Econometrica, July York: Springer-Verlag, 1985, 251-85.
1981, 49, 1057-72. Escribano, Alvaro, "Non-Linear Error-Cor-
Dolado, Juan J. and Jenkinson, Tim, "Coin- rection: The Case of Money Demand in
tegration: A Survey of Recent Develop- the U.K. (1878-1970)," mimeo, Univer-
ments," Applied Economics Discussion sity of California, San Diego, December
Paper No. 39, Institute of Economics and 1985.
Statistics, University of Oxford, Novem- Fisher, R. A., "The Goodness of Fit of Re-
ber 1987. gression Formulae, and the Distribution
Dufour, Jean-Marie, "Recursive Stability of Regression Coefficients," Journal of
Analysis of Linear Regression Relation- the Royal Statistical Society, 1922, 85 (4),
ships: An Exploratory Methodology," 597-612.
Journal of Econometrics, May 1982, 19, Florens, Jean-Pierre and Mouchart, Michel,
31-76. (1985a) "Conditioning in Dynamic Mod-
Durbin, James and Watson, Geoffrey S., "Test- els," Journal of Time Series Analysis, 1985,
ing for Serial Correlation in Least Squares 6 (1), 15-34.
Regression. I," Biometrika, December _ and , (1985b) "A Linear The-
1950, 37, 409-28. ory for Noncausality," Econometrica, Jan-
and , "Testing for Serial Cor- uary 1985, 53, 157-75.
relation in Least Squares Regression. II," Friedman, Milton, Essays in Positive Eco-
Biometrika, June 1951, 38, 159-78. nomics, Chicago: University of Chicago
Engle, Robert F., "Autoregressive Condi- Press, 1953.
tional Heteroscedasticity with Estimates 9 "The Quantity Theory of
of the Variance of United Kingdom In- Money-A Restatement," in M. Fried-
flation," Econometrica, July 1982, 50, man, ed., Studies in the Quantity Theory
987-1007. of Money, Chicago: University of Chicago
,"Wald, Likelihood Ratio, and La- Press, 1956, 3-21.
grange Multiplier Tests in Econometrics," and Schwartz, Anna J., Monetary
in Z. Griliches and M. D. Intriligator, Trends in the United States and the United
eds., Handbook of Econometrics, Vol. 2, Kingdom: Their Relation to Income, Prices,
Amsterdam: North-Holland, 1984, 775- and Interest Rates, 1867-1975, Chicago:
826. University of Chicago Press, 1982.
9 "On the Theory of Cointegrated Gilbert, Christopher L., "Professor Hendry's
Economic Time Series, 1987," mimeo, Econometric Methodology," Oxford Bul-
letin of Economics and Statistics, August Statistics and Nuffield College, University
1986, 48, 283-307. of Oxford, 1989.
Godfrey, Leslie G., "Testing Against General _ and Ericsson, Neil R., "Assertion
Autoregressive and Moving Average Er- without Empirical Basis: An Econometric
ror Models when the Regressors Include Appraisal of 'Monetary Trends in... the
Lagged Dependent Variables," Econo- United Kingdom' by Milton Friedman and
metrica, November 1978, 46, 1293-1301. Anna Schwartz," in Bank of England
Goodhart, Charles A. E., "Monetary Trends in Panel of Academic Consultants, Mone-
the United States and the United Kingdom: tary Trends in the United Kingdom, Panel
A British Review," Journal of Economic Paper No. 22, October 1983, 45-101 (with
Literature, December 1982, 20, 1540-51. additional references).
, Monetary Theory and Practice: The and , "Assertion without Em-
UK Experience, London: Macmillan, 1984. pirical Basis: An Econometric Appraisal
Granger, Clive W. J., "Some Properties of of Monetary Trends in... the United King-
Time Series Data and Their Use in dom by Milton Friedman and Anna J.
Econometric Model Specification," Jour- Schwartz," International Finance Discus-
nal of Econometrics, May 1981, 16, sion Paper No. 270, Board of Governors
121-30. of the Federal Reserve System, Washing-
Hallman, Jeff, "Attractor Sets: A Nonlinear ton, DC, December 1985.
Extension of the Cointegration Concept," _ and , "Prolegomenon to a
mimeo, University of California, San Reconstruction: Further Econometric
Diego, November 1987. Appraisal of Monetary Trends in ... the
Harvey, Andrew C., The Econometric Analysis United Kingdom by Milton Friedman and
of Time Series, Oxford: Philip Allan, 1981. Anna J. Schwartz," mimeo, Board of
Hawtrey, R. G., A Century of Bank Rate, Governors of the Federal Reserve Sys-
London: Longmans, Green, 1938. tem, Washington, DC, March 1986.
Hendry, David F., "Predictive Failure and and Mizon, Grayham E., "Serial Cor-
Econometric Modelling in Macroeco- relation as a Convenient Simplification,
nomics: The Transactions Demand for Not a Nuisance: A Comment on a Study
Money," in P. Ormerod, ed., Economic of the Demand for Money by the Bank of
Modelling, London: Heinemann, 1979, England," Economic Journal, September
217-42. 1978, 88, 549-63.
, "Econometric Modelling: The and , "Procrustean Econo-
'Consumption Function' in Retrospect," metrics: Or Stretching and Squeezing
Scottish Journal of Political Economy, Data," in Clive W. J. Granger, ed.,
November 1983, 30, 193-220. Modelling Economic Series: Readings
, ed., Economic Modelling with Coin- in Econometric Methodology, Oxford:
tegrated Variables (special issue), Oxford Clarendon Press, 1990, 121-36.
Bulletin of Economics and Statistics, Au- _ and Neale, Adrian J., "Monte Carlo
gust 1986, 48. Experimentation Using PC-NAIVE," in
,_ "Econometric Methodology: A Per- G. F. Rhodes, Jr., ed., Advances in
sonal Perspective," in T. F. Bewley, ed., Econometrics, Vol. 6, Greenwich, CT: JAI
Advances in Econometrics, Vol. 2, Cam- Press, 1987, 91-125.
bridge: Cambridge University Press, 1987, , and Srba, Frank, "Econo-
29-48. metric Analysis of Small Linear Systems
, "The Encompassing Implications of Using PC-FIML," Journal of Economet-
Feedback versus Feedforward Mecha- rics, May/June 1988, 38, 203-26.
nisms in Econometrics," Oxford Eco- , Pagan, Adrian R. and Sargan, J. Denis,
nomic Papers, March 1988, 40, 132-49. "Dynamic Specification," in Z. Griliches
, PC-GIVE: An Interactive Econo- and M. D. Intriligator, eds., Handbook of
metric Modelling System, Version 6.0/ Econometrics, Vol. 2, Amsterdam: North-
6.01, Oxford: Institute of Economics and Holland, 1984, 1023-1100.
and Richard, Jean-Fransois, "On the Methodology of Scientific Research Pro-

