Sunteți pe pagina 1din 81

Microeconomic Theory: Basic Game

Theory
Juuso Valimaki
Contents
Chapter 1. Game Theory: An Introduction 1
1. Game Theory and Parlor Games - a Brief History 2
2. Game theory in microeconomics 2
3. Basic Elements of Noncooperative Games 2
Chapter 2. Normal Form Games 3
1. Pure Strategies 3
2. Dominance 4
3. Nash equilibrium 7
4. Mixed Strategies 10
5. Mixed Strategies with a Continuum of Pure Strategies 16
6. Appendix: Dominance Solvability and Rationalizability 23
7. Appendix: Existence of Mixed Strategy Equilibria 25
Chapter 3. Extensive Form Games 27
1. Introduction 27
2. Denitions for Extensive Form Games 28
3. Subgame Perfect Equilibrium 33
4. The Limits of Backward Induction: Additional Material 35
5. Reading 37
Chapter 4. Repeated Games 39
1. Finitely Repeated Games 39
2. Innitely Repeated Games 41
3. Collusion among duopolists 42
Chapter 5. Sequential Bargaining 47
1. Bargaining with Complete Information 48
2. Dynamic Programming 51
Chapter 6. Games of Incomplete Information 57
1. Introduction 57
2. Basics 57
3. Bayesian Game 58
4. A Game of Incomplete Information: First Price Auction 60
5. Conditional Probability and Conditional Expectation 61
6. Double Auction 64
iii
iv CONTENTS
Chapter 7. Adverse selection (with two types) 69
1. Monopolistic Price Discrimination 69
Chapter 8. Theoretical Complements 73
1. Mixed Strategy Bayes Nash Equilibrium 73
2. Sender-Receiver Games 73
3. Perfect Bayesian Equilibrium 74
Bibliography 77
CHAPTER 1
Game Theory: An Introduction
Game theory is the study of multi-person decision problems. The fo-
cus of game theory is on interdependence, situations in which an entire
group of people is aected by the choices made by every individual within
that group. Such situations appear frequently in economics: many trading
processes (auction, bargaining), labor and nancial markets, oligopolistic
markets and models of political economy all involve game theoretic reason-
ing. There are multi-agent decision problems within an organization, several
individuals may compete for a promotion, dierent divisions compete for in-
vestment capital. In international economics countries choose taris and
trade policies, in macroeconomics, the central bank attempts to control the
price level. The fundamental questions to be addressed in a game theoretic
model include:
(1) What do the individuals believe about the other individuals ac-
tions?
(2) What action will each person take?
(3) What is the outcome of these actions?
In addition we may ask
(1) Does it make a dierence if the group interacts more than once?
(2) What if each individual is uncertain about the characteristics of
the other players?
(3) What do observed actions tell about unobservable characteristics
of the player?
Three basic distinctions can be made at the outset
(1) non-cooperative vs. cooperative games
(2) strategic (or normal form) games and extensive (form) games
(3) games with complete or incomplete information
In all game theoretic models, the basic entity is a player. In noncooper-
ative games the individual player and her actions are the primitives of the
model, whereas in cooperative games coalitions of players, their joint actions
and the resulting utilities are the primitives. In other words, cooperative
game theory can be interpreted as a theory that allows for binding contracts
between individuals. These notes will consider only non-cooperative game
theory.
1
2 1. GAME THEORY: AN INTRODUCTION
1. Game Theory and Parlor Games - a Brief History
E. Zermelo (1913) chess, the game has a solution, solution concept:
backwards induction.
E. Borel (1913) mixed strategies, conjecture of non-existence.
J. v. Neumann (1928) existence of solutions in zero-sum games.
J. v. Neumann / O. Morgenstern (1944) Theory of Games and
Economic Behavior: Axiomatic expected utility theory, Zero-sum
games, cooperative game theory.
J. Nash (1950) Nonzero sum games and the concept of Nash equi-
librium.
R. Selten (1965,75) dynamic games, subgame perfect equilibrium.
J. Harsanyi (1967/68) games of incomplete information, Bayesian
equilibrium.
2. Game theory in microeconomics
(1) Decision theory (single agent)
(2) Game theory (few agents)
(3) General equilibrium theory (many agents)
3. Basic Elements of Noncooperative Games
A game is a formal representation of a situation in which a number of
individuals interact in a setting of strategic interdependence. The welfare
of an agent depends not only on his own action but on the actions of other
agents as well. The degree of strategic interdependence may often vary.
Example 1. Monopoly, Oligopoly, Perfect Competition
To describe a strategic situation we need to describe the players, the
rules of the game, the outcomes, and the payos or utilities.
Example 2. Matching Pennies, Tick-Tack-Toe (Zero Sum Games).
Example 3. Meeting in New York with unknown meeting place (Non
Zero Sum Game, Coordination Game, Imperfect Information).
CHAPTER 2
Normal Form Games
We start with games that have nitely many strategies for all players
and where players choose their strategies simultaneously at the start of the
game. In these games, the solution of the game can simply be obtained by
searching (in a clever way) among all possible solutions.
1. Pure Strategies
We begin with a coordination game played simultaneously between Alice
and Bruce. Both Alice and Bruce have a dichotomous choice to be made:
either go to an art exhibition or to ballet. This strategic interaction can be
represented by a double matrix as follows:
Art Ballet
Art 2, 2 0, 0
Ballet 0, 0 1, 1
Figure 1. Coordination
The rst entry in each cell is the payo to the row player. The second
entry is the payo to the column player from the outcome identied by
the cell. How should this game be played? What should each individual
player do?
But rst we need to give a denition of a game. In order to describe a
game, the set of players must be specied:
(1) 1 = 1, 2, ..., I .
In order to describe what might happen in the interaction, we need to specify
the set of possible choices, called strategies for each player
(2) i, s
i
S
i
,
where each individual player i has a set of pure strategies S
i
available to him
and a particular element in the set of pure strategies is denoted by s
i
S
i
.
Finally there are payo functions for each player i:
(3) u
i
: S
1
S
2
S
I
R.
A prole of pure strategies for the players is given by
s = (s
1
, ..., s
I
)
I

i=1
S
i
3
4 2. NORMAL FORM GAMES
or alternatively by separating the strategy of player i from all other players,
denoted by i:
s = (s
i
, s
i
) (S
i
, S
i
) .
A strategy prole is said to induce an outcome in the game. Hence for any
prole we can deduce the payos received by the players. This representation
of the game is known as normal or strategic form of the game.
Definition 1 (Normal Form Representation). The normal form
N
represent a game as:

N
= 1, S
i

i
, u
i
()
i
.
2. Dominance
The most primitive solution concept to games does not require knowl-
edge of the actions taken by the other players. A dominant strategy is the
best choice for a player in a game regardless of what the others are doing.
The next solution concepts assumes that rationality of the players is com-
mon knowledge. Common knowledge of rationality is essentially a recursive
notion, so that every player is rational, every player knows that every player
is rational, every player knows that every player knows that every player
is rational..., possibly ad innitum. Finally, the Nash equilibrium concept
requires that each players choice be optimal given his belief about the other
players behavior, and that this belief be correct in equilibrium.
Consider next a truly famous game which gures prominently in the
debate on the possibility of cooperation in strategic situations, the prisoners
dilemma game: Besides its somewhat puzzling conclusion this game has
Cooperate Defect
Cooperate 3, 3 0, 4
Defect 4, 0 1, 1
Figure 2. Prisoners Dilemma
another interesting property which comes to light when we analyze the payo
of each individual player, for the row player and for the column player In
C D
C 3 0
D 4 1
Figure 3. Row players payos in Prisoners Dilemma
both cases, one strategy is better than the other strategy inrrespective of
the choice made by the opponent.
2. DOMINANCE 5
C D
C 3 4
D 0 1
Figure 4. Column players payos in Prisoners Dilemma
Definition 2 (Dominant Strategy). A strategy s
i
is a dominant strategy
for player i in
N
if for all s
i
S
i
and for all s

i
,= s
i
,
u
i
(s
i
, s
i
) > u
i
_
s

i
, s
i
_
.
Definition 3 (Weakly Dominant Strategy). A strategy s
i
is a weakly
dominant strategy for player i in
N
if for all s
i
S
i
and for all s

i
,= s
i
,
u
i
(s
i
, s
i
) u
i
_
s

i
, s
i
_
,
and u
i
_
s
i
, s

i
_
> u
i
_
s

i
, s

i
_
, for some s

i
S
i
.
The most straightforward games to analyze are the ones that have dom-
inant strategies for all players. In those games, there is little need to think
about the strategic choices of others as the dominant strategy can never be
beaten by another choice of strategy.
Definition 4 (Dominant Strategy Equilibrium). If each player i in a
game
N
has a (weakly) dominant strategy s
i
, then s = (s
1
, ..., s
N
) is said
to be a (weakly) dominant strategy equilibrium of
N
.
Example 4. Second price auction (with complete information) is an
example of a game with a weakly dominant strategy equilibrium.
Definition 5 (Purely Strictly Dominated). A strategy s
i
S
i
is a
purely strictly dominated strategy for player i in
N
if there exists a pure
strategy s

i
,= s
i
, such that
(4) u
i
_
s

i
, s
i
_
> u
i
(s
i
, s
i
)
for all s
i
S
i
.
Definition 6 (Purely Weakly Dominated). A strategy s
i
is a weakly
dominated strategy for player i in
N
if there exists a pure strategy s

i
,= s
i
,
such that
u
i
(s
i
, s
i
) u
i
_
s

i
, s
i
_
, for all s
i
S
i
and u
i
_
s
i
, s

i
_
< u
i
_
s

i
, s

i
_
, for some s

i
S
i
Now consider a game

obtained from the original game after elim-


inating all strictly dominated actions. Once again, if a player knows that
all the other players are rational, then he should not choose a strategy in

that is never-best response to in

. Continuing to argue in this way leads


to the outcome in should survive an unlimited number of rounds of such
elimination:
6 2. NORMAL FORM GAMES
Definition 7 (Iterated Pure Strict Dominance). The process of iterated
deletion of purely strictly dominated strategies proceeds as follows: Set S
0
i
=
S
i
. Dene S
n
i
recursively by
S
n
i
=
_
s
i
S
n1
i

i
S
n1
i
, s.th. u
i
_
s

i
, s
i
_
> u
i
(s
i
, s
i
) , s
i
S
n1
i
_
,
Set
S

i
=

n=0
S
n
i
.
The set S

i
is then the set of pure strategies that survive iterated deletion of
purely strictly dominated strategies.
Observe that the denition of iterated strict dominance implies that in
each round all strictly dominated strategies have to be eliminated. It is
conceivable to dene a weaker notion of iterated elimination, where in each
step only a subset of the strategies have to be eliminated.
Here is an example of iterated strict dominance:

2

(3)
2

(1)
2

1
2, 3 3, 2 0, 1

(2)
1
0, 0 1, 3 4, 2
Figure 5. Iterated strict dominance
which yields a unique prediction of the game, and the superscript indi-
cates the order in which dominated strategies are removed. We might be
tempted to suggest a similar notion of iteration for weakly dominated strate-
gies, however this runs into additional technical problems as order and speed
in the removal of strategies matters for the shape of the residual game as
the following example shows: If we eliminate rst
1
, then
2
is weakly

2

2

1
3, 2 2, 2

1
1, 1 0, 0

1
1, 1 0, 0
Figure 6. Iterated weak dominance
dominated, similarly if we eliminate
1
, then
2
is weakly dominated, but
if we eliminate both, then neither
2
nor
2
is strictly dominated anymore.
A similar example is The discussion of domination demonstrated that less
restrictive notion of play may often give good prediction.. However as the
following example shows, it clearly is not enough for analyzing all strategic
interactions. The next subsection develops a solution concept that singles
out (M, M) as the relevant solution of the above game.
3. NASH EQUILIBRIUM 7
L R
T 1, 1 0, 0
M 1, 1 2, 1
B 0, 0 2, 1
Figure 7. Iterated weak dominance 2
L M R
T 3, 0 0, 2 0, 3
M 2, 0 1, 1 2, 0
B 0, 3 0, 2 3, 0
Figure 8. No dominated strategies
3. Nash equilibrium
Let us now go back to the coordination game. How to solve this game.
A solution should be such that every player is happy playing his strategy
and has no desire to change his strategy in response to the other players
strategic choices. A start is to do well given what the strategy chosen by the
other player. So let us look at the game form the point of one player, and
as the game is symmetric it is at the same time, the game of his opponent:
Art Ballet
Art 2 0
Ballet 0 1
Figure 9. Individual Payos in Battle of Sexes
Definition 8 (Best Response). In a game
N
, strategy s
i
is a best
response for player i to his rivals strategies s
i
if
u
i
(s
i
, s
i
) u
i
_
s

i
, s
i
_
for all s

i
S
i
. Strategy s
i
is never a best response if there is no s
i
for
which s
i
is a best response.
We can also put it dierently by saying that s
i
is a best response if
(5) s
i
arg max
s

i
S
i
u
i
_
s

i
, s
i
_
s
i
BR(s
i
) s
i
BR(s) .
We could now proceed and ask that the best response property holds for
every individual engaged in this play.
Definition 9 (Pure Strategy Nash Equilibrium). A strategy prole s =
(s
1
, ..., s
n
) constitutes a pure strategy Nash equilibrium of the game
N
if
for every i 1,
(6) u
i
(s
i
, s
i
) u
i
_
s

i
, s
i
_
, for all s

i
S
i
.
8 2. NORMAL FORM GAMES
In other words, a strategy prole is a Nash equilibrium if each players
strategy s
i
is a best response to the strategy prole s
i
of all the remaining
players. In the language of a best response, it says that
s BR(s) ,
and hence a pure strategy Nash equilibrium s is a xed-point under the best
response mapping. Let us now consider some more examples to test our
ability with these new notions.
3.1. Finite Normal Form Game. Consider the Hawk-Dove game
where two animals are ghting over some prey. Each can behave like a dove
or like a hawk. How is this game played and what is logic behind it.
Hawk Dove
Hawk 3, 3 1, 4
Dove 4, 1 0, 0
Figure 10. Hawk-Dove
There are two conicting interpretations of solutions for strategic and
extensive form games:
steady state (evolutionary).
deductive (eductive).
The latter approach treats the game in isolation and attempts to infer
the restrictions that rationality imposes on the outcome. The games we
are considering next are economic applications. They are distinct from our
previous examples insofar as the strategy space is now a continuum of strate-
gies. The rst problem is indeed very continuous and regular optimization
instruments apply directly. The second example has continuous but not dif-
ferentiable, payo functions and we shall proceed directly without using the
regular optimization calculus.
3.2. Cournot Oligopoly. Consider the following Cournot model in
which the sellers oer quantities and the market price is determined as a
function of the market quantities.
P(q
1
, q
2
) = a q
1
q
2
The prot functions of the rms are:

i
= (a q
1
q
2
)q
i
and dierentiated we obtain:

i
= a 2q
1
q
2

(7) q
1
(q
2
) =
a q
2
2
and q
2
(q
1
) =
a q
1
2
,
3. NASH EQUILIBRIUM 9
where q
1
(q
2
) and q
2
(q
1
) are the reaction functions or best response functions
of the players.
The steeper function is q
1
(q
2
) and the atter function is q
2
(q
1
). While
we can solve (7) directly, we realize the symmetry of the problem, which
reduces the system of equation to a single equation, and a symmetric equi-
librium, so that
q =
a q
2
and
q =
a
3
and hence the prots are
(8)
i
_
q
1
=
a
3
, q
2
=
a
3
_
=
a
2
9
.
3.3. Auctions with perfect information. Consider the following
setting in which there are n bidders with valuations for a single good
0 < v
1
< v
2
< .... < v
n
< .
We shall consider two dierent auction mechanisms: the rst and the second
price auction. The object is given to the bidder with the highest index among
all those who submit the highest bid. In the rst price auction, the winner
pays his own bid, in the second price auction, the winner pays the second
highest bid.
The payo functions in the rst price auction are
(9) u
i
(v
i
, b
1
, ..., b
n
) =
_
_
_
v
i
b
i
if b
i
> b
j
, j ,= i,
v
i
b
i
if b
i
b
j
, and if b
i
= b
j
i > j.
0, otherwise
and in the second price auction they are
(10) u
i
(v
i
, b
1
, ..., b
n
) =
_
_
_
v
i
b
k
if b
i
> b
j
, j ,= i,
v
i
b
k
if b
i
b
j
, and if b
i
= b
j
i > j.
0, otherwise
where b
k
satises the following inequality
b
i
b
k
b
j
, j ,= i, k.
We can then show that the equilibrium in the rst-price auction has the
property that the bidder with the highest value always wins the object.
Notice that this may not be true in the second price auction. However we
can show that in the second price auction, the bid b
i
= v
i
weakly dominates
all other strategies.
Reading: Chapter 0 in [2] is a splendid introduction into the subject.
Chapter b of [8] and Chapter 1-1.2 in [4] covers the material of the last two
lectures as does Chapter 1-1.2 in [3].
10 2. NORMAL FORM GAMES
4. Mixed Strategies
So far we were only concerned with pure strategies, i.e. situations where
each player i picks a single strategy s
i
. However some games dont have
equilibria in pure strategies. In these cases it is necessary to introduces an
element of surprise or blu through considering randomized decisions
by the players. Consider the game of matching pennies:
Heads Tails
Heads 1, 1 1, 1
Tails 1, 1 1, 1
Figure 11. Matching Pennies
It is immediate to verify that no pure strategy equilibrium exists. A
similar game is Rock, Paper, Scissors:
R P S
R 0, 0 1, 1 1, 1
P 1, 1 0, 0 1, 1
S 1, 1 1, 1 0, 0
Figure 12. Rock, Paper, Scissors
4.1. Basics. Let us now introduce the notation for mixed strategies.
Besides playing a pure strategy s
i
each player is also allowed to randomize
over the set of pure strategies. Formally, we have
Definition 10. A mixed strategy for player i,
i
: S
i
[0, 1] assigns
to each pure strategy s
i
S
i
, a probability
i
(s
i
) 0 that it will be played,
where

s
i
S
i

i
(s
i
) = 1
The mixed strategy can be represented as a simplex over the pure strate-
gies:
(S
i
) =
_
(
i
(s
1
) , ...,
i
(s
n
)) R
n
:
i
(s
ik
) 0, k, and
n

k=1

i
(s
ik
) = 1
_
.
This simplex is also called the mixed extension of S
i
. For [S
i
[ = 3, it
can be easily represented as a two-dimensional triangle sitting in a three
dimensional space.
i
(s
i
) is formally a probability mass function and
i
is
a probability vector. Such a randomization is called a mixed strategy
i
which is a element of set of distributions over the set of pure strategies:

i
(S
i
) .
4. MIXED STRATEGIES 11
An alternative notation for the set of mixed strategies, which you may nd
frequently is

i

i
.
In order to respect the idea that the players are choosing their strate-
gies simultaneously and independently of each other, we require that the
randomized strategies chosen by the dierent players satisfy statistical inde-
pendence.
1
In other words, we require that the probability of a pure strategy
prole s

= (s

1
, ..., s

N
) is chosen is given by
N

i=1

i
_
s

i
_
.
The expected utility of any pure strategy s
i
when some of the remaining
players choose a mixed strategy prole
i
is
u
i
(
1
, ...,
i1
, s
i
,
i+1
, ...,
I
) =

s
i
S
i
_
_

j=i

j
(s
j
)
_
_
u
i
(s
1
, ..., s
i1
, s
i
, s
i+1
, ..., s
I
) .
Similarly, the expected utility of any mixed strategy
i
is
u
i
(
1
, ...,
i1
,
i
,
i+1
, ...,
I
) =

sS
_
_

j
(s
j
)
_
_
u
i
(s
1
, ..., s
i1
, s
i
, s
i+1
, ..., s
I
) .
As another example, consider the famous battle of the sexes game:
This game has clearly two pure strategy equilibria. To see if it has a mixed
Opera Football
Opera 2, 1 0, 0
Football 0, 0 1, 2
Figure 13. Battle of Sexes
strategy equilibrium as well, calculate the payo to Sheila if Bruce is choos-
ing according to mixed strategy
B
():
u
S
(O,
B
) = 2
B
(O) + 0
B
(F) ,
and similarly
u
S
(F,
B
) = 0
B
(O) + 1
B
(F) .
Again this allows for an illuminating graphical representation by the best-
response function. Consider the best-response mapping of the two players,
1
Much of what follows can be generalized to the case where this independence con-
dition is dropped. The resulting solution concept is then called correlated equilibrium.
Since independence is a special case of such correlated random strategies, the analysis
below is a special case of correlated equilibrium. The notion of sunspots, familiar from
macroeconomic models is an example of correlated strategies. There all players observe a
signal prior to the play of the game, and condition their play on the signal even though
the signal may be completely unrelated to the underlying payos.
12 2. NORMAL FORM GAMES
where we plot the probability of attending the football game for Sheila on the
x- axis and for Bruce on the y- axis. Every point of intersection indicates
a Nash equilibrium. An equilibrium mixed strategy by Sheila would be
written as
(11)
S
=
_

S
(O) =
2
3
,
S
(F) =
1
3
_
and if the ordering of the pure strategies is given, then we can write (11)
simply as

S
=
_
2
3
,
1
3
_
.
Before we give a systematic account of the construction of a mixed strategy
equilibrium, let us extend our denitions for pure strategies to those of mixed
strategies.
Definition 11 (Normal Form Representation). The normal form rep-
resentation
N
species for each player i a set of strategies
i
, with
i

i
and a payo function u
i
(
1
, ...,
n
) giving the von Neumann-Morgenstern
utility levels associated with the (possible random) outcome arising from
strategies (
1
, ...,
n
) :

N
=
_
1, (
i
, u
i
)
iI
_
.
Definition 12 (Best Response). In game
N
=
_
1, (
i
, u
i
)
iI
_
, strat-
egy
i
is a best response for player i to his rivals strategies
i
if
u
i
(
i
,
i
) u
i
_

i
,
i
_
for all

i

i
.
Definition 13 (Nash Equilibrium). A mixed strategy prole

= (

1
, ...,

N
)
constitutes a Nash equilibrium of game
N
=
_
1, (
i
, u
i
)
iI
_
if for every
i = 1, ..., I;
(12) u
i
_

i
,

i
_
u
i
_

i
,

i
_
for all
i

i
.
Lemma 1. Condition (12) is equivalent to:
(13) u
i
_

i
,

i
_
u
i
_
s
i
,

i
_
,
for all s
i
S
i
, since any mixed strategy is composed of pure strategies.
Since this is the rst proof, we will be somewhat pedantic about it.
Proof. Suppose rst that (12) holds. Then, it is sucient (why?) to
show that for every s

i
S
i
, there exists

i

i
such that
u
i
_

i
,

i
_
u
i
_
s

i
,

i
_
But since the set of pure strategies is a strict subset of the set of mixed
strategies, or S
i

i
, we can simply set

i
= s

i
, and hence it follows that
(13) has to hold.
4. MIXED STRATEGIES 13
Next suppose that (13) holds and we want to show that this implies that
(12) holds. It is then sucient (again, why) to show that for every

i

i
,
there exists s

i
S
i
such that
u
i
_
s

i
,

i
_
u
i
_

i
,

i
_
.
But since every mixed strategy
i
is composed out of pure strategies, in
particular since
u
i
_

i
,

i
_
=

s
i
S
i

i
(s
i
) u
i
_
s
i
,

i
_
,
consider the pure strategy s
i
among all those with
i
(s
i
) > 0 which achieves
the highest payo u
i
_
s
i
,

i
_
. Then it must be the case that
u
i
_
s
i
,

i
_
u
i
_

i
,

i
_
,
and hence the conclusion.
The following proposition indicates how to construct mixed strategy
equilibria. Let
S
+
i
(
i
) = s
i
S
i
[
i
(s
i
) > 0 ,
then S
+
i
(
i
) is the set of all those strategies which receive positive probabil-
ity under the mixed strategy
i
. As S
+
i
(
i
) contains all points for which
i
assigns positive probability, it is mathematically referred to as the support
of
i
.
Proposition 2 (Composition). Let S
+
i
(

i
) S
i
denote the set of pure
strategies that player i plays with positive probability in a mixed strategy
prole

= (

1
, ...,

I
). Strategy prole

is a Nash equilibrium in game

N
=
_
1, (
i
, u
i
)
iI
_
if and only if for all i = 1, ..., I
(i) u
i
_
s
i
,

i
_
= u
i
_
s

i
,

i
_
for all s
i
, s

i
S
+
i
(

i
) ;
(ii) u
i
_
s
i
,

i
_
u
i
_
s

i
,

i
_
for all s
i
S
+
i
(

i
) , s

i
S.
4.2. Construction of a mixed strategy Nash equilibrium. We
return to the example of the battle of sexes.
Sheila
Bruce
Opera Football
Opera 2, 1 0, 0
Football 0, 0 1, 2
Figure 14. Battle of Sexes
Then a strategy for Sheila has the following payo if Bruce is playing a
mixed strategy
B
()
u
S
(
B
, O) = 2
B
(O) + 0
B
(F) ,
and similarly
u
S
(
B
, F) = 0
B
(O) + 1
B
(F) .
14 2. NORMAL FORM GAMES
We want to create a mixed strategy for Sheila and Bruce. If Sheila is sup-
posed to use a mixed strategy, it means that she picks either alternative with
a positive probability. As the strategy is supposed to be a best-response to
her, it has to be the case that the expected payos are identical across the
two alternative, for otherwise Sheila would clearly be better o to use only
the pure strategy which gives her the strictly higher payo. In short, it has
to be that
u
S
(
B
, O) = u
S
(
B
, F)
or explicitly
(14) 2
B
(O) + 0
B
(F) = 0
B
(O) + 1
B
(F) .
Thus when we consider whether Sheila will randomize, we examine con-
ditions on the mixing behavior of Bruce under which Sheila is indierent.
Let

B
(F) =
B
and consequently

B
(O) = 1
B
.
The condition (14) can then be written as
(15) 2 (1
B
) =
B

B
=
2
3
.
Thus if Sheila is to be prepared to randomize, it has be that Bruce is ran-
domizing according to (15). Now Bruce is willing to randomize with
B
=
1
3
in equilibrium if and only if he is indeed indierent between the pure strategy
alternatives. So, now we have to ask ourselves when is this the case. Well,
we have to investigate in turn how the indierence condition of Bruce deter-
mines the randomizing behavior of Sheila. (The construction of the mixed
strategy equilibrium highlights the interactive decision problem, which is
essentially solved as a xed point problem.) A similar set of conditions as
before gives us
u
B
(
S
, O) = u
B
(
S
, F)
or explicitly
(16)
S
(O) + 0
S
(F) = 0
S
(O) + 2
S
(F) .
Thus as we consider whether Bruce is willing to randomize, we examine
conditions on the mixing behavior of Sheila to make Bruce indierent. Let

S
(F) =
S
and consequently

S
(O) = 1
S
.
The condition (16) can then be written as
(17) (1
S
) = 2
S

S
=
1
3
.
4. MIXED STRATEGIES 15
Thus we reached the conclusion that the following is a mixed strategy Nash
equilibrium

B
=
_

B
(O) =
1
3
,
B
(F) =
2
3
_

S
=
_

S
(O) =
2
3
,
S
(F) =
1
3
_
.
The payo to the two players is then
2
3
in this mixed strategy equilibrium.
This examples presents the general logic behind the construction of a mixed
strategy Nash equilibrium. The only issue which we still have to discuss is
the support of the mixed strategy equilibrium.
The following game is a helpful example for this question:

1, 1 1, 1 x, x
1, 1 1, 1 x, x
x, x x, x x, x
Figure 15. Augmented Matching Pennies
For x < 0, cannot be part of any mixed strategy equilibrium, and for
x > 0, the original matching pennies equilibrium cannot be a mixed strategy
equilibrium anymore. However for x = 0, there can be many mixed strategy
equilibria, and they dier not only in the probabilities, but in the support
as well.
2
4.3. Interpretation of mixed strategy equilibria. Finally, consider
the competing interpretations of a mixed strategy Nash equilibrium:
(1) mixed strategies as objects of choice
(2) steady state, probability as a frequency of certain acts
(3) mixed strategies as pure strategies in a game where the randomness
is introduced through some random private information (Harsanyis
purication argument)
Remark 1. In the battle of sexes, the mixed strategy equilibrium used
private randomization by each player.
Suppose instead they would have a common (or public) randomization
device, say a coin, then they could realize a payo of (
3
2
,
3
2
) by e.g. both
going to opera if heads and both going to football if tails. Observe that this
payo is far superior to the mixed strategy equilibrium payo The notion of
public randomization is used in the correlated equilibrium.
Finally we consider games where each player has a continuum of actions
available. The rst one is an example of an all pay auction.
2
Verify that there cant be an equilibrium in which randomization occurs between
and only, or alternatively between and .
16 2. NORMAL FORM GAMES
5. Mixed Strategies with a Continuum of Pure Strategies
5.1. All pay auction. Two investors are involved in a competition
with a prize of v. Each investor can spend any amount in the interval [0, v].
The winner is the investor who spends the most. In the event of tie each
investor receives v/2. Formulate this situation as a strategic game and nd
its mixed strategy Nash equilibria.
The all pay auction is, despite its initially funny structure, a rather com-
mon strategic situation: Rae, R&D patent races, price wars of attrition,
wars, winner take all contests with participation costs.
The payo function of the very agent is therefore given by
u
i
(v, b
i
, b
j
) =
_
_
_
v b
i
for b
j
< b
i
1
2
v b
i
, for b
i
= b
j
b
i
for b
i
< b
j
The mixed (or pure) strategy by player i can be represented by the distri-
bution function of the bid of i:
F
i
(b
i
) ,
where we recall that the distribution function (or cumulative distribution
function) F
i
(b
i
) is dening the event
F
i
(b
i
) = Pr (b b
i
)
We recall the following notation:
F
_
b

_
= lim
ab
F (a) ,
and
F
_
b
+
_
= lim
ab
F (a) .
Before we can analyze and derive the equilibrium, we have to ask what is
the expected payo from submitting a bid b
1
given that player 2 uses an
arbitrary distribution function F
2
. From the payo functions, we see that it
can be calculated as:
E[u
1
(v, b
1
) [F
2
] = F
2
(b
2
< b
1
) (v b
1
) +
(F
2
(b
2
b
1
) F
2
(b
2
< b
1
))
_
1
2
v b
1
_
+ (1 F (b
2
b
1
)) (b
1
) ,
which can be written more compactly as
(18) E[u
1
(v, b
1
) [F
2
] = F
2
_
b

1
_
v +
_
F
2
(b
1
) F
2
_
b

1
__
1
2
v b
1
.
This expected payo function shows that in this auction the bidder has to
nd the best trade-o between increasing the probability of winning and the
cost associated with increasing the bid.
We suppose for the moment that the distribution function F
2
() is
continuous-(ly dierentiable) everywhere. Then (18) can be rewritten as
E[u
1
(v, b
1
) [F
2
] = F
2
(b
1
) v b
1
.
5. MIXED STRATEGIES WITH A CONTINUUM OF PURE STRATEGIES 17
As agent i chooses his bids optimally in equilibrium, his bid must maximize
his payo. In other words, his optimal bidding strategy is characterized
by the rst order conditions which represent the marginal benet equals
marginal cost condition
(19) f
2
(b
1
) v 1 = 0,
which states that the increase in probability of winning which occurs at the
rate of f (b
1
) multiplied by the gain of v must equal the marginal cost of
increasing the bid which is 1. As we attempt to construct a mixed strat-
egy equilibrium, we know that for all bids with f
1
(b
1
) > 0, the rst order
condition has to be satised. Rewriting (19) we get
(20) f
2
(b
1
) =
1
v
.
As we look for a symmetric equilibrium, in which agent i and j behave
identically, we can omit the subscript in (20) and get a uniform (constant)
density
f

(b) =
1
v
with associated distribution function:
(21) F

(b) =
b
v
.
As F (b = 0) = 0 and F (b = v) = 1, it follows that F

(b) in fact constitutes


a mixed equilibrium strategy.
We went rather fast through the derivation of the equilibrium. Let us
now complete the analysis. Observe rst that every pure strategy can be
represented as a distribution function. We start by explaining why there
cant be a pure strategy equilibrium in the all-pay auction, or in other words
why there can be no atoms with probability 1.
Lemma 3 (Non-existence of pure strategy equilibrium). a pure strategy
equilibrium in the all pay auction.
Proof. We argue by contradiction. Suppose therefore that there is a pure
strategy equilibrium. Then either (i) b
i
> b
j
or (ii) b
i
= b
j
. Suppose rst
that
b
i
> b
j
then b
i
would lower his bid and still win the auction. Suppose then that
b
i
= b
j
.
The payo of bidder i would then be
1
2
v b
i
,
but by increasing the bid to b
i
+ for small , he would get the object for
sure, or
v b
i
.
18 2. NORMAL FORM GAMES
As 0, and when taking the dierence, we see that i would have a jump
in his net utility of
1
2
v as
v b
i

_
1
2
v b
i
_
=
1
2
v
1
2
v,
which concludes the proof.
As we have now shown that there is no pure strategy equilibrium, we
can concentrate on the derivation of a mixed strategy equilibrium. In the
derivation above, we omitted some possibilities: (i) it could be that the
distribution function of agent i is not continuous and hence has some mass
points or atoms, (ii) we have not explicitly derived the support of the dis-
tribution. We will in turn solve both problems.
A similar argument which excludes a pure strategy (and hence atoms
with mass one) also excludes the possibility of atoms in mixed strategy
equilibrium.
Lemma 4 (Non-existence of mass points in the interior). a mixed
strategy equilibrium with a mass point b in the interior of the support of
either players mixed strategy ( i.e. F
i
(b) F
i
(b