Formulation of Empirical Models in Dy- grammes," in I. Lakatos and A. Mus-
namic Econometrics," Journal of Econo- grave, eds., Criticism and the Growth of
metrics, October 1982, 20, 3-33. Knowledge, Cambridge: Cambridge Uni-
and , "The Econometric versity Press, 1970, 91-196.
Analysis of Economic Time Series," In- Leamer, Edward E., Specification Searches: Ad
ternational Statistical Review, August 1983, Hoc Inference with Nonexperimental Data,
51, 111-63. New York: Wiley, 1978.
and , "Recent Developments ,_ "Let's Take the Con Out of Econo-
in the Theory of Encompassing," in metrics," American Economic Review,
Bernard Cornet and Henry Tulkens, eds., March 1983, 73, 31-43.
Contributions to Operations Research and Longbottom, Andrew and Holly, Sean, "Econo-
Economics: The Twentieth Anniversary of metric Methodology and Monetarism:
CORE, Cambridge, MA: MIT Press, 1989, Professor Friedman and Professor Hendry
393-440. on the Demand for Money," Discussion
and Wallis, Kenneth F., eds., Econo- Paper No. 131, London Business School,
metrics and Quantitative Economics, Ox- February 1985.
ford: Blackwell, 1984. Lubrano, Michel, Pierse, Richard G. and
Herschel, J., A Preliminary Discourse on the Richard, Jean-Francois, "Stability of a U.K.
Study of Natural Philosophy, London: Money Demand Equation: A Bayesian
Longman, Rees, Brown & Green and Approach to Testing Exogeneity," Review
John Taylor, 1830. of Economic Studies, August 1986, 53,
Jarque, Carlos M. and Bera, Anil K., "Efficient 603-34.
Tests for Normality, Homoscedasticity MacKinnon, James G., "Model Specification
and Serial Independence of Regression Tests Against Non-Nested Alternatives,"
Residuals," Economics Letters, 1980, 6 Econometric Reviews, 1983, 2 (1), 85-158.
(3), 255-9. _ and White, Halbert, "Some Hetero-
Johnston, J., Econometric Methods, New skedasticity-Consistent Covariance Matrix
York: McGraw-Hill, 1963. Estimators with Improved Finite Sample
Judd, John P. and Scadding, John L., "The Properties," Journal of Econometrics,
Search for a Stable Money Demand September 1985, 29, 305-25.
Function: A Survey of the Post-1973 Lit- Mayer, Thomas, "Monetary Trends in the
erature," Journal of Economic Literature, United States and the United Kingdom: A
September 1982, 20, 993-1023. Review Article," Journal of Economic
Kiviet, Jan F., "Model Selection Test Proce- Literature, December 1982, 20, 1528-39.
dures in a Single Linear Equation of a McAleer, Michael, Pagan, Adrian R. and Volker,
Dynamic Simultaneous System and Their Paul A., "What Will Take the Con Out of
Defects in Small Samples," Journal of Econometrics?" American Economic Re-
Econometrics, June 1985, 28, 327-62. view, June 1985, 75, 293-307.
, Testing Linear Econometric Models, Milbourne, Ross, "Optimal Money Holding
Amsterdam: Amsterdam University Press, Under Uncertainty," International Eco-
1987. nomic Review," October 1983, 24, 685-98.
Klovland, Jan T., "The Demand for Money Miller, Merton H. and Orr, Daniel, "A Model
in the United Kingdom, 1875-1913," Ox- of the Demand for Money by Firms,"
ford Bulletin of Economics and Statistics, Quarterly Journal of Economics, August
August 1987, 49, 251-71. 1966, 80, 413-35.
Koopmans, Tjalling C., "When Is an Equation Mizon, Grayham E., "The Encompassing Ap-
System Complete for Statistical Pur- proach in Econometrics," in D. F. Hendry
poses?" in Tjalling C. Koopmans, ed., and K. F. Wallis, eds., Econometrics and
Statistical Inference in Dynamic Economic Quantitative Economics, Oxford: Black-
Models, New York: Wiley, 1950, 393-409. well, 1984, 135-72.
Lakatos, Imre, "Falsification and the _ and Richard, Jean-Francois, "The En-
compassing Principle and Its Application Methodology," in P. E. Hart, G. Mills