) > 0 for some i) in the


all pay auction.
Proof. Suppose to the contrary that the strategy of player j contains an
atom so that for some b,
F
j
(b) F
j
_
b

_
> 0,
If F
j
() has an atom at b, then for any > 0, it must be that
F
i
(b) F
i
(b ) > 0,
otherwise j could lower the bids at the atom without changing the prob-
ability of winning. But consider the winnings of i at b

< b. They are at


best
vF
j
_
b

_
b

but by increasing the bid to b

> b, i would receive


vF
j
_
b

_
b

.
But letting b

b and b

b, we can calculate the expected payo dierence


between
(22)
lim
b

b, b

b
_
vF
j
_
b

_
b

_
vF
j
_
b

_
b

_
= v
_
F
j
(b) F
j
_
b

__
> 0,
and hence there would be a protable deviation for player i.
Notice the argument applies to any b < v, but not for b v, and hence
there could be atoms at the upper bound of the support of the distribution.
3
3
Recall the support of a (density) function is the closure of the set
{x : f (x) = 0} ,
5. MIXED STRATEGIES WITH A CONTINUUM OF PURE STRATEGIES 19
It then remains to nd the mixed strategy equilibrium which has an interval
as its support. Suppose F
i
(b
i
) is the distribution function which denes the
mixed strategy by player i. For her to mix over any interval
(23) b
i

_
b
i
, b
i

she has to be indierent:


(24) vF
j
(b) b = c,
where c is the expected payo from any mixed strategy equilibrium. Argue
that c = 0, then we solve the equation to
(25) F
j
(b) =
b
v
,
which tells us how player j would have to randomize to make player i in-
dierent. The argument is actually already made by the derivation of the
rst order conditions and the observation that there cant be any atom in
the interior of [0, v]. By symmetry, (25) then denes the equilibrium pricing
strategy by each player.
Suppose we were to increase the number of bidders, then (24) changes
under the assumption of symmetry to
(26) v (F (b))
n1
b = c,
where c is the expected payo from any mixed strategy equilibrium. Argue
again that c = 0, then we solve the equation to
(27) F (b, n) =
_
b
v
_ 1
n1
and the expected payment of each individual bidder is
_
v
0
bdF (b) =
_
v
0
bf (b) db
and since
F

(b) = f (b) =
_
b
v
_
1
n1
(n 1) b
we obtain
(28)
_
v
0
_
b
v
_
1
n1
(n 1)
db =
v
n
.
and thus the support is
supp f = {x : f (x) = 0}
20 2. NORMAL FORM GAMES
5.2. War of attrition. Suppose two players are competing for a prize
of value v. They have to pay c per unit of time to stay in the competition.
If either of the two player drops out of the competition, then the other
player gets the object immediately. This is a simple example of a timing
game, where the strategy is dened by the time
i
at which the player drops
out. There are pure strategy Nash equilibria, which have
_

i
= 0,
j

c
v
_
,
however the only symmetric equilibrium is a mixed strategy equilibrium,
where each player must be indierent between continuing and dropping out
of the game at any point in time, or
f
i
(t)
1 F
i
(t)
v = c
f
i
(t)
1 F
i
(t)
=
c
v
,
where the indierence condition may be interpreted as a rst order dier-
ential equation
dF
i
(t)
(1 F
i
(t))
=
c
v
.
It is very easy to solve this equation once it is observed that the left hand
side in the equation is

d
dt
ln (1 F
i
(t)) .
Hence integrating both sides and taking exponentials on both sides yields
1 F
i
(t) = e

c
v
t
, or F
i
(t) = 1 e

c
v
t
.
Alternatively, we may appeal to our knowledge of statistics, recognize
f
i
(t)
1 F
i
(t)
as the hazard rate
4
, and notice that only the exponential distribution func-
tion has a constant hazard ratio
F (t) = 1 e
t
and
f (t) = e
t
.
The symmetric equilibrium is then given by:
F (t) = 1 e

c
v
t
4
To understand why this is called the hazard rate, think of t as the time a machine
fails. Then as time t passes, and moving from t to t + dt, one nds that the conditional
probability that the machine fails in this interval, conditional on not having failed before
is
f (t) dt
1 F (t)
.
Als recall that the conditional probability is
P (A|B) =
P (A B)
P (B)
5. MIXED STRATEGIES WITH A CONTINUUM OF PURE STRATEGIES 21
and the expected time of dropping out of the game, conditional on the other
player staying in the game forever is
_

0
tdF (t) =
_

0
tf (t) dt =
v
c
.
5.3. Appendix: Basic probability theory. The general framework
of probability theory involves an experiment that has various possible out-
comes. Each distinct outcome is represented by a point in a set, the sample
space. Probabilities are assigned to certain outcomes according to certain
axioms. More precisely X is called a random variable, indicating that the
value of X is determined by the outcome of the experiment. The values X
can take are denoted by x. We can then refer to events such as X = x or
X x and the sample space is A.
We distinguish two cases of probability distributions: discrete and con-
tinuous.
5.3.1. Discrete Case. In the discrete case, the numbers of possible out-
comes is either nite or countable and so we can list them. Consider the
following experiment in which the numbers x 1, 2, 3, 4 are drawn with
probabilities p (x) as indicated. Any point x with p (x) > 0 is called a mass
point or an atom. The function p () is a probability mass function
p : A [0, 1] ,
such that

xX
p (x) = 1.
This is an example of discrete probabilities. The cumulative distribution
function is then given by
F (z) =

xz
p (z) .
Notice that F (x) is a piecewise constant function, which is monotoni-
cally increasing and at the points x 1, 2, 4 discontinuous. The function
F (x) is zero before 1 and 1 at and after 4.
5.3.2. Continuous Case. Suppose next that we wish to choose any num-
ber between x [0, 4]. The assignment of probabilities occurs via the density
function f (x) 0, so that
_
4
0
f (x) dx = 1.
The likelihood that we pick a number x x

is then given by
F
_
x

_
= F
_
x x

_
=
_
x

0
f (x) dx
22 2. NORMAL FORM GAMES
The likelihood that a number x [x

, x

] is chosen in the experiment is


given by
F
_
x

_
F
_
x

_
=
_
x

f (x) dx
In other words, any interval of numbers [x, x

] has some positive probability


of being chosen. An example of such a distribution function is then given
by:
(29) F (x) =
_
_
_
0, for x < 0
1
_
1
1
4
x
_
3
, for 0 x 4
1, for x > 4.
We may then ask what is the probability that a number is drawn out of
the interval (x, x

), and it is given by
F
_
x

_
F (x) .
The density function then identies the rate at which probability is added
to the interval by taking the limit as = x

x converges to zero:
f (x) = lim
0
F (x

) F (x)
x

x
=
F (x + ) F (x)

.
Notice that the density is only the rate at which probability is added to the
interval, but is not the probability of a particular x, call it x occurring since
the probability of x occurring is of course 0 under a continuous distribution
function, since
_
x
x
f (x) dx = 0.
Thus we have to distinguish between events which have probability zero, but
are possible and impossible events. The density function associated with the
distribution function F (x) is given by
f (x) =
_
_
_
0, for x < 0
3
4
_
1
1
4
x
_
2
, for 0 x 4
0 for x > 4.
5.3.3. Expectations. The density function f (x) is of course
F (x) =
_
x

f (z) dz
as we had in the discrete case the analogous expression in the sum
F (x) =

zx
p (z) .
Finally we can take the expectation of the lottery as
E[x] =
_

zf (z) dz
6. APPENDIX: DOMINANCE SOLVABILITY AND RATIONALIZABILITY 23
or for discrete random variables
E[x] =

z
zp (z)
Reading: Recall: A brief review of basic probability concepts, including
sample space and event, independent events, conditional probabilities and
Bayes formula, discrete and continuous random variables, probability mass
function, density and distribution function, all introduced in [12], Chapter
1 -2.4 is highly recommended. [10], chapter 3 contains a very nice discus-
sion of the interpretation and conceptual aspects of the equilibrium notion,
especially in its mixed version.
6. Appendix: Dominance Solvability and Rationalizability
In this section, we elaborate on the notion of iterated deletion of dom-
inated strategies. We dene dominance in a way that allows for the use of
mixed strategies as well and we also connect this concept to the notion of
rationalizable strategies where the question is phrased in terms of nding a
set of mutually compatible strategies in the sense that all such strategies are
best responses to some element in the other players allowed set of strategies.
Nash equilibrium is a special case of this concept where it is also required
that the sets of allowed strategies are singletons for all players.
Definition 14 (Strictly Dominated). A strategy s
i
S
i
is a strictly
dominated strategy for player i in game
N
=
_
1, (
i
, u
i
)
iI
_
if there exists
a mixed strategy

i
,= s
i
, such that
(30) u
i
_

i
, s
i
_
> u
i
(s
i
, s
i
)
for all s
i
S
i
.
Remark 2. This notion of dominance is based on domination by pure
or mixed strategies. Clearly, more strategies can be dominated if we allow
mixed strategies to be in the set of dominating strategies.
Definition 15 (Never-Best response). A strategy s
i
is never-best re-
sponse if it is not a best response to any belief by player i. In other words,
for all
i

i
, there is a s

i
such that
u
i
_
s

i
,

i
_
> u
i
_
s
i
,

i
_
.
Lemma 5. In a two player game, a strategy s
i
of a player is never-best
response if and only if it is strictly dominated.
The notion of strict dominance has hence a decision-theoretic basis that
is independent of the notion of mixed strategy.
5
5
The above Lemma extends to more than two players if all probability distributions
on S
i
and not only the ones satisfying independence are allowed.
24 2. NORMAL FORM GAMES
Now consider a game

obtained from the original game after elimi-


nating all strictly dominated actions. Once again if a player knows that all
the other players are rational, then he should not choose a strategy in

that is never-best response to in

. Continuing to argue in this way leads


to the outcome in should survive an unlimited number of rounds of such
elimination:
Definition 16 (Iterated Strict Dominance). The process of iterated
deletion of strictly dominated strategies proceeds as follows: Set S
0
i
= S
i
,
and
i
=
0
i
. Dene S
n
i
recursively by
S
n
i
=
_
s
i
S
n1
i

i

n1
i
, s.th. u
i
(
i
, s
i
) > u
i
(s
i
, s
i
) , s
i
S
n1
i
_
,
and similarly

n
i
=
i

i
[
i
(s
i
) > 0 only if s
i
S
n
i
.
Set
S

i
=

i=1
S
i
,
and

i
=
_

i

i

i

i
s.th. u
i
_

i
, s
i
_
> u
i
(
i
, s
i
) , s
i
S

i
_
The set S

i
is then the set of pure strategies that survive iterated deletion
of strictly dominated strategies. The set

i
is the set of players i mixed
strategies that survive iterated strict dominance.
6.1. Rationalizable Strategies. The notion of rationalizability be-
gins by asking what are all the strategies that a rational player could play?
Clearly, a rational player will play a best response to his beliefs about the
play of other players and hence she will not use a strategy that is never
a best response. Hence such strategies may be pruned from the original
game. If rationality of all players is common knowledge, then all players
know that their opponents will never use strategies that are never best re-
sponses. Hence a player should use only strategies that are best responses
to some belief of other players play where the others are not using never
best response strategies. Iteration of this reasoning leads to a denition of
rationalizable strategies. Denote the convex hull of A by co (A) . In other
words, co (A) is the smallest convex set containing A.
Definition 17 (Rationalizable Strategies). Set
0
i
=
i
and for each i
dene recursively
(31)

n
i
=
_

i

n1
i

i

_

j=i
co
_

n1
j
__
, s.th. u
i
(
i
,
i
) u
i
_

i
,
i
_
,

i

n1
i
_
.
The rationalizable strategies for player i are
(32) R
i
=

n=0

n
i
.
7. APPENDIX: EXISTENCE OF MIXED STRATEGY EQUILIBRIA 25
_
1, (
i
, u
i
)
iI
_
, the strategies in (S
i
) that survive the iterated removal of
strategies that are never a best response are known as player i

s rationalizable
strategies.
The reason for having the convex hull operator in the recursive deni-
tion is the following. Each point in the convex hull can be obtained as a
randomization over undominated strategies for j ,= i. This randomization
corresponds to i

s subjective uncertainty over j

s choice. It could fully well


be that this random element is dominated as a mixed strategy for player j.
Hence co
_

n1
j
_
,=
n1
j
in some cases.
Rationalizability and iterated strict dominance coincide in two player
games. The reason for the coincidence is the equivalence of never-best re-
sponse and strictly dominated in two player games which does not hold in
three player games as the following example suggests. In the game below,
we have pictured the payo to player 3, and player 3 chooses either matrix
A, B, C or D.

2

2

1
9 0

1
0 0
A

2

2

1
0 9

1
9 0
B

2

2

1
0 0

1
0 9
C

2

2

1
6 0

1
0 6
D
Figure 16. Rationalizable vs. undominated strategies
where we can show that action D is not a best response to any mixed
strategy of players 1 and 2, but that D is not dominated. If each player
i can have a correlated belief about the others choices, the equivalence
of correlated rationalizability and iterated deletion of strictly dominated
strategies can be established.
7. Appendix: Existence of Mixed Strategy Equilibria
Up to this point, we have not shown that mixed strategy equilibria exist
for general games. We have simply assumed their existence and character-
ized them for some games. It is important to know that for most games,
mixed strategy Nash equilibria do exist.
The rst existence result was proved by Nash in his famous paper [9].
Theorem 6 (Existence in Finite Games). Finite games, i.e. games with
a nite number of players and a nite number of strategies for each player,
have a mixed strategy Nash equilibrium.
26 2. NORMAL FORM GAMES
The proof of this result is a standard application of Kakutanis xed
point theorem, and it is given in almost all graduate level textbooks, for
instance the appendix to Chapter 8 in [8] give the proof.
Many strategic situations are more easily modeled as games with a con-
tinuum of strategies for each player (e.g. the Cournot quantity competition
mode above). For these games, slightly dierent existence theorems are
needed. The simplest case that was also proved by Nash is the following
(and also covered in the same appendix of [8]).
Theorem 7 (Continuous-Quasiconcave case). Let
N
=
_
1, (S
i
, u
i
)
iI
_
be a normal form game where 1 = 1, ..., N for some N < and S
i
=
[a
i
, b
i
] R for all i. Then if u
i
: S R is continuous for all i and quasi-
concave in s
i
for all i,
N
has a pure strategy Nash equilibrium.
Observe that this theorem actually guarantees the existence of a pure
strategy equilibrium. This results from the assumption of continuity and
quasiconcavity. If either of these requirements is dropped, pure strategy
equilibria do not necessarily exist.
Unfortunately, the payos in many games of interest are discontinuous
(e.g. the all pay auction and the war of attrition examples above show
this). For these games, powerful existence results have been proved, but the
mathematical techniques are a bit too advanced for this course. State of the
art results are available in the recent article [11]. The basic conclusion of
that paper is that it will be very hard to nd economically relevant games
where existence of mixed strategy equilibria would be doubtful.
An articial game where no equilibria (in either pure or mixed strategies)
exists is obtained by considering a modication of the Cournot quantity
game. Recall that in that game, the payos to the two players are given by

i
(q
i
, q
j
) = (a q
1
q
2
) q
i
.
If the payos are modied to coincide with these at all (q
1
, q
2
) except
_
a
3
,
a
3
_
and
i
_
a
3
,
a
3
_
= 0 for both rms, then one can show that the game has
no Nash equilibria. The argument to show this is that all of the strategies
except for q
i
=
a
3
are deleted in nitely many rounds of iterated deletion of
strictly dominated strategies. Hence none of those strategies can be in the
support of any mixed strategy Nash equilibrium. But it is also clear that
_
a
3
,
a
3
_
is not a Nash equilibrium of the game.
Reading: Chapter 8 in [8], chapter 2 in [3] and to a lesser extent Chapter
1 in [4] cover the material of this chapter.
CHAPTER 3
Extensive Form Games
1. Introduction
This lecture covers dynamic games with complete information. As dy-
namic games, we understand games which extend over many periods, either
nitely many or innitely many. The new concept introduced here is the
notion of subgame perfect Nash equilibrium (SPE), which is closely related
to the principle of backward induction.
The key idea is the principle of sequential rationality: equilibrium strate-
gies should specify optimal behavior from any point in the game onward.
The notion of subgame perfect Nash equilibrium is a strengthening of the
Nash equilibrium to rule out incredible strategies. We start with nite
games, in particular games with a nite horizon.
We then consider two of the most celebrated applications of dynamic
games with complete information: the innite horizon bargaining game of
Rubinstein and the theory of repeated games.
Example 5. Predation. There is an entrant rm and an incumbent
rm. Sharing the market (1, 1) is less protable than monopoly (0, 2), but
it is better than a price war (3, 1). Depict the game in the normal form
and the extensive form. Characterize the equilibria of the game. Distinction
between subgame perfect and Nash equilibrium. Credible vs. incredible threat.
Future play (plan) aects current action.
Example 6. Cournot Model. Stackelberg Equilibrium. Consider
again the quantity setting duopoly with aggregate demand
P(q
1
, q
2
) = a q
1
q
2
and the individual prot function:

i
(q
i
, q
j
) = (a q
1
q
2
)q
i
The best response function in the quantity setting game are, as we observed
before,
q
j
= R(q
i
) =
a q
i
2
Suppose player i moves rst, and after observing the quantity choice of player
i , player j chooses his quantity optimal. Thus i takes into account the
reaction (function) of player j, and hence

i
(q
i
, R(q
i
)) = (a q
i

a q
i
2
)q
i
= (
a
2

q
i
2
)q
i
27
28 3. EXTENSIVE FORM GAMES
and the associated rst-order conditions are

i
(q
i
, R(q
i
)) =
a
2
q
i
= 0.
The Stackelberg equilibrium is then
q
i
=
a
2
,
i
_
a
2
,
a
4
_
=
a
2
8
q
j
=
a
4
,
j
_
a
2
,
a
4
_
=
a
2
16
The sequential solution method employed here is called backward induction
and the associated Nash equilibrium is called the subgame perfect Nash
equilibrium. Comparing with the Cournot outcome we nd
q
S
i
+q
S
j
q
C
i
+q
C
i
=
a
3
+
a
3
but

S
i
>
C
i
=
a
2
9
,
S
j
<
C
j
.
This example illustrates the value of commitment, which is in this game
simply achieved by moving earlier than the opponent.
Notice that the Cournot equilibrium is still a Nash equilibrium. But is it
reasonable? One has to think about the issue of non credibility and empty
threats.
2. Denitions for Extensive Form Games
2.1. Introduction. The extensive form of the game captures (i) the
physical order of play, (ii) the choices each player can take, (iii) rules deter-
mining who is to move and when, (iv) what players know when they move,
(v) what the outcome is as a function of the players actions and the players
payo from each possible outcome, (vi) the initial condition that begin the
game (move by nature).
2.2. Graph and Tree. We introduce all the elements of the conceptual
apparatus, called the game tree. The language of a tree is borrowed from
graph theory, a branch of discrete mathematics and combinatorics.
A graph G is given by (V, E), where V = v
1
, ..., v
n
is a nite set of
nodes or vertices and E = v
i
v
j
, v
k
v
l
, ..., v
r
v
s
is a set of pairs of vertices
(or 2-subsets of V), called branches or edges which indicates which nodes
are connected to each other. In game theory, we are mostly interested in
graphs that are directed. Edges in a directed graph are ordered pairs of
nodes (v
1
, v
j
) and we say that the edge starts at v
i
and ends at v
j
. A walk
in a graph G is a sequence of vertices,
v
1
, v
2
, ...., v
n
such that v
i
and v
i+1
are adjacent, i.e., form a 2-subset. If all its vertices
are distinct, a walk is called a path. A walk v
1
, v
2
, ...., v
n
such that v
1
= v
n
is called a cycle.
2. DEFINITIONS FOR EXTENSIVE FORM GAMES 29
A graph T is a tree if it has two properties
(T1) T is connected,
(T2) there are no cycles in T,
where connected means that there is a path from every vertex to every other
vertex in the graph. A forest is a graph satisfying (T2) but not necessarily
(T1).
2.3. Game Tree. The game form is a graph T endowed with a physical
order represented by , formally (T, ) The order represents precedence
on nodes. We write y x if y precedes x in the game form, i.e. if every path
reaching x goes through y. The relation totally orders the predecessors
of each member of V . Thus is asymmetric and transitive.
The following notation based on the game form (T, ) is helpful:
name notation denition
terminal nodes (outcomes) Z t T : s (t) =
decision nodes X TZ
initial nodes W t T : p (t) =
predecessors of t p (t) x T : x t
immediate predecessor of t p
1
(t) x T : x t and y t y x
n-th predecessor p
n
(t) p
1
(p
n1
(t)) with p
n1
(t) / W, p
0
(t) = t
number of predecessors l (t) p
l(t)
(t) W
immediate successor s (x) y T : x y and t y y t
terminal successor z (x) z Z : x z for x X.
In order to dene formally the extensive form game, we need the follow-
ing ingredients.
(i) The set of nodes V consists of initial nodes W, decision nodes X,
and terminal nodes Y .
T = W, X, Y .
(ii) A mapping
p : TW T,
which is called the predecessor mapping (it is a point to set mapping). The
immediate predecessor nodes are p
1
(x) and the immediate successor nodes
of x are
s (x) .where p
1
(s (x)) = x.
(iii) An assignment function
: TW A,
where we require that is a one-to-one function onto the set s (x) of x. The
term (s (x)) represents the sets of all actions which can be taken at x.
30 3. EXTENSIVE FORM GAMES
(iv) A function
: X I
assigning each decision node to the player who moves at that node.
Definition 18. A partition H of a set X is a collection of subsets of
X, with
H = h
1
, ..., h
k
,
s.th.
h
k
h

k
= ,
and
_
k
h
k
= X.
Definition 19. A partition H of X is called a collection of information
sets if x

, x

(33) x

h
_
x

_

_
x

_
=
_
x

_
and
_
s
_
x

__
=
_
s
_
x

__
.
Information sets as histories. By the consistency requirement im-
posed in (33) we dene the set of actions which can be taken at an informa-
tion set h as A(h), where
A(h) (s (h)) A(h)
_
s
_
x

__
=
_
s
_
x

__
, for all x

h
_
x

_
.
The partition H can be further partitioned into H
i
such that H
i
=

1
(i), where H
i
is the set of all decision nodes at which i is called to make
a move. A prominent way of identifying an information set is through the
history leading to it. Hence h H is often called a particular history of the
game.
1
(vi) A collection of payo functions
u = u
1
() , ..., u
I
() ,
assigning utilities to each terminal node that can be reached
u
i
: Z R.
Observe that since each terminal node can only be reached through a single
path, dening payos on terminal nodes is equivalent to dening payos on
paths of actions.
(vii) An initial assessment as a probability measure over the set of
initial nodes
: W [0, 1] ,
assigning probabilities to actions where nature moves.
Definition 20 (Extensive Form Game). The collection T, ; A, ; I, ; H; u
i
;
is an extensive form game.
1
For notational convenience, we assume that is onto and that for each a A,
A
1
(a) is a singleton on H, but not on X of course. That is, each action can be taken
only in a single information set.
2. DEFINITIONS FOR EXTENSIVE FORM GAMES 31
Definition 21 (Perfect Information). A game is one of perfect infor-
mation if each information set contains a single decision node. Otherwise,
it is a game of imperfect information.
Definition 22 (Almost Perfect Information). A game is of almost
perfect information if the players choose simultaneously in evert period t,
knowing all the actions chosen by everybody at dates 0 to t 1.
2.4. Notion of Strategy. A strategy is a complete contingent plan
how the player will act in every possible distinguishable circumstance in
which she might be called upon to move. Consider rst the following game
of perfect information:
Definition 23 (Pure Strategy). A strategy for player i is a function
(34) s
i
: H
i
/,
where / =
h
A(h) , such that s
i
(h) A(h) for all h H
i
.
Definition 24 (Mixed Strategy). Given player i

s (nite) pure strategy


set S
i
, a mixed strategy for player i,

i
: S
i
[0, 1]
assigns to each pure strategy s
i
S
i
a probability
i
(s
i
) 0 that it will be
played where

s
i
S
i

i
(s
i
) = 1.
Definition 25 (Outcome). The outcome induced by a mixed strategy
prole in an extensive form game is the distribution induced by the initial
assessment and the mixed strategies on the terminal nodes.
In general the number of pure strategies available to each player is given
by
[S
i
[ =

h
i
H
i
[A(h
i
)[
where [S[ represents in general the cardinality of a set S, i.e. the number of
members in the set. If the number of actions available at each information
set h
i
is constant and given by [A(h
i
)[, then [S
i
[ simplies to
[S
i
[ = [A(h
i
)[
|H
i
|
.
In extensive form games, it is often customary to use a slightly dierent
notion for a mixed strategy. In the denition above, the number of compo-
nents needed to specify a mixed strategy is [S
i
[ 1. This is often a very large
number. A simpler way to dene a random strategy is the following. Rather
than considering a grand randomization at the beginning of the game for
each player, we could consider a sequence of independent randomizations,
one at each information set for each of the players. This leads to the notion
of a behavior strategy.
32 3. EXTENSIVE FORM GAMES
Definition 26 (Behavior Strategy). A behavior strategy of player i is
a mapping
b
i
: H
i
(/)
such that for all h
i
H
i
, b (h
i
) (A(h
i
)) .
Observe that in order to specify a behavior strategy for player i, only

h
i
H
i
[A(h
i
) 1[
components must be specied. It is immediately obvious that each behavior
strategy generates a mixed strategy in the sense of the above denition. It is
a much deeper insights into extensive form games that most often it is enough
to consider behavior strategies in the sense that each outcome induced by
a mixed strategy prole is also induced by some behavior strategy prole.
This is true as long as the game is one of perfect recall.
Definition 27 (Perfect Recall). An extensive form game T, ; A, ; I, ; H; u
i
;
has perfect recall if
i) i, h
i
H
i
, x, x

h
i
(x x

).
ii) If x

h(x

) and x p (x

) , then x h(x) such that x p (x

)
and all the actions on the path from x to x

coincide with the actions from


x to x

.
The rst requirement says intuitively that no player forgets his moves,
and the second says that no player forgets past knowledge.
Theorem 8 (Kuhns Theorem). If the extensive form game T, ; A, ; I, ; H; u
i
;
has perfect recall and an outcome (Z) is induced by and =
(
1
, ...,
N
) , then there is a behavior strategy prole b = (b
1
, ..., b
N
) such
that is induced by and b.
In other words, in games of perfect recall, there is no loss of generality
in restricting attention to behavior strategies.
The following example is due to Rubinstein and Piccione (see the special
issue of Games and Economic Behavior, July 1997, on games of imperfect
recall).
Example 7 (Absent Minded Driver). This is a single person game with
two decision nodes for the player. The player must decide whether to keep
on driving D on the highway or whether to exit E. If the driver chooses D
twice, he ends in a bad neighborhood and gets payo of 1. If he exits at
the rst node, he gets 0. M If he chooses D followed by E, then his payo
is 2.
The twist in the story is that the driver is absent minded. He cannot
remember if he has already decided or not in the past. This absent minded-
ness is represented by a game tree where both of the nodes are in the same
information set even though the two nodes are connected by an action by the
single player. Hence the game is not one of perfect recall as dened above.
3. SUBGAME PERFECT EQUILIBRIUM 33
It is a useful exercise to draw the game tree for this problem and consider
the various notions of strategies. The rst step in the analysis of any games
is to decide what the possible strategies are. As usual, a strategy should be
an assignment of a (possibly mixed) action to each information set. With
this denition for a strategy, it is easy to show that the optimal strategy in
this game cannot be a pure strategy. Observe also that if the driver were able
to commit to a mixed strategy, then she would choose to exit with probability
p =
2
3
resulting in a payo of
2
9
.
If the driver were to adopt the strategy of exiting with probability
2
3
,
then she would assign probabilities 3/4 and 1/4 respectively to the rst and
the second node in her information set. In order to calculate correctly the
payos to the driver, behavior at other nodes must be kept constant when
calculating the current payo to exiting or driving on. Thus choosing D and
then following the proposed strategy yields a payo of
3
4
(2
2
3

1
3
)
1
4
=
1
2
.
Choosing E yields similarly a payo of
1
2
.
The key to understanding the workings of the game lies in the inter-
pretation of a deviation from the proposed equilibrium strategy. Most game
theorists believe that in games of this type, a deviation should be modeled as
a single deviation away from the proposed strategy. In particular, this would
imply that a deviation can take place at a single node in the information set
without necessarily happening at the other. Piccione and Rubinstein oer a
dierent interpretation of the game, see the discussions in the issue of GEB
as cited above.
3. Subgame Perfect Equilibrium
Before we can dene the notion of a subgame perfect equilibrium we
need to know what a subgame is. The key feature of a subgame is that,
contemplated in isolation it forms its own game.
Definition 28 (Subform). A subform of an extensive form is a col-
lection of nodes

T T, together with ; A, ; 1, ; H all dened on the
restriction to

T, satisfying closure under succession and preservation of in-
formation sets:
if x

T, then s (x)

T, h(x)

T,
A proper subform is a subform

T consisting solely of some node x and its
successors.
Definition 29 (Subgame). Given a proper subform, the associated proper
subgame starts at x, with the payos restricted to

T Z, and the initial as-
sessment (x) = 1.
34 3. EXTENSIVE FORM GAMES
Definition 30 (Subgame Perfect Nash Equilibrium). A prole of strate-
gies
= (
1
, ...,
n
)
in an extensive form game
E
is a SPE if it induces a Nash equilibrium in
every subgame of
E
.
Clearly, every SPE is an NE, but the converse doesnt hold. Consider
rst the extensive form games with perfect information. Chess is a nite
extensive form game as is centipede game introduced above. What is the
unique subgame perfect equilibrium in this game?
It is a description of the equilibrium strategy for each player:
s

1
=
_
s, s

, s

_
, s

2
=
_
s, s

_
,
on and o the equilibrium path. The equilibrium path is the set of con-
nected edges which lead from an initial node to a nal node. The equilibrium
path is thus s and the equilibrium outcome is the payo realization
(1, 1). Thus a description of the equilibrium strategy asks for the behavior
of every player on and o the equilibrium path.
Theorem 9 (Zermelo, Kuhn). Every nite game of perfect information

E
has a pure strategy subgame perfect Nash equilibrium.
3.1. Finite Extensive Form Game: Bank-Run. Consider the fol-
lowing bank-run model with R > D > r > D/2. The interpretation is as
follows. The bank acts as an intermediary between lender and borrower.
Each lender deposit initially D. The bank also attempts to transform ma-
turity structures. The deposits are short-term, but the loan is long-term.
Suppose the deposits are applied towards an investment project which has a
return of 2r if the project is liquidated after one period and 2R if liquidated
after two periods. The return is 2D if the project is continued for more than
two terms. If either of the depositor asks for a return of its deposit, then
the project is liquidated, and the withdrawing depositor either gets D or
his share in liquidation proceeds, whichever is larger. Denote the rst stage
actions by w for withdrawing and n for not withdrawing. Let the second
period actions for the two players be W and N respectively. Thus the stage
payos are given by the two matrices There are two pure strategy subgame
w n
w r, r D, 2r D
n 2r D, D second period
Figure 1. First period
perfect equilibria of this game
s

1
= w, W , s

2
w, W
and
s

1
= n, W , s

2
n, W
4. THE LIMITS OF BACKWARD INDUCTION: ADDITIONAL MATERIAL 35
W N
W R, R 2R D, D
N D, 2R D D, D
Figure 2. Second period
Consider the following two-period example with imperfect information.
The rst period game is a prisoners dilemma game and the second game is
coordination game: and then Specify all subgame perfect equilibria in this

2, 2 1, 3
3, 1 0, 0
Figure 3. First period

x, x 0, 0
0, 0 x, x
Figure 4. Second period
game. Does the class of subgame perfect equilibria coincide with the class of
Nash equilibria? For x 2, cooperation is sustainable in a subgame perfect
equilibria by switching from the pure to the mixed strategy equilibrium in
the second stage game.
4. The Limits of Backward Induction: Additional Material
Consider the following example of game, referred to as Centipede Game.
There are two players each start with 1 dollar. They alternate saying stop
or continue, starting with player 1. If they say continue, 1 dollar is taken
from them and 2 dollar is given to the other player. As soon as either player
says stop the game is terminated and each player receives her current pile of
money. Alternatively, the game ends if both players pile reach 100 dollars:
(1) (2) (1) (2) (1) .... (2) (1) (100, 100)

(1, 1) (0, 3) (2, 2) (1, 4) (98, 101) (100, 100)
The rst entry in the payo vector refers to player 1, the second to player
2. Instead of looking at the true centipede game, let us look at a smaller
version of the game:
(1) c (2) c (1) c

(2) c

(1) c

(2, 4)
s s s

(1, 1) (0, 3) (2, 2) (1, 4) (3, 3)


36 3. EXTENSIVE FORM GAMES
What is a strategy for player 1 in this small centipede game. A strategy
is a complete description of actions at every information set (or history).
Player 1 has two histories at which he could conceivable make a choice
H
1
= , cc and likewise H
2
= c, ccc