to Testing Non-Nested Hypotheses," and J. K. Whitaker, eds., Econometric
Econometrica, May 1986, 54, 657-78. Analysis for National Economic Planning,
Moore, Basil J., "Monetary Trends in the Colston Papers, Vol. 16, London: Butter-
United States and in the United Kingdom, worths, 1964, 25-63; reprinted in D. F.
A Review," The Financial Review, May Hendry and K. F. Wallis, eds., Economet-
1983, 18, 146-66. rics and Quantitative Economics, Oxford:
Nicholls, Desmond F. and Pagan, Adrian R., Blackwell, 1984, 275-314.
"Heteroscedasticity in Models with , (1980a) "Some Approximations to
Lagged Dependent Variables," Econo- the Distribution of Econometric Criteria
metrica, July 1983, 51, 1233-42. which Are Asymptotically Distributed as
Nickell, Stephen, "Error Correction, Partial Chi-Squared," Econometrica, July 1980,
Adjustment and All That: An Expository 48, 1107-38.
Note," Oxford Bulletin of Economics and , (1980b) "Some Tests of Dynamic
Statistics, May 1985, 47, 119-29. Specification for a Single Equation,"
Patterson, Kerry D., "The Stability of Some Econometrica, May 1980, 48, 879-97.
Annual Consumption Functions," Oxford _ and Bhargava, Alok, "Testing Residu-
Economic Papers, March 1986, 38, 1-30. als from Least Squares Regression for
Pesaran, M. Hashem, "Comparison of Local Being Generated by the Gaussian Ran-
Power of Alternative Tests of Non-Nested dom Walk," Econometrica, January 1983,
Regression Models," Econometrica, Sep- 51, 153-74.
tember 1982, 50, 1287-1305. Sims, Christopher A., "Macroeconomics and
Phillips, Peter C. B., "Time Series Regression Reality," Econometrica, January 1980, 48,
with a Unit Root," Econometrica, March 1-48.
1987, 55, 277-301. Smith, Gregor W., "A Dynamic Baumol-Tobin
_ "Reflections on Econometric Model of Money Demand," Review of
Methodology," Economic Record, De- Economic Studies, July 1986, 53, 465-9.
cember 1988, 64, 344-59. Spanos, Aris, Statistical Foundations of
and Hansen, Bruce E., "Statistical In- Econometric Modelling, Cambridge: Cam-
ference in Instrumental Variables Re- bridge University Press, 1986.
gression with I(1) Processes," Review of Starr, Ross M., "Variation in the Maturity
Economic Studies, January 1990, 57, Composition of Debt and Behavior of the
99-125. Monetary Aggregates: The Maturity Shift
Popper, Karl R., The Logic of Scientific Dis- Hypothesis," Economic Review (San
covery, London: Hutchinson, 1959. Francisco), conference supplement, Fall
Ramsey, J. B., "Tests for Specification Errors 1982, 202-33.
in Classical Linear Least-Squares Regres- Stock, James H., "Asymptotic Properties of
sion Analysis," Journal of the Royal Sta- Least Squares Estimators of Cointegrat-
tistical Society, Series B, 1969, 31 (2), ing Vectors," Econometrica, September
350-71. 1987, 55, 1035-56.
Salmon, Mark "Error Correction Mecha- Trundle, John M., "The Demand for Ml in
nisms," Economic Journal, September the UK," mimeo, Bank of England, 1982.
1982, 92, 615-29. White, Halbert, "A Heteroskedasticity-Con-
Sargan, J. Denis, "The Estimation of Eco- sistent Covariance Matrix Estimator and
nomic Relationships Using Instrumental a Direct Test for Heteroskedasticity,"
Variables," Econometrica, October 1958, Econometrica, May 1980, 48, 817-38.
26, 393-415. , Specification Analysis in Economet-
, "Wages and Prices in the United rics, manuscript, 1988; Cambridge: Cam-
Kingdom: A Study in Econometric bridge University Press, forthcoming.