. Thus a possible strategy for player


1 is
h
1
= h
1
= cc h
1
= ccc


s
1
= c , c

.
Another possible strategy is however:
h
1
= h
1
= cc h
1
= ccc


s
1
= s , c

.
The notion of a strategy in the extensive form game is stretched beyond
what we would call a plan. We saw in the centipede game that it indeed
requires a player to specify his action after histories that are impossible if he
carries out his plan. A dierent interpretation of the extensive form game
may help our understanding. We may think of each player having as many
agents (or multiple selves) as information sets. The strategy of each player
is to give instruction to each agents (which he may not be able to perfectly
control) as to what he should do if he is called to make a decision. All agents
of the same player have the same payo function. A plan is then a complete
set of instruction to every agent who acts on behalf of the player.
Notice that we did not discuss notions similar to rationalizability in the
context of extensive form games. The reason for this is that making sense
of what common knowledge of rationality means in extensive form games is
much more problematic than in normal form games. A simple example will
illustrate.
Robert Aumann has maintained that common knowledge of rationality
in games of perfect information implies backwards induction. Hence e.g.
the centipede game should have a unique solution where all players stop the
game at all nodes. Suppose that this is indeed the case. Then after observing
a move where the rst player continues the game, the second player ought
to conclude that the rst player might not be rational. But then it might
well be in the second players best interest to continue in the game. But
if this is the case, then it is not irrational for the rst player to continue
at the rst node. It should also be pointed out that knowledge and belief
with probability 1 are more or less the same in normal form games, but
in extensive form games these two notions yield very dierent conclusions.
Appropriate models of knowledge for extensive form games form a very
dicult part of interactive logic (or epistemic logic) and Chapter 14 in [3]
is a slightly outdated introduction into this area.
5. READING 37
5. Reading
The notion of extensive form game is rst (extensively) developed in
[7] and nicely developed in [5] and [6]. The relation between mixed and
behavior strategies is further discussed in [3], Ch.3.4.3.
CHAPTER 4
Repeated Games
1. Finitely Repeated Games
We rst briey consider games which are repeated nitely many time.
Then we move on the theory of innitely repeated games. We rst introduce
some notation. Consider the prisoners dilemma game: and repeat the stage
C D
C 4, 4 0, 5
D 5, 0 1, 1
Figure 1. Prisoners Dilemma
game T times. A pure strategy s
i
=
_
s
1
i
, ..., s
T
i
_
for a player i in the repeated
game is then a mapping from the history of the game H
t
into the strategies
in the stage game S
i
:
s
t
i
: H
t1
S
i
.
A complete specication of repeated game strategy has to suggest a strategy
choice after every possible history of the game, (and not only the histories
which will emerge in equilibrium). A complete specication can be thought
of as a complete set of instructions written before playing the game, so that
no matter what happens in the game, once it started a prescription of play
is found the set of instructions. The history of the game is formed by the
past observed strategy choices of the players:
H
t1
=
_
s
1
1
, ..., s
1
I
; ...; s
t1
1
, ..., s
t1
I
_
.
Definition 31. Given a stage game =
_
I, (S
i
)
iI
, (u
i
)
iI
_
, we denote
by (T) the nitely repeated game in which is played T times.
The outcomes of all preceding plays are observed before the next play
begins. The payos for (T) are simply the sums of the payos from the T
stage games.
Definition 32. In a repeated game (T) a players strategy species
the action the player will take in each stage, for each possible history of
play through the previous stage:
s
i
=
_
s
0
i
, ...., s
T
i
_
where:
s
t
i
: H
t1
S
i
.
39
40 4. REPEATED GAMES
Theorem 10. If the stage game has a unique Nash equilibrium then,
for any nite T, the repeated game (T) has a unique subgame-perfect out-
come: the Nash equilibrium of is played in every stage.
Corollary 11. Every combination of stage game Nash equilibria forms
a subgame perfect equilibrium.
Consider now the following game with multiple equilibria in the stage
game Can coordination among the players arise in a stage game?
C D E
C 4, 4 0, 5 3, 3
D 5, 0 1, 1 3, 3
E 3, 3 3, 3 -2,-2
Figure 2. Modied Prisoners Dilemma
Example 8. To understand the complexity of repeated games, even if
they are repeated only very few times, let us consider the prisoners dilemma
played three times. What are all possible histories of a game. How many
strategy choice have to be made in order for the repeated game strategy to be
a complete specication?
The history H
1
is naturally the empty set:
H
1
= .
The history H
0
is the set of all possible outcomes after the game has been
played once. Each outcome is uniquely dened by the (pure) strategy choices
leading to this outcome, thus
H
0
= CC, CD, DC, DD .
The cardinality of possible outcomes in period 0 is of course

H
0

= [S
1
[ [S
2
[ = 4.
We then observe that the number of possible histories grows rather rapidly.
In period 1, it is already

H
1

= [S
1
[ [S
2
[ [S
1
[ [S
2
[ = 16,
and in general it is

H
t

= ([S
1
[ [S
2
[)
t+1
.
However the complexity of strategies grows even faster. A complete strategy
for period t species a particular action for every possible history leading to
period t. Thus the number of contingent choices to be made by each player
in period t is given by

S
t
i

= [S
i
[
[H
t1
[
= [S
i
[
|S
1
||S
2
|
t+1
.
Notice that a strategy even in this two period game is not merely an instruc-
tion to act in a specic manner in the rst and second period, but needs to
specify the action after each contingency.
2. INFINITELY REPEATED GAMES 41
In order to sustain cooperation two conditions need to be satises: (i)
there are rewards from cooperation and (ii) there are credible strategies to
punish deviators. This theme of carrots and sticks is pervasive in the
entire theory of repeated games. As the punishment occurs in the future,
agents must value the future enough so that lower payos in the future
would indeed harm the agents welfare suciently. If the discount factor
represents the agents attitude towards the future, then higher discount fac-
tors make cooperation easier to achieve as lower payos in the future gure
more prominently in the agents calculation.
2. Innitely Repeated Games
Next we consider games which are repeated innitely often
Definition 33. Given a stage game , let () denote the innitely
repeated game in which is repeated forever and where each of the players
discounts future at discount factor < 1.
The payo in an innitely repeated game with stage payos
t
and
discount factor < 1 is given by

t=0

t
.
Definition 34. Given the discount factor , the average payo of the
innitely repeated sequence of payos
0
,
1
,
2
, ... is given by
(1 )

t=0

t
.
Notice that in innitely repeated games every subgame is identical to
the game itself.
2.1. The prisoners dilemma. Let us consider again the stage game
in gure 1 which represented the prisoners dilemma and consider the fol-
lowing strategy suggested to the players. I cooperate in the rst period
unconditionally and then I continue to cooperate if my opponent cooper-
ated in the last period. In case he defected in the last period, I will also
defect today in all future periods of the play. When can we sustain cooper-
ation? This strategy is often called the grim-trigger strategy, which can be
formally described by
s
i
t
=
_
C if H
t1
= CC, ...., CC or H
1
=
D if otherwise
We now want to show that this set of strategies form a subgame perfect
equilibrium. To prevent a deviation the following has to hold:
5 +

1

4
1
,
42 4. REPEATED GAMES
or
5 (1 ) + 4,
so that

1
4
.
Thus if the players are suciently patient, cooperation can be sustained in
this repeated game.
The equilibrium we suggested implies cooperation forever along the equi-
librium path. However, this is clearly not the only equilibrium. There are
many more, and many of them of very dierent characteristics. In the limit
as 1, one can show, and this is the content of the celebrated folk
theorems of the repeated games, that the entire convex hull of all individu-
ally rational payos of the stage game can be achieved as a subgame perfect
equilibrium. Notice, that this may involve very asymmetric payos for the
players:
3. Collusion among duopolists
The special feature of the prisoners dilemma game is that minimum
payo a player can guarantee himself in any stage, independent of the ac-
tions chosen by all other players, is equal to the payo in the stage game
Nash equilibrium. This minimum payo is called the reservation payo. It
constitutes the limit to the damage an opponent can inict on the player.
The individual rational payo v
i
is the lowest payo that the other players
can force upon player i:
v
i
= min
s
i
S
i
max
s
i
S
i
u
i
(s
i
, s
i
) .
3.1. Static Game. Consider the following Cournot oligopoly model
with prices determined by quantities as
p (q
1
, q
2
) = a q
1
q
2
.
The Cournot stage game equilibrium is found by solving the rst order
conditions of the prot functions, where we normalize the marginal cost of
production to zero,

i
(q
1
, q
2
) = q
i
(a q
1
q
2
) .
We saw earlier that the Cournot outcome is
q
C
i
=
a
3
,
and the associated prots are:

i
_
q
C
1
, q
C
2
_
=
a
2
9
.
In contrast, the strategy of the monopolist would be to maximize
(q) = q (a q) ,
3. COLLUSION AMONG DUOPOLISTS 43
which would result in the rst order conditions
a 2q = 0
and hence
q
M
=
a
2
and the prots are

_
q
M
_
=
M
=
a
2
4
.
3.2. Best Response. Suppose the two rms agree to each supply q to
the market, then the prots are given by:

i
(q, q) = q (a 2q) = aq 2q
2
.
By the best response function, the optimal response to any particular q is
(35) q

(q) =
a q
2
and the resulting prots are
(36)
1
(q

(q) , q) =
(a q)
2
4
.
where:
(a q)
2
4

_
aq 2q
2
_
=
1
4
(3q a)
2
0
and with equality at q =
a
3
, which is the Cournot output.
3.3. Collusion with trigger strategy. The discrepancy between mo-
nopoly and duopoly prot suggest that the duopolists may collude in the
market to achieve higher prices. How could they do it. Consider the follow-
ing strategy.
q
i
t
=
_
q
M
2
if h
t
=
__
q
M
2
,
q
M
2
_
, ....,
_
q
M
2
,
q
M
2
__
or h
t
=
q
C
otherwise

and consider whether the symmetric strategy can form a subgame perfect
equilibrium? We have to consider the alternatives available to the players.
Consider an history in which
h
t
=
__
q
M
2
,
q
M
2
_
, ....,
_
q
M
2
,
q
M
2
__
,
then the resulting payo would be
(37)
1
1
1
2

M

d
+

1

C
.
Thus to evaluate whether the player follow the equilibrium strategy, we have
to evaluate
d
.
44 4. REPEATED GAMES
3.3.1. SPE. This suggests the following trigger strategy equilibrium
(38)
1
1
1
2

M

d
+

1

C
,
which can be sustained for
9
17
. The question then arises whether we can
do better or more precisely whether there are punishment strategies which
would act even harsher on deviators. We still have to verify that following
a deviation, i.e. all histories dierent from collusive histories
(39)
1
1

C

d
+

1

C
,
but since q
i
= q
C
, (38) is trivially satised.
3.4. Collusion with two phase strategies. We next consider a re-
peated game where the equilibrium strategy consists of two dierent phases:
a maximal reward and a maximal punishment phase, a carrot and stick
strategy. The strategy may formally be described as follows:
q
i
t
=
_

_
q
M
2
if h
t
=
__
q
M
2
,
q
M
2
_
, ....,
_
q
M
2
,
q
M
2
__
q
M
2
if
_
q
i
t1
, q
j
t1
_
=
_
q, q
_
q if otherwise

Let x denote the level of output in the punishment phase (to be determined
as part of the equilibrium). We may then write the equilibrium conditions
as:
(40)
1
1
1
2

M

d
+
Punishment
..

_
ax 2x
2
_
+
Continuation
..

2
1
1
2

M
and
(41)
Punishment
..
_
ax 2x
2
_
+
Continuation
..

1
1
2

Deviation
..
(a x)
2
4
+
Punishment
..

_
ax 2x
2
_
+
Continuation
..

2
1
1
2

M
Equation (40) may be rewritten as,
(1 ) (1 +)
1
1
2

M

d
+
_
ax 2x
2
_
and nally as:
(42)
_
1
2

_
ax 2x
2
_
_

d

1
2

M
.
The inequality (42) suggests that the relative gains from deviating in con-
trast to collusion today must be smaller than the relative losses tomorrow
3. COLLUSION AMONG DUOPOLISTS 45
induced by a deviation today and again in contrast to continued collusion.
Similarly for the punishment to be credible, starting from (41), we get
(43)
_
ax 2x
2
_
+
1
2

(a x)
2
4
+
_
ax 2x
2
_
.
and which permits an intuitive interpretation in terms of todays and tomor-
rows payo, or alternatively as the one shot deviation strategy. By inserting
the value of
M
we obtain two equalities:

_
1
2
a
2
4

_
ax 2x
2
_
_
=
_
9
64
a
2

1
2
a
2
4
_

_
1
2
a
2
4

_
ax 2x
2
_
_
=
(ax)
2
4

_
ax 2x
2
_
which we can solve and obtain the following results as the harshest level of
punishment x =
5
12
a and the lowest level of discount factor =
9
32
for which
cooperation is achievable.
Reading: [4], Ch. 2.3 is a nice introduction in the theory of repeated
games. A far more advanced reading is [3], Ch.5. The characterization of
extremal equilibria is rst given in [1].
CHAPTER 5
Sequential Bargaining
The leading example in all of bargaining theory is the following trading
problem in the presence of bilateral monopoly power. A single seller and a
single buyer meet to agree on the terms of trade for a single unit of output.
The seller incurs a cost of c to produce the good while the buyer has a
value of v for the good. If the two parties have payo functions that are
quasilinear in money and the good, then there are positive gains from trade
if and only if v > c. The two parties have also an opportunity to refuse any
trade which yields them an outside option value of 0. Both parties prefer
(weakly) trading at price p to not trading at all if p satises v p c.
Which (if any) of such prices will be agreed upon when the two parties
negotiate?
The conventional wisdom up to late 70s was that the determination of
the price depends largely on psychological and sociological factors determin-
ing the bargaining power of the two parties and as a result economics had
little to say on the issue. When the concept of subgame perfect equilib-
rium was established as the standard solution concept in games of perfect
information, the bargaining issue was taken up again. Preliminary work was
done on bargaining in simple two period models, but the real breakthrough
was the innite horizon bargaining model with alternating oers proposed
by Ariel Rubinstein in [13]. There he showed that the innite horizon game
has a unique subgame perfect equilibrium and that the division of surplus
between the bargainers can be explained in terms of the time discount rates
of the bargainers.
It should be pointed out that other approaches to bargaining were also
adopted. An entirely dierent methodology starts by describing a set of
possible outcomes X that can be obtained in the bargaining process between
the two players and an alternative d that represent the outcome in the case
of no agreement. Each of the bargainers has preference relation _
i
dened
on X d. Let the set of possible preference relations of i be denoted by
R
i
. A solution to the bargaining problem is a function that associates an
outcome to each prole of preferences. In other words, the solution is a
function f such that
f : R
1
R
2
X d.
The method of analysis in this approach is often taken to be axiomatic. The
analyst decides in the abstract a set of criteria that a good solution concept
should satisfy. An example of such criteria could be Pareto eciency of
47
48 5. SEQUENTIAL BARGAINING
the solutions, i.e. if x ~
i
x

and x _
j
x

, then x

/ f (_
i
, _
j
) . Once the
criteria have been established, the analyst determines the set of functions f
that satisfy this criterion. Preferably this set is small (ideally a singleton)
so that some clear properties of the solutions can be established. This ap-
proach precedes the noncooperative approach to be followed in this handout.
The rst contribution to this cooperative or axiomatic approach to bargain-
ing was made by John Nash in The Bargaining Problem in Econometrica
1950. Various other solutions were proposed for two player and many player
bargaining models. A good survey of these developments can be found in
chapters 8-10 in Roger Myersons book Game Theory: Analysis of Con-
ict. Rubinsteins solution of the noncooperative bargaining game picks
the same solution as Nashs original cooperative solution for that game. A
recurring theme in the noncooperative bargaining literature has been the
identication of fully strategic models that yield the classical cooperative
bargaining solutions as their equilibrium outcomes.
1. Bargaining with Complete Information
In this section, we focus our attention on the case where two players
bargain over the division of 1 unit of surplus. (Sometimes more colorful
language is used and it is said that the players bargain over their shares
of a pie that has size of 1). We start by considering the following simple
extensive form to the bargaining procedure. There are just two stages in the
game. Player 1 proposes a sharing rule for the surplus we assume that all
real numbers between 0 and 1 represent feasible shares. Denote the shares
going to 1 and 2 by x and 1x respectively. Player 2 then sees the proposal
and decides whether to accept it or not. If the proposal is accepted, the
outcome in the game is a vector (x, 1 x) for the two players, if the oer
is rejected, the outcome is (0, 0) . It is easy to see that the only subgame
perfect equilibrium in this simple game results in a proposal x = 1 which is
accepted in equilibrium by player 2. In this case, we say that player 1 has
all the bargaining power because she is in the position of making a take-it-
or-leave-it oer to player 2 and as a result, player 1 gets the entire surplus
in the game.
Consider next the version of the game that extends over 2 periods. The
rst period is identical to the one described in the paragraph above except
that if player 2 refuses the rst period oer, the game moves into the second
period. The second period is identical to the rst, but now player 2 makes
the oer and 1 decides whether to accept or not. At the end of the second
stage, the payos are realized (obviously if player 2 accepts the rst oer,
then the payos are realized in period 1). If the players are completely
patient, i.e. their discount factor is 1, then it is clear that the subgame
perfect equilibria in the game have all the same outcomes. Since player 2
can refuse the rst period oer which put her in the position of player 1 in
the paragraph above, it must be that the equilibrium shares in the game are
1. BARGAINING WITH COMPLETE INFORMATION 49
(0, 1) . Hence we could say that player 2 has all the bargaining power since
she is the one in a position to make the last take it or leave it oer. The
situation changes somewhat if the players are less than perfectly patient.
Assume now that player i has a discount factor
i
between the periods in
the game. If player 2 refuses the rst oer and gets all of the surplus in
the second period, her payo from this strategy is
2
. Hence she will accept
any rst period oer x >
2
and refuse any oer x <
2
. At x =
2
, she is
indierent between accepting and rejecting the oer, but it is again easily
established that in the unique subgame perfect equilibrium of the game, she
will accept the oer x =
2
and the outcome of the game is (1
2
,
2
) .
Notice that the introduction of time discounting lessens the advantage from
being in the position to make the take it or leave it oer since the value of
that eventual surplus is diminished.
If there were 3 periods in the game, the unique equilibrium in the sub-
game that starts following a rejection of the rst oer would be the solution
of the 2 period game (by uniqueness of the subgame perfect equilibrium in
the 2 period game with discounting). Hence player 2 could secure a payo
of 1
1
from the second period onwards. As a result, player 1 must oer to
keep exactly 1
2
(1
1
) to himself in the rst period and the outcome is
thus (1
2
(1
1
) ,
2
(1
1
)) . At this point, we are ready to guess the
form of the equilibrium in a game with T periods.
Proposition 12. Let x
T
denote the unique subgame perfect equilibrium
share of player 1 in the bargaining game of length T. Then we have
x
T+2
= 1
2
_
1
1
x
T
_
.
Furthermore, the oer of x
T
is accepted immediately.
Proof. Consider the game with T +2 periods. In period T +1, player
2 must oer a share of
1
x
T
to player 1. This results in surplus
_
1
1
x
T
_
to player 2. Hence player 1 must oer a share
2
_
1
1
x
T
_
to player 2
in period T + 2 proving the rst claim. To see that in equilibrium, the
rst period oer must be accepted, assume to the contrary and denote the
equilibrium continuation payo vector by (v
1
, v
2
) after the refusal by player
2. Because of time discounting and feasibility v
1
+ v
2
< 1. But then it is
possible for 1 to oer x

such that x

> v
1
and (1 x

) > v
2
contradicting
the optimality of the original oer.
The proposition gives the equilibrium shares as a solution to a dierence
equation. The most interesting question about the shares is what happens as
the horizon becomes arbitrarily long, i.e. as T . Let x

= lim
T
x
T
.
Then it is easy to see that
x

=
1
2
1
1

2
.
Observe that in the case of identical discount factors, this share is monoton-
ically decreasing in and that (by using LHopitals rule) lim
1
x

() =
1
2
.
50 5. SEQUENTIAL BARGAINING
Hence the unique outcome in nite but long bargaining games with equally
patient players is the equal division of surplus.
Notice that nite games of this type always put special emphasis on the
eects of last stages. By the process of backwards induction, these eects
have implications on the earlier stages in the game as well. If the bargaining
can eectively take as much time as necessary, it is questionable if it is
appropriate to use a model which features endgame eects. The simplest
way around these eects is to assume that the bargaining process may take
innitely many periods. A simple argument based on the continuity of the
payo functions establishes that making the oers from the nite game in
all periods of the innite game continues to be an equilibrium in the innite
game. The more dicult claim to establish is the one showing that there
are no other subgame perfect equilibria in the game. In order to establish
this, we start by dening the strategies a bit more carefully.
The analysis here is taken from Fudenberg and Tirole, Chapter 4. The
innite horizon alternating oer bargaining game is a game with perfect in-
formation, i.e. at all stages, a single player moves and all of the previous
moves are observed by all players. In odd periods, player 1 makes the oer
and player 2 either accepts (thereby ending the game) or rejects and the
game moves to the next period. In even periods, the roles are reverse. De-
note by h
t
the vector of all past oers and acceptance decisions. A strategy
for player i is then a sequence of functions:
s
t
1
: h
t
[0, 1] for t odd,
s
t
1
: h
t
[0, 1] A, R for t even.
The strategies of player 2 are similarly dened.
We say that an action a
t
i
is conditionally dominated at history h
t
if
in the beginning of the subgame indexed by h
t
, every strategy that as-
signs positive probability to a
t
i
is strictly dominated. Iterated deletion of
conditionally dominated strategies removes in each round all conditionally
dominated strategies in all subgames given that the opponents strategies
have survived previous deletions. Observe that this process never deletes
subgame perfect equilibrium strategies. We will show that the alternating
oer bargaining game has a unique strategy prole that survives iterated
deletion of conditionally dominated strategies.
In what follows, we take each oer to be an oer for player 1s share in
the game. Observe rst that it is conditionally dominated for one to refuse
any oer from 2 that gives 1 a share exceeding
1
and similarly 2 must accept
any share for 1 that is below 1
2
. At the second round, 2 will never oer
more than
1
and 2 will reject all oers above 1
2
(1
1
) . Also 1 never
oers below 1
2
and never accepts oers below
1
(1
2
) .
Suppose then that after k iterations of the deletion process, 1 accepts
all oers above x
k
and 2 accepts all oers below y
k
where y
k
< x
k
. Then
after one more round of deletions, 2 never oers more than x
k
and rejects
2. DYNAMIC PROGRAMMING 51
all oers above 1
2
_
1 x
k
_
. Player 1 never oers below y
k
and rejects
all oers below
1
y
k
.
At the next round, player 1 must accept all oers above x
k+1

1
(1
2
)+

2
x
k
. To see this, consider the implications of refusing an oer by player
2. This has three possible implications. First, it may be that agreement
is never reached. The second possibility is that 2 accepts some future oer
from 1. This has a current value of at most
1
_
1
2
_
1 x
k
__
. The third
possibility is that player 1 accepts one of player 2s future oers. The pay-
o from this is at most
2
1
x
k
. The payo from the second possibility is the
largest of the three. Hence 1 must accept all oers above x
k+1
.
A similar argument shows that player 2 must accept all oers below
y
k+1
1
2
+
1

2
y
k
.
The sequences x
k
and y
k
are monotonic sequences with limits
x

=

1
(1
2
)
1
1

2
, y

=
1
2
1
1

2
.
Hence player 2 rejects any oer where 1s share is above y

=
1
2
1
1

2
and
accepts all shares below y

. Hence the unique outcome is again for 1 to


propose the share
_
1
2
1
1

2
, 1
1
2
1
1

2
_
and for 2 to accept immediately.
1.1. Extending the Basic Model. It is a relatively easy exercise to
show that if the surplus is divisible only in nitely many pieces, then any
sharing of the surplus can be realized in a subgame perfect equilibrium as
1. Again, the interpretation of this result relies on the view one takes on
modeling. If it is thought that the fact that monetary sums can be divided
only into pounds and pennies is important in the mind of the bargainers,
then this result is probably troublesome. If on the other hand it is thought
that this is not that important for the bargainers, then it is probably best
to use a model where the indivisibility is not present.
A more troublesome observation is that the result is somewhat sensitive
to the extensive form of the bargaining game. If the protocol of oers and
rejections is modied, then other outcomes can be supported in equilibrium
as well. Finally, the extensions to 3 or more player bargaining have been
unsuccessful. The uniqueness seems to be beyond the reach of the theory.
Probably the most successful models in this class have been the ones where
a random proposer suggests a sharing of the surplus in each period and the
acceptance is decided by majority vote. Models along these lines have been
developed by Baron and Ferejohn in the literature on political economy.
2. Dynamic Programming
How to extend the notion of backwards induction to an innite time
horizon. Consider rst a schematic decision problem in a decision tree and
observe what the process of backward reduction does to reduce the com-
plexity of the decision problem. It represents and reduces subtrees through
52 5. SEQUENTIAL BARGAINING
values, where the values represent the maximal value one can attain in any
subtree. When does this reduction lead to the optimal decision through the
entire tree: if and only if we replace each subtree with the optimal value.
Theorem 13. Every nite decision tree (with nitely many periods and
nitely may choices) has an optimal solution which can be obtained by back-
wards induction.
We can extend this theorem certainly to an interactive decision prob-
lem, provided that it maintains the structure of single person decision tree,
namely perfect information.
Theorem 14 (Zermelo, Kuhn). Every nite game of perfect information

E
has a pure strategy subgame perfect Nash equilibrium.
The theorem applies to many parlor games, in particular to chess. Thus
in principle we can apply backward induction to solve for a subgame per-
fect Nash equilibrium in chess. How does a computer program approach
the problem of evaluating a move? By approximating the value of a move
through a forward looking procedure which assesses the value of the posi-
tion in three or four moves from then on. We may apply the same idea of
approximation to every problem with an innite horizon. In fact, we can
obtain in many situations an exact approximation.
We now return to the bargaining problem as dened above. Two players,
1 and 2 bargain over a prize of value 1.. The rst player starts by making
an oer to the second player who can either accept or reject the oer. If he
rejects the oer, then in the next period he has the right to make an oer
which can again be accepted or rejected. The process of making oers then
starts again. The agents are impatient and have common discount factor of
= 1/ (1 +r) < 1.
Consider rst the game with an innite horizon. The surprising result and
the even more surprising simplicity of the argument is given rst. We can
adopt the denition of stationarity also for equilibrium strategies.
Definition 35 (Stationary strategies). A set of strategies:
_
s
t
i
_
I,
i=1,t=0
is stationary if
s
t
i
= s
i
, i, t.
Theorem 15. The unique subgame perfect equilibrium to the alternate
move bargaining game is given by for the oering party i to suggest
v =

1 +
in every period and for the receiving party i to always accept if v / (1 +)
and reject if v < /1 .
2. DYNAMIC PROGRAMMING 53
Notice that you have to specify the strategy for every period in the
innite time horizon model. The proof proceeds in two parts. First we
show that a stationary equilibrium exists, and then we show that it is the
unique subgame perfect equilibrium, stationary or not. To do the later part
of the argument, rst upper and lower bounds are obtained for each players
equilibrium payo. Then it is shown that the upper and lower bounds are
equal, which shows that an equilibrium exists and that it is unique.
Proof. We start with the stationary equilibrium and use directly the
payos and then later on derive the strategies which support these payo.
Suppose then that there is a stationary equilibrium. Then the following has
to be true for the person who accepts today:
v = (1 v) ,
which leaves us with
v =

1 +
.
Next we show that this is indeed the unique subgame perfect equilibrium.
Let us start in the rst period. Let v be the largest payo that player 1
gets in any SPE. By the symmetry, or stationarity of the bargaining game,
this is also the largest sum seller 2 can expect in the subgame which begins
after she refused the oer of player 1. By backwards induction the payo of
player 1 cannot be less than
v = 1 v,
or
(44) 1 = v + v.
Next we claim that v cannot be larger than 1 v if the oer is accepted in
the rst period. But what about making an oer that is rejected in the rst
period? As the second player gets at least v in the next period, this cannot
give the player more than (1 v) as he has to wait until the next period.
Hence
v 1 v.
Note that this implies that
(45) v 1 v = v + v v,
where the last equality has been obtained through (44). But rewriting (45)
we obtain
v (1 ) v (1 ) ,
which be the denition of the upper and lower bounds imply that v = v.
Let us denote this payo by v

. Since v

= 1 v

, we nd that player 1
must earn
v

=
1
1 +
,
54 5. SEQUENTIAL BARGAINING
and player 2 must earn
v

=

1 +
.
The derivation of the equilibrium strategies follows directly from here.
2.1. Additional Material: Capital Accumulation. Consider the
following investment problem. The return of a capital k
t
in period t is k
t
.
The capital stock depreciates in every period by , so that
k
t+1
= (1 ) k
t
+i
t
and can be replenished by an investment level i
t
. The cost of investment is
given by
c (i
t
) =
1
2
i
2
t
.
The value of the investment problem can therefore be written as
V (k
0
) max
{i
t
}

t=0
_

t=0

t
_
k
t

1
2
i
2
t
_
_
and likewise for all future periods
V (k

) max
{i
t
}

t=
_

t=0

(t)
_
k
t

1
2
i
2
t
_
_
Thus if we knew the values in the future say V (k
1
), we could certainly nd
the optimal policy for period 0, as we could formulate the following problem
(46) V (k
0
) max
i
0
_
k
0

1
2
i
2
0
+V ((1 ) k
0
+i
0
)
_
Supposing that V is dierentiable, we would get as a rst order condition
V

((1 ) k
0
+i
0
) i
0
= 0.
Consider now a stationary solution.
Definition 36 (Stationary). A policy (or strategy) is stationary if for
all t = 0, 1, ..., :
k
t
= k
In other words, the strategy is time invariant and clearly we have that
i
t
= i = k for all t.
Suppose there exists a k

such that
V (k

) =

t=0

t
_
k

1
2
(k

)
2
_
or
V (k

) =
k

1
2
(k

)
2
1
2. DYNAMIC PROGRAMMING 55
However, it still remains to verify what the optimal solution is. But we can
do this now by analyzing the rst order condition of the Bellman equation

1
k

= 0
Thus we get
k

.
CHAPTER 6
Games of Incomplete Information
1. Introduction
This is a very modest introduction into the theory of games with in-
complete information or Bayesian games. We start by dening incomplete
information, then suggest how to model incomplete information (as a move
by nature) to transform it into a game of imperfect information. We then
introduce the associated equilibrium concept of Bayesian Nash equilibrium.
1.1. Example. Harsanyis insight is illustrated by the following exam-
ple. Suppose payos of a two player two action game are either:

1,1 0,0
0,1 1,0
or

1,0 0,1
0,0 1,1
i.e. either player II has dominant strategy to play H or a dominant strategy
to play T. Suppose that II knows his own payos but player I thinks there
is probability p that payos are given by the rst matrix, probability 1 p
that they are given by the second matrix. Say that player II is of type 1 if
payos are given by the rst matrix, type 2 if payos are given by the second
matrix. Clearly equilibrium must have: II plays H if type 1, T if type 2; I
plays H if p >
1
2
, T if p <
1
2
. But how to analyze this problem in general?
All payo uncertainty can be captured by a single move of nature at the
beginning of the game.
2. Basics
A (nite) static incomplete information game consists of
Players 1, ..., I
Actions sets A
1
, ..., A
I
Sets of types T
1
, ..., T
I
A probability distribution over types p (T), where T = T
1

... T
I
Payo functions g
1
, ..., g
I
, each g
i
: AT R
57
58 6. GAMES OF INCOMPLETE INFORMATION
Interpretation. Nature chooses a prole of players types, t (t
1
, ..., t
I
)
T according to probability distribution p (.). Each player i observes his
own type t
i
and chooses an action a
i
. We also refer to t
i
as the private
information of agent i. Now player i receives payos g
i
(a, t). By introducing
ctional moves by nature, rst suggested by Harsanyi (1967) we transform
a game of incomplete information into a game of imperfect information. In
the transformed game one players incomplete information about the other
players type becomes imperfect information about natures moves.
2.1. Strategy. A pure strategy is a mapping s
i
: T
i
A
i
. Write S
i
for the set of such strategies, and let S = S
1
.. S
I
. A mixed strategy can
also depend on his private information:

i
: T
i
(S
i
).
2.2. Belief. A players type t
i
is any private information relevant
to the players decision making. The private information may refer to his
payo function, his belief about other agents payo function, the likelihood
of some relevant event, etc. We assume that for the players types t
i

I
i=1
there is some objective distribution function
p (t
1
, ..., t
I
)
where t
i
T
i
, which is called the prior distribution or prior belief. The
conditional probability
p
_
t

i
[t
i
_
=
p
_
t

i
, t
i
_

t
i
T
i
p (t
i
, t
i
)
by Bayes rule denotes the conditional probability over the other players
types give ones own type t
i
. This expression assumes that T
i
are nite.
Similar expressions hold for innite T
i
. If the players types are independent
then
p (t
i
[t
i
) = p
i
(t
i
) , t
i
T
i
.
2.3. Payos. The description of the Bayesian game is completed by
the payo function
g
i
(a
1
, ..., a
I
; t
1
, ..., t
I
) .
If player i knew the strategy of the other players as a function of his type then
he could calculate the expected payo from his decision given the conditional
probability p (t
i
[t
i
).
3. Bayesian Game
Definition 37. A Bayesian game in normal form is given by