American Economic Association The American Economic Review

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

American Economic Association The American Economic Review

Încărcat de

Drepturi de autor:

Formate disponibile

American Economic Association

By DAVID F. HENDRY AND NEIL R. ERICSSON *

This paper evaluates an empirical model of UK money demand developed by

(a) ( n-p-=n) = 0.16 + 0.88(i - n) - 11.16RN- 0.22G(p + i) + 1.4W + 215 p. 282

(a = 5.54, R2 = 0.970, dw = 1.51, flJ[18, 12] = 6.3)

(b) (p + i) = 0.38 + 1.01m + 14.17RN+ 0.53G(p + i) - 1.1W- 19S p. 349

(a = 5.99, R2 = 0.9977, dw = 1.33, dJ[l8, 12] = 5.7)

(c) p = -5.99 - 0.015i + 1.02m + 0.94RS- 1.10G(p + i) p. 420

(a = 6.0, dw = 1.01, lJ[l,5 10] = 3.1)

(d) i = 6.50 + 0.017- - 0.05m + 2.34R3+ 2.20G(p + i) p. 420

(v = 5.9, dw = 0.97, lJ[l,5 10] = 2.1)

We could closely replicate equation (a): conditional on treating it as provisionally

the interest rate coefficient from phases is exogenous. Treating M as endogenous

~_ ---coefficient estimate ?2 standard errors

1.0 Annual data (vt)

1880 1900 1920 1940 1960 1980

implicit) sequence of reductions and trans- exogeneity, D) parameter constancy, E) data

TABLE 2-CRITERIA FOR EVALUATING AND DESIGNING ECONOMETRIC MODELS

B first-order residual autocorrelation dw James Durbin and Geoffrey Watson

B q invalid parameter restrictions ?73[q, T - k - q] J. Johnston (1963 p. 126)

B qth-order ARCH 64[q]; ?74[q, T - k - 2q] Engle (1982)

B skewness and excess kurtosis 65[2] Jarque and Bera (1980)

B heteroscedasticity quadratic in 716[q, T - k - q - 1] White (1980), Desmond Nicholls and

B qth-order RESET ?77[q, T - k - q] J. B. Ramsey (1969)

C q instrumental variables not 8[q- k]; J. Denis Sargan (1958, 1964);

D predictive failure over a subset Y71[q, T - k - q] Chow (1960 pp. 594-5)

In the context of discovery, satisfying a long-run relationship between variables with

(5) Ayt=a+f3Azt+y(yt-i-Zt-l)+vt C. Weak Exogeneity

TABLE 3-A GENERAL AUTOREGRESSIVE-DISTRIBU

lag i (or index)

Mt-l -1.0 1.316 -0.621 0.295 -0.091 0.00047 -0.100

Pt-i 0.447 -0.410 0.201 -0.176 0.083 - 0.050 0.095

1t-i 0.087 0.031 0.047 - 0.024 0.034 - 0.065 0.110

rst-i -0.019 0.014 - 0.00417 0.00510 - 0.00449 0.00407 - 0.00403

rl,-i -0.069 - 0.075 0.147 - 0.084 0.071 -0.00664 -0.016

Ut-_ 10.077 0.485 -1.821

Di 3.993 0.607 3.624

t -- -- - --- ---- -- - ---

-0.3 coefficient estimate

--- coefficient estimate ? 2 standard errors

0 ______ coefficient estimate

0.04 ______ residual

oo t- ------- ;j :--------------- - - ----1 --- --------

0.02 006 L-----------

1910 1920 1930 1940 1950 1960 1970

2.5 coefficient estimate

--coefficient estimate ? 2 standard errors

- 1.15(uGt_ -0.2)ut2_ 1 + 0.007

and Richard, Jean-Fransois, "On the Methodology of Scientific Research Pro-

compassing Principle and Its Application Methodology," in P. E. Hart, G. Mills

S-ar putea să vă placă și