N
=
_
1, (A
i
)
I
i=1
, (T
i
)
I
i=1
, p () , (g
i
(; ))
I
i=1
_
.
The timing of a static Bayesian game is then as follows:
(1) (a) nature draws a type vector t = (t
1
, ..., t
n
) , t
i
T
i
,
3. BAYESIAN GAME 59
(b) nature reveals t
i
to player i but not to any other player,
(c) players choose simultaneously actions s
i
S
i
,
(d) payos g
i
(a
1
, ..., a
I
; t
1
, ..., t
I
) are received.
Again, by imperfect information we mean that at some move in the game
the player with the move does not know the complete history of the game
thus far. Notice that with this specication, the only proper subgame of a
game of incomplete information is the whole game. Any Nash equilibrium
is then subgame perfect and the subgame perfection has no discriminating
power.
3.1. Bayesian Nash Equilibrium. Player is payo function of the
incomplete information game, u
i
: S R, is
u
i
(s)

tT
p (t) g
i
(s (t) , t)
where s = (s
1
, ..., s
I
) and s (t) = (s
1
(t
1
) , ..., s
I
(t
I
)). Recall:
(Old) Denition:: Strategy prole s

is a pure strategy Nash equi-


librium if
u
i
_
s

i
, s

i
_
u
i
_
s
i
, s

i
_
for all s
i
S
i
and i = 1, .., I
This can be re-written in the context of incomplete information as:
(47)

tT
p (t) g
i
__
s

i
(t
i
) , s

i
(t
i
)
_
, t
_


tT
p (t) g
i
__
s
i
(t
i
) , s

i
(t
i
)
_
, t
_
for all s
i
S
i
and i = 1, .., I
Writing p (t
i
) =

t

i
p
_
t
i
, t

i
_
and p (t
i
[t
i
)
p(t
i
,t
i
)
p(t
i
)
, this can be re-written
as:
(48)

t
i
T
i
p (t
i
[t
i
) g
i
__
s

i
(t
i
) , s

i
(t
i
)
_
, t
_


t
i
T
i
p (t
i
[t
i
) g
i
__
a
i
, s

i
(t
i
)
_
, t
_
for all t
i
T
i
, a
i
A
i
and i = 1, ..., I
Definition 38. A pure strategy Bayesian Nash equilibrium of
= A
1
, ..., A
I
; T
1
, ..., T
I
; u
1
, ..., u
I
; p
is a strategy prole s

= (s

1
, ..., s

I
) such that, for all i, for all t
i
T,
condition (48) is satised.
Notice the dierences in the denition between the Bayesian Nash equi-
librium and the Nash equilibrium. In the former the payo function may
depend on the type. Moreover the payo is calculated as an expected payo
given his own information, with the conditional distribution over all other
agents type.
60 6. GAMES OF INCOMPLETE INFORMATION
4. A Game of Incomplete Information: First Price Auction
Suppose there are two bidders for an object sold by an auctioneer and
the valuation of the object is private information. Each buyer i = 1, 2 has a
valuation v
i
| ([0, 1]) and submits a bid b
i
for the object. The valuations
are assumed to be statistically independent. If bidder i submits the higher
of the bids, buyer i receives v
i
b
i
and the seller receives b
i
.
u
i
(v
i
) =
_

_
v
i
b
i
, if b
i
b
i
,
1
2
(v
i
b
i
) , if b
i
= b
i
,
0, if b
i
< b
i
.
A Bayesian Nash equilibrium is then a strategy pair b
1
(v
1
) , b
2
(v
2
). The
optimal strategy is found for each player by solving the following problem
(49) max
b
i
(v
i
b
i
) Pr (b
i
> b
j
(v
j
)) +
1
2
(v
i
b
i
) Pr (b
i
= b
j
(v
j
)) .
We simplify the analysis by looking at linear equilibria:
b
i
= a
i
+c
i
v
i
.
Hence we can write the objective function in (49) as
max
b
i
(v
i
b
i
) Pr (b
i
a
j
+c
j
v
j
)
Notice that
(50) Pr (b
i
a
j
+c
j
v
j
) = Pr
_
v
j

b
i
a
j
c
j
_
,
and by the uniformity this results in
max
b
i
(v
i
b
i
)
b
i
a
j
c
j
.
The rst-order conditions for this problem result in
(51) b
i
(v
i
) = (v
i
+a
j
) /2.
Similarly and by symmetry we have
(52) b
j
(v
j
) = (v
j
+a
i
) /2.
But in equilibrium we must have
(v
i
+a
j
) /2 = a
i
+c
i
v
i
, v
i
,
and
(v
j
+a
i
) /2 = a
j
+c
j
v
j
, v
j
,
and hence
c
i
= c
j
=
1
2
,
and consequently
a
i
= a
j
= 0.
5. CONDITIONAL PROBABILITY AND CONDITIONAL EXPECTATION 61
Hence the equilibrium strategy is to systematically underbid
b
i
(v
i
) =
1
2
v
i
.
5. Conditional Probability and Conditional Expectation
Before we analyze the next game, the double auction, recall some basic
concepts in probability theory, namely conditional probability and condi-
tional expectation.
5.1. Discrete probabilities - nite events. We start with a nite
number of events, and then consider a continuum of events Consider the
following example. Suppose a buyer is faced with a lottery of prices
0 < p
1
< ... < p
i
< ... < p
n
<
where each price p
i
has an unconditional probability 0 <
i
< 1, and

n
i=1

i
= 1, by which the price p
i
is chosen. Suppose the buyer is will-
ing to accept all prices, then the expected price to be paid by the buyer
is
(53) E[p] =
n

i=1

i
p
i
,
which is the unconditional expectation of the price to be paid. Suppose
next, that the buyer adopts a rule to accepts all price higher or equal to
p
k
and reject all other prices. (Recall this is just an example.) Then the
unconditional expectation of the price to be paid is
(54) E[p] =
n

i=k

i
p
i
,
The expectation in (54) is still unconditional since no new information en-
tered in the formation of the expectation. Suppose we represent the accep-
tance rule of the buyer by the index k below the expectations operator E[],
as in E
k
[]. The expected payment is then
E
k
[p] =
k1

i=1

i
0 +
n

i=k

i
p
i
,
since all price strictly lower than p
k
are rejected. Clearly
E
1
[p] > E
k
[p]
as
(55) E
1
[p] E
k
[p] =
n

i=1

i
p
i

i=k

i
p
i
=
k1

i=1

i
p
i
> 0.
since the buyer is willing to pay p
i
in more circumstances in (53) and (54).
62 6. GAMES OF INCOMPLETE INFORMATION
Consider now the average price which is paid under the two dierent
policies conditional on accepting a price and paying it, in other words, we
ask for the conditional expectation:
E
k
[p [p p
k
]
We consider now exclusively the events, i.e. realization of prices p, where
p p
k
for some k. The conditional probability of a price p
i
being quoted
conditional on p
i
p
k
is then given by
(56) Pr (p
i
[p
i
p
k
) =
Pr (p
i
)
Pr (p
i
p
k
)
,
since
Pr (p
i
p
i
p
k
) = Pr (p
i
)
if indeed p
i
satises p
i
p
k
, which is the nothing else than Bayes rule.
1
Notice that we can write more explicitly as:
(58) Pr (p
i
[p
i
p
k
) =

i

n
j=k

j
,
and that
n

i=k
(p
i
[p
i
p
k
) = 1.
The conditional expectation of the prices, conditioned on accepting then
price, is then nothing else but taken the expectation with respect to the
conditional probabilities rather than with the unconditional probabilities,
so that
E
k
[p [p p
k
] =
n

i=k

n
j=k

j
p
i
,
where E
k
[p [p p
k
] is now the expected price paid conditional on accepting
the price. You may now verify that the expected price or average price paid,
conditional on paying is increasing in k, as we might have expected it.
Next we generalize the notion of conditional expectation to continuum of
prices where the prices are distributed according to a distribution function.
5.2. Densities - continuum of events. You will see that the gener-
alization is entirely obvious, but it may help to familiarize yourself with the
continuous case. Suppose a buyer is faced with a lottery of prices
p
_
p, p

where the distribution of prices is given by a (continuous) distribution func-


tion F (p) and an associated density function f (p). Again suppose the buyer
1
Recall that Bayes rule says for any two events, A and B, that
(57) Pr (A|B) =
Pr (A B)
Pr (B)
5. CONDITIONAL PROBABILITY AND CONDITIONAL EXPECTATION 63
is willing to accept all prices, then the expected price to be paid by the buyer
is
(59) E[p] =
_
p
p
pf (p) dp =
_
p
p
pdF (p)
which is the unconditional expectation of the price to be paid. Notice that
the summation is now replaced with the integral operation. Suppose next,
that the buyer adopts a rule to accept all prices higher or equal to p and re-
ject all other prices. (Recall this is just an example.) Then the unconditional
expectation of the price to be paid is
(60) E
p
[p] =
_
p
p
pdF (p)
Again, the expectation in (15) is still unconditional since no new information
entered in the formation of the expectation and we represent the acceptance
rule of the buyer by the index p below the expectations operator E[], as in
E
p
[]. The expected payment is then
E
p
[p] =
_
p
p
0dF (p) +
_
p
p
pdF (p)
since all price strictly lower than p are rejected. Clearly
E[p] > E
p
[p] , for all p > p.
as
(61) E[p] E
p
[p] =
_
p
p
pdF (p) > 0.
since the buyer is willing to pay p in more circumstances in (59) and (60).
Consider now the average price which is paid under the two dierent
policies conditional on accepting a price and paying it, in other words, we
ask for the conditional expectation:
E
p
[p [p p]
We consider now exclusively the events, i.e. realization of prices p, where
p p for some p. Notice that we have now a continuum of events, each of
which has probability zero, but all of which have associated densities. We
then switch to densities. The conditional densities of a price p being quoted
conditional on p p is then given by
(62) f (p [p p) =
f (p)
1 F ( p)
.
Compare this expression to (56) and (58) above and notice that this follows
from
p p p = p
64 6. GAMES OF INCOMPLETE INFORMATION
as long as p p, and otherwise the intersection is of course the empty set
Notice that, as before the conditional densities integrate up to 1 since
(63)
_
p
p
f (p [p p) dp =
_
p
p
f (p)
1 F ( p)
dp =
1
1 F ( p)
_
p
p
f (p) dp = 1,
where the last equality of course follows from
2
:
_
p
p
f (p) dp = 1 F ( p) .
Finally, the conditional expectation of the prices, conditioned on accepting
then price, is then nothing else but taken the expectation with respect to the
conditional probabilities rather than with the unconditional probabilities, so
that
E
p
[p [p p] =
_
p
p
pf (p [p p) dp =
_
p
p
p
f (p)
1 F ( p)
dp =
1
1 F ( p)
_
p
p
pf (p) dp.
Again you may verify that the expected price or average price paid, con-
ditional on paying is increasing in p, as we might have expected it. In
particular,
(64) lim
p p
E
p
[p [p p] = p
where in contrast, we have of course
(65) lim
p p
E
p
[p] = 0.
You may wish to verify (64) and (65) to realize the dierence between ex-
pected value (price) and conditional expected value (price) in the second
example.
6. Double Auction
6.1. Model. We present here another important trading situation where
both sides of the market have private information. The trading game is also
called a double auction. We assume that the sellers valuation is
(66) v
s
F (v
s
) = | ([0, 1]) ,
which we may think of the cost of producing one unit of the good. Buyers
valuation
(67) v
b
F (v
b
) = | ([0, 1]) .
The valuations v
s
and v
b
are independent. If sale occurs at p, buyer receives
v
b
p and seller receives p v
s
. Trade occurs at p =
1
2
(p
s
+p
b
) , if p
s
p
b
.
2
Recall the similarity between the second to last term in (27) to the term appearing
when computing the expected value of the bid p
s
(v
s
).
6. DOUBLE AUCTION 65
There is no trade if p
s
> p
b
. If there is no trade both, buyer and seller receive
zero payo. Given v
b
, p
b
and p
s
, the buyers payo is:
u
b
(v
b
, p
b
, p
s
) =
_
v
b

1
2
(p
s
+p
b
) if p
b
p
s
0 if p
b
< p
s
and similar for the seller
u
s
(v
s
, p
b
, p
s
) =
_
1
2
(p
s
+p
b
) v
s
if p
b
p
s
0 if p
b
< p
s
6.2. Equilibrium. A strategy for each agent is then a mapping
(68) p
i
: V
i
R.
Denote an equilibrium strategy by
p

i
(v
i
)
We can then dene a Bayes-Nash equilibrium as follows.
Definition 39. A Bayes-Nash equilibrium is given by
p

b
(v
b
) , p

s
(v
s
)
such that
(69) E
p

s
[u
b
(v
b
, p

b
(v
b
) , p

s
)] E
p

s
_
u
b
_
v
b
, p

b
, p

s
_
, v
b
, p

b
,
and
(70) E
p

b
[u
s
(v
s
, p

s
(v
s
) , p

b
)] E
p

b
_
u
s
_
v
s
, p

s
, p

b
_
, v
s
, p

s
.
Remark 3. The Bayes part in the denition of the Bayes-Nash equilib-
rium through (69) and (70) is that the buyer has to nd an optimal bid for
every value v
b
while he is uncertainty about the ask (p

s
) of the seller. From
the buyers point of view, he is uncertain about p

s
as the ask, p

s
= p

s
(v
s
)
will depend on the value the seller assigns to the object, but as this is the
private information of the seller, the buyer has to make his bid only know-
ing that the seller will adopt the equilibrium strategy: p

s
: V
s
[0, 1], but
not knowing the realization of v
s
and hence not knowing the realization of
p

s
= p

s
(v
s
).
6.3. Equilibrium Analysis. To nd the best strategy for the buyer,
and likewise for the seller, we have to able to evaluate the net utility of the
buyer for every possible bid he could make, given a particular value v
b
he
assigns to the object,
(71) E
p

s
[u
b
(v
b
, p
b
, p

s
)] =
_
{p

s
|p

s
p
b
}
_
v
b

1
2
(p
b
+p

s
)
_
dG

(p

s
) +
+
_
{p

s
|p

s
>p
b
}
0dG

(p

s
)
where
G

(x) = Pr (p

s
x)
66 6. GAMES OF INCOMPLETE INFORMATION
is the distribution of the equilibrium asks of the seller. Notice now that the
equilibrium ask, p

s
, of the seller will depend on the realization of the sellers
value v
s
.Thus we can equivalently, the expectation in (71) with respect to
v
s
, or
(72) E
v
s
[u
b
(v
b
, p
b
, p

s
(v
s
))] =
_
{v
s
|p

s
(v
s
)p
b
}
_
v
b

1
2
(p
b
+p

s
(v
s
))
_
dF (v
s
)
The advantage of the later operation is that we have direct information
about the distribution of values, whereas we have only indirect information
about the distribution of the equilibrium asks. Using (66), we get
(73) E
v
s
[u
b
(v
b
, p
b
, p

s
(v
s
))] =
_
{v
s
|p

s
(v
s
)p
b
}
_
v
b

1
2
(p
b
+p

s
(v
s
))
_
dv
s
.
However, as we dont know the exact relationship between v
s
and p

s
, we cant
take the expectation yet. Similar to the auction example, we therefore make
the following guess, which we shall verify eventually about the relationship
of v
s
and p

s
:
(74) p

s
(v
s
) = a
s
+c
s
v
s
.
Inserting (74) into (73) leads us to
(75)
_
{v
s
|p

s
(v
s
)p
b
}
_
v
b

1
2
(p
b
+a
s
+c
s
v
s
)
_
dv
s
.
The nal question then is what is the set of v
s
over which we have to inte-
grate, or
v
s
[p

s
(v
s
) p
b
= v
s
[a
s
+c
s
v
s
p
b

=
_
v
s

v
s

p
b
a
s
c
s
_
and hence (75) is
_
p
b
a
s
c
s
0
_
v
b

1
2
(p
b
+a
s
+c
s
v
s
)
_
dv
s
Integrating yields
_
v
b
v
s

1
2
_
p
b
v
s
+a
s
v
s
+
1
2
c
s
v
2
s
__
p
b
a
s
c
s
0
or
(76) E
v
s
[u
b
(v
b
, p
b
, p

s
(v
s
))] =
p
b
a
s
c
s
_
v
b

1
2
_
p
b
+a
s
+
1
2
c
s
p
b
a
s
c
s
__
The expected utility for the buyer is now expressed as a function of v
b
, p
b
and the constants in the strategy of the seller, but independent of p

s
or v
s
.
6. DOUBLE AUCTION 67
Simplifying, we get
(77) u
b
(v
b
, p
b
, a
s
, c
s
) =
probability of trade
..
p
b
a
s
c
s

expected net utility conditional on trade


..
_
v
b

1
2
_
3
2
p
b
+
1
2
a
s
__
,
which essentially represents the trade-o for the buyer. A higher bid p
b
increases the probability of trade but decreases the revenue conditional on
trade. The rst order conditions are
u
b
(v
b
, p
b
, a
s
, c
s
)
p
b
= v
b

_
3p
b
+a
s
4
_

3
4
(p
b
a
s
) = 0
or
(78) p

b
(v
b
) =
2v
b
3
+
a
s
3
.
We can perform a similar steps of arguments for the seller to obtain
_
1
p
s
a
b
c
b
_
1
2
(p
s
+a
b
+c
b
v
b
) v
s
_
dv
b
and we get the sellers payo as a function of v
s
, p
s
, a
b
, c
b
:
u
s
(v
s
, p
s
, a
b
, c
b
) =
_
1
2
_
p
s
v
b
+a
b
v
b
+
1
2
c
b
v
2
b
_
v
s
v
b
_
1
p
s
a
b
c
b
and when evaluated it give us
u
s
(v
s
, p
s
, a
b
, c
b
) =
probability of trade
..
c
b
p
s
+a
b
c
b

expected net utility conditional on trade


..
1
4
(3p
s
+c
b
+a
b
4v
s
)
where the trade-o is now such that a higher ask leads to lower probability
of trade, but a higher net utility conditional on trade. Taking the rst order
conditions, leads us to
u
s
(v
s
, p
s
, a
b
, c
b
)
p
s
= 0
or equivalently
(79) p

s
(v
s
) =
1
3
c
b
+
1
3
a
b
+
2
3
v
s
From the solution of the two pricing policies, we can now solve for a
b
and a
s
as well as
1
3
c
b
+
1
3
a
b
= a
s
a
s
3
= a
b
and hence
a
s
=
1
4
, a
b
=
1
12
68 6. GAMES OF INCOMPLETE INFORMATION
and thus the complete resolution is
p

b
(v
b
) =
1
12
+
2
3
v
b
and
p

s
(v
s
) =
1
4
+
2
3
v
s
Next we can evaluate the eciency of the equilibrium. Since trade occurs if
p

b
(v
b
) p

s
(v
s
)
1
12
+
2
3
v
b

1
4
+
2
3
v
s
v
b
v
s

1
4
trade occurs if the gains from trade exceed
1
4
. The following strategies also
form an equilibrium for any x [0, 1] .
p
b
=
_
x if v
b
x
0 otherwise
p
s
=
_
x if v
s
x
1 otherwise
.
Readings. Chapter 3 in [4], [3] Chapter 6, [8], Chapter 8.E.
CHAPTER 7
Adverse selection (with two types)
Often also called self-selection, or screening. In insurance economics,
if a insurance company oers a tari tailored to the average population, the
tari will only be accepted by those with higher than average risk.
1. Monopolistic Price Discrimination
A simple model of wine merchant and wine buyer, who could either have
a coarse or a sophisticated taste, which is unobservable to the merchant.
What qualities should the merchant oer and at what price?
The model is given by the utility function of the buyer, which is
(80) v (
i
, q
i
, t
i
) = u(
i
, q
i
) t
i
=
i
q
i
t
i
, i l, h
where
i
represent the marginal willingness to pay for quality q
i
and t
i
is the
transfer (price) buyer i has to pay for the quality q
i
. The taste parameters

i
satises
(81) 0 <
l
<
h
< .
The cost of producing quality q 0 is given by
(82) c (q) 0, c

(q) > 0, c

(q) > 0.
The ex-ante (prior) probability that the buyer has a high willingness to pay
is given by
p = Pr (
i
=
h
)
We also observe that the dierence in utility for the high and low valuation
buyer for any given quality q
u(
h
, q) u(
l
, q)
is increasing in q. (This is know as the Spence-Mirrlees sorting condition.).
If the taste parameter
i
were a continuous variable, the sorting condition
could be written in terms of the second cross derivative:

2
u(, q)
q
> 0,
which states that taste and quality q are complements. The prot for the
seller from a bundle (q, t) is given by
(t, q) = t c (q)
69
70 7. ADVERSE SELECTION (WITH TWO TYPES)
1.1. First Best. Consider rst the nature of the socially optimal so-
lution. As dierent types have dierent preferences, they should consume
dierent qualities. The social surplus for each type can be maximized sepa-
rately by solving
max
q
i

i
q
i
c (q
i
)
and the rst order conditions yield:
q
i
= q

i
c

(q

i
) =
i
q

l
< q

h
.
The ecient solution is the equilibrium outcome if either the monopolist
can perfectly discriminate between the types (rst degree price discrimina-
tion) or if there is perfect competition. The two outcomes dier only in
terms of the distribution of the social surplus. With a perfectly discriminat-
ing monopolist, the monopolist sets
(83) t
i
=
i
q
i
and then solves for each type separately:
max
{t
i
,q
i
}
(t
i
, q
i
) max
{t
i
,q
i
}

i
q
i
c (q
i
) ,
using (83). Likewise with perfect competition, the sellers will break even,
get zero prot and set prices at
t
i
= c (q

i
)
in which case the buyer will get all the surplus.
1.2. Second Best: Asymmetric information. Consider next the
situation under asymmetric information. It is veried immediately that
perfect discrimination is now impossible as
(84)
h
q

l
t

l
= (
h

l
) q

l
> 0 =

h
q

h
t
h
but sorting is possible. The problem for the monopolist is now
(85) max
{t
l
,q
l
,t
h
,q
h
}
(1 ) t
l
c (q
l
) + (t
h
(c (q
h
)))
subject to the individual rationality constraint for every type
(86)
i
q
i
t
i
0 (IR
i
)
and the incentive compatibility constraint
(87)
i
q
i
t
i

i
q
j
t
j
(IC
i
)
The question is then how to separate. We will show that the binding con-
straint are IR
l
and IC
h
, whereas the remaining constraints are not binding.
We then solve for t
l
and t
h
, which in turn allows us to solve for q
h
, and
leaves us with an unconstrained problem for q
l
.Thus we want to show
(88) (i) IR
l
binding, (ii) IC
h
binding, (iii) q
h
q
l
(iv) q
h
= q

h
1. MONOPOLISTIC PRICE DISCRIMINATION 71
Consider (i). We argue by contradiction. As
(89)
h
q
h
t
h

IC
h

h
q
l
t
l

h
>
l

l
q
l
t
l
suppose that
l
q
l
t
l
> 0, then we could increase t
l
, t
h
by a equal amount,
satisfy all the constraints and increase the prots of the seller. Contradic-
tion.
Consider (ii) Suppose not, then as
(90)
h
q
h
t
h
>
h
q
l
t
l

h
>
l

l
q
l
t
l
(IR
l
)
= 0
and thus t
h
could be increased, again increasing the prot of the seller.
(iii) Adding up the incentive constraints gives us (IC
l
) + (IC
l
)
(91)
h
(q
h
q
l
)
l
(q
h
q
l
)
and since:
(92)
h
>
l
q
h
q
l
0.
Next we show that IC
l
can be neglected as
(93) t
h
t
l
=
h
(q
h
q
l
)
l
(q
h
q
l
) .
This allows to say that the equilibrium transfers are going to be
(94) t
l
=
l
q
l
and
t
h
t
L
=
h
(q
h
q
L
) t
h
=
h
(q
h
q
L
) +
l
q
l
.
Using the transfer, it is immediate that
q
h
= q

h
and we can solve for the last remaining variable, q
l
.
max
q
l
(1 p) (
l
q
l
(c (q
l
)) +p (
h
(q

h
q
L
) +
l
q
l
c (q

h
)))
but as q

h
is just as constant, the optimal solution is independent of constant
terms and we can simplify the expression to:
max
q
l
(1 p) (
l
q
l
c (q
l
)) p (
h

l
) q
l

Dividing by (1 p) we get
max
q
l
_

l
q
l
c (q
l
)
p
1 p
(
h

l
) q
l
_
for which the rst order conditions are

l
c

(q
l
)
p
1 p
(
h

l
) q
l
= 0
This immediately implies that the solution q
l
:
c

( q
l
) <
l
q
l
< q

l
and the quality supply to the low valuation buyer is ineciently low (with
the possibility of complete exclusion).
72 7. ADVERSE SELECTION (WITH TWO TYPES)
Consider next the information rent for the high valuation buyer, it is
I (q
l
) = (
h

l
) q
l
and therefore the rent is increasing in q
l
which is the motivation for the
seller to depress the quality supply to the low end of the market.
The material is also covered in [14], Chapter 2.
CHAPTER 8
Theoretical Complements
1. Mixed Strategy Bayes Nash Equilibrium
Definition 40. A Bayesian Nash equilibrium of
= S
1
, ..., S
I
; T
1
, ..., T
I
; u
1
, ..., u
I
; p
is a strategy prole

= (

1
, ...,

I
) such that, for all i, for all t
i
T and
all s
i
supp

i
(t
i
):
Bayes
..

t
i
T
i
p
i
(t
i
[t
i
)
mixed strategy Nash
..
_
_

s
i
S
i
u
i
__
s

i
, s
i
_
; (t
i
, t
i
)
_

i
(s
i
[t
i
)
_
_

Bayes
..

t
i
T
i
p
i
(t
i
[t
i
)
mixed strategy Nash
..
_
_

s
i
S
i
u
i
__
s

i
, s
i
_
; (t
i
, t
i
)
_

i
(s
i
[t
i
)
_
_
, s

i
where

i
(s
i
[t
i
)
_

1
(s
1
[t
1
) , ...,

i1
(s
i1
[t
i1
) ,

i+1
(s
i+1
[t
i+1
) , ...,

n
(s
n
[t
n
)
_
.
Since the strategy choices are independent across players, we can write
that:

i
(s
i
[t
i
) =

j=i

j
(s
j
[t
j
) .
2. Sender-Receiver Games
2.1. Model. A signaling games is a dynamic game of incomplete infor-
mation involving two players: a sender (S) and receiver (R). The sender is
the informed party, the receiver the uniformed party. (Describe the game in
its extensive form). The timing is as follows:
(1) nature draws a type t
i
T = t
1
, ..., t
I
according to a probability
distribution p (t
i
) 0,

i
p (t
i
) = 1,
(2) sender observes t
i
and chooses a message m
j
M = m
1
, ..., m
J
,
(3) receiver observes m
j
(but not t
i
) and chooses an action a
k
A =
a
1
, ..., a
K
,
(4) payos are given by u
S
(t
i
, m
j
, a
k
) and u
R
(t
i
, m
j
, a
k
).
73
74 8. THEORETICAL COMPLEMENTS
3. Perfect Bayesian Equilibrium
3.1. Informal Notion. The notion of a perfect Bayesian equilibrium
is then given by:
Condition 16. At each information set, the player with the move must
have a belief about which node in the information set has been reached by
the play of the game. For a non singleton information set, a belief is a
probability distribution over the nodes in the information set; for a singleton
information set, the players belief puts probability one on the single decision
node.
Condition 17. Given their beliefs, the players strategies must be se-
quentially rational. That is, at each information set the action taken by the
player with the move (and the players subsequent strategy) must be optimal
given the players belief at that information set and the other players subse-
quent strategies (where a subsequent strategy is a complete plan of action
covering every contingency that might arise after the given information set
has been reached).
Definition 41. For a given equilibrium in a given extensive-form game,
an information set is on the equilibrium path if it will be reached with positive
probability if the game is played according to the equilibrium strategies, and
is o the equilibrium path if it is certain not to be reached if the game is
played according to the equilibrium strategies (where equilibrium can mean
Nash, subgame-perfect, Bayesian, or perfect Bayesian equilibrium).
Condition 18. At information sets on the equilibrium path, beliefs are
determined by Bayes rule and the players equilibrium strategies.
Condition 19. At information sets o the equilibrium path beliefs are
determined by Bayes rule and the players equilibrium strategies where pos-
sible.
Definition 42. A perfect Bayesian equilibrium consists of strategies and
beliefs satisfying Requirements 1 through 4.
3.2. Formal Denition.
Condition 20. After observing any message m
j
from M, the Receiver
must have a belief about which types could have sent m
j
. Denote this belief
by the probability distribution (t
i
[m
j
), where (t
i
[m
j
) 0 for each t
i
in
T and
(95)

t
i
T
(t
i
[m
j
) = 1
Condition 21.
3. PERFECT BAYESIAN EQUILIBRIUM 75
(1) (a) For each m
j
in M, the Receivers action a

(m
j
) must maximize
the Receivers expected utility, given the belief (t
i
[m
j
) about
which types could have sent m
j
. That is, a

(m
j
) solves
(96) max
a
k
A

(t
i
[m
j
)
t
i
T
U
R
(t
i
, m
j
, a
k
) .
(2) For each t
i
in T, Senders message m

(t
i
) must maximize the Senders
utility, given the Receivers strategy a

(m
j
) . That is, m

(t
i
) solves:
(97) max
m
j
M
U
s
(t
i
, m
j
, a

(m
j
))
Condition 22. For each m
j
in M, if exists t
i
in T such that m

(t
i
) =
m
j
, then the Receivers belief at the information set corresponding to m
j
must follow from Bayes rule and the Senders strategy:
(98) (t
i
[m
j
) =
p (t
i
)

t
i
T
p (t
i
)
.
We can summarize these results to get the denition of a Perfect Bayesian
Equilibrium:
Definition 43. A pure-strategy perfect Bayesian equilibrium in a sig-
naling game is a pair of strategies m

(t
i
) and a

(m
j
) and a belief (t
i
[m
j
)
satisfying signaling requirements (20)-(22).
Alternatively, we can state it more succinctly as saying that:
Definition 44 (PBE). A pure strategy Perfect Bayesian Equilibrium
is a set of strategies m

(t
i
) , a

(m
j
) and posterior beliefs p

(t
i
[m
j
) such
that:
(1) t
i
, p

(t
i
[m
j
) , s.th. p

(t
i
[m
j
) 0, and

t
i
T
p

(t
i
[m
j
) = 1;
(2) m
j
, a

(m
j
) arg max

t
i
T
u
R
(t
i
, m
j
, a
k
) p

(t
i
[m
j
) ;
(3) t
i
, m

(t
i
) arg max u
S
(t
i
, m
j
, a

(m
j
))
(4) m
j
if t
i
T s.th. at m

(t
i
) = m
j
, then:
p (t
i
[m
j
) =
p (t
i
)

i
[m

(t

i
)=m
j
p (t

i
)
.
We can now introduce additional examples to rene the equilibrium set.
Example 9. Job Market Signalling
Example 10. Cho-Kreps (Beer-Quiche) - Entry Deterrence. Remark
the payos in [4] have a problem, they dont allow for a separating equilib-
rium, modify)
Definition 45 (Equilibrium Domination). Given a PBE, the message
m
j
is equilibrium-dominated for type t
i
, if
U

S
(t
i
) > max
a
k
A
U
S
(t
i
, m
j
, a
k
) .
76 8. THEORETICAL COMPLEMENTS
Definition 46 (Intuitive Criterion). If the information set following m
j
is o the equilibrium path and m
j
equilibrium dominated for type t
i
, then
(t
i
[m
j
) = 0.
This is possible provided that m
j
is not equilibrium dominated for all
types in T.
Readings: [8] Chapter 9.C, [3] Chapter 8.
Bibliography
[1] D. Abreu. Extremal equilibria of oligopolistic supergames. Journal of Economic The-
ory, 39:191225, 1986.
[2] K. Binmore. Fun and Games. D.C. Heath, Lexington, 1992.
[3] D. Fudenberg and J. Tirole. Game Theory. MIT Press, Cambridge, 1991.
[4] R. Gibbons. Game Theory for Applied Economists. Princeton University Press,
Princeton, 1992.
[5] D. M. Kreps. A Course in Microeconomic Theory. Princeton University Press, Prince-
ton, 1990.
[6] D.M. Kreps and R. Wilson. Sequential equilibria. Econometrica, 50:863894, 1982.
[7] H. Kuhn. Extensive games and the problem of information. In H. Kuhn and A. Tucker,
editors, Contributions to the Theory of Games, pages 193216. Princeton University
Press, Princeton, 1953.
[8] A. Mas-Collel, M.D. Whinston, and J.R. Green. Microeconomic Theory. Oxford Uni-
versity Press, Oxford, 1995.
[9] J. Nash. Equilibrium points in n-person games. Proceedings of the National Academy
of Sciences, 36:4849, 1950.
[10] M.J. Osborne and A. Rubinstein. A Course in Game Theory. MIT Press, Cambridge,
1994.
[11] P. Reny. On t he existence of pure and mixed strategy nash equilibria in discontinuous
games. Econometrica, 67:10291056, 1999.
[12] S.M. Ross. Probability Models. Academic Press, San Diego, 1993.
[13] A. Rubinstein. Perfect equilibrium in a bargaining model. Econometrica, 50:97109,
1982.
[14] B. Salanie. The Economics of Contracts. MIT Press, Cambridge, 1997.
77

S-ar putea să vă placă și