Bayesian Games
Abdul Quadir
XLRI
7 January, 2020
Introduction
I Game with incomplete information
I Bayes-Nash equilibrium
I Screening and signaling
Reading: Dixit and Skeath, Chapter 8.
Example: Quantity Competition with Incomplete
Information
I Two firms produce an identical good at constant marginal cost
I Inverse demand is P(Q) = α − Q, where Q = q1 + q2 is the
total quantity.
I Firm 1 has marginal cost c that is common knowledge.
I Firm 2’s marginal cost is private information: cl with
probability θ and ch with probability 1 − θ.
I Game has two players and two states (l and h), actions
qi ∈ [0, ∞).
I Payoffs for the two players are
u1 (q1 , q2 , t) = qi (P(q1 + q2 ) − c)
u2 (q1 , q2 , t) = q2 (P(q1 + q2 ) − ct )
where t ∈ {l, h} is firm’s 2type.
Example: Quantity Competition With Incomplete
Information
I we can think of strategy profile a triple (q1∗ , ql∗ , qh∗ ).
I We can get the three best response functionas
B1 (ql , qh ) = arg max q1 (θP(q1 + ql ) + (1 − θ)(P(q1 + qh ) − c)
q1 ≥0
Bl (q1 ) = arg max ql (P(q1 + ql ) − cl )
ql ≥0
Bh (q1 ) = arg max qh (P(q1 + qh ) − ch )
qh ≥0
I Bayes Nash equilibria are triples (q1∗ , ql∗ , qh∗ ) such that
B1 (ql∗ , qh∗ ) = q1∗ , Bl (q1∗ ) = ql∗ , Bh (q1∗ ) = qh∗ .
π1 = q1 [θ(α − (q1 + ql ) + (1 − θ)(α − (q1 + qh )) − c)
πl = ql (α − (q1 + ql ) − cl )
πh = qh (α − (q1 + qh ) − ch )
Example: Quantity Competition With Incomplete
Information
I we can think of strategy profile a triple (q1∗ , ql∗ , qh∗ ).
I We can get the three best response functionas
B1 (ql , qh ) = arg max q1 (θP(q1 + ql ) + (1 − θ)(P(q1 + qh ) − c)
q1 ≥0
Bl (q1 ) = arg max ql (P(q1 + ql ) − cl )
ql ≥0
Bh (q1 ) = arg max qh (P(q1 + qh ) − ch )
qh ≥0
I Bayes Nash equilibria are triples (q1∗ , ql∗ , qh∗ ) such that
B1 (ql∗ , qh∗ ) = q1∗ , Bl (q1∗ ) = ql∗ , Bh (q1∗ ) = qh∗ .
π1 = q1 [θ(α − (q1 + ql ) + (1 − θ)(α − (q1 + qh )) − c)
πl = ql (α − (q1 + ql ) − cl )
πh = qh (α − (q1 + qh ) − ch )
Example: Quantity Competition With Incomplete
Information
I we can think of strategy profile a triple (q1∗ , ql∗ , qh∗ ).
I We can get the three best response functionas
B1 (ql , qh ) = arg max q1 (θP(q1 + ql ) + (1 − θ)(P(q1 + qh ) − c)
q1 ≥0
Bl (q1 ) = arg max ql (P(q1 + ql ) − cl )
ql ≥0
Bh (q1 ) = arg max qh (P(q1 + qh ) − ch )
qh ≥0
I Bayes Nash equilibria are triples (q1∗ , ql∗ , qh∗ ) such that
B1 (ql∗ , qh∗ ) = q1∗ , Bl (q1∗ ) = ql∗ , Bh (q1∗ ) = qh∗ .
π1 = q1 [θ(α − (q1 + ql ) + (1 − θ)(α − (q1 + qh )) − c)
πl = ql (α − (q1 + ql ) − cl )
πh = qh (α − (q1 + qh ) − ch )
Example: Quantity Competition with Incomplete
Information
I Therefore, the three best response function are given as:
1
q1 (ql , qh ) = [θ(α − ql ) + (1 − θ)(α − qh ) − c]
2
1
ql (q1 ) = (α − q1 − cl )
2
1
qh (q1 ) = (α − q1 − ch )
2
I Therefore, Bayes Nash equilibrium is
1
q1∗ = (α − 2c + θcl + (1 − θ)ch )
3
1 1−θ
ql∗ = (α − 2cl + c) − (ch − cl )
3 6
1 θ
qh∗ = (α − 2ch + c) + (ch − cl )
3 6
I Note ql∗ > qh∗ , type with lower cost produces more.
Example: Quantity Competition with Incomplete
Information
I Therefore, the three best response function are given as:
1
q1 (ql , qh ) = [θ(α − ql ) + (1 − θ)(α − qh ) − c]
2
1
ql (q1 ) = (α − q1 − cl )
2
1
qh (q1 ) = (α − q1 − ch )
2
I Therefore, Bayes Nash equilibrium is
1
q1∗ = (α − 2c + θcl + (1 − θ)ch )
3
1 1−θ
ql∗ = (α − 2cl + c) − (ch − cl )
3 6
1 θ
qh∗ = (α − 2ch + c) + (ch − cl )
3 6
I Note ql∗ > qh∗ , type with lower cost produces more.
Example: Quantity Competition with Incomplete
Information
I In the game with complete information, the Nash equilibrium
involves players producing
1
qi = (α − 2ci + cj )
3
I With incomplete information, firm 2 produces more than this
when its cost is ch and less when its cost is cl .
I This is because firm 1 produces a moderated output.
I When firm 2 has cost ch , firm 1 produces less than it would if
it knew ch , so firm 2 gets to produce a bit more.
I When firm 2 has cost cl , firm 1 produces more than it would
if it knew cl , so firm 2 gets to produce a bit less.
Example: Quantity Competition with Incomplete
Information
I In the game with complete information, the Nash equilibrium
involves players producing
1
qi = (α − 2ci + cj )
3
I With incomplete information, firm 2 produces more than this
when its cost is ch and less when its cost is cl .
I This is because firm 1 produces a moderated output.
I When firm 2 has cost ch , firm 1 produces less than it would if
it knew ch , so firm 2 gets to produce a bit more.
I When firm 2 has cost cl , firm 1 produces more than it would
if it knew cl , so firm 2 gets to produce a bit less.
Example: Market for Used Cars
I Consider a used car market where a car is good with
probability q ∈ [0, 1] and bad with probability (1 − q).
I There is one buyer and one seller.
I The seller knows the types of car but buyer does not know.
I There is a price for every used car p.
I The probability q is common knowledge.
I The seller has to possible actions: SELL and NOT SELL.
I The buyer has two possible actions: BUY and NOT BUY.
I If the car is good, then the buyer enjoys a value of 6 and the
seller enjoys a value of 5.
I If the car is bad, the buyer enjoys a value of 4 and the seller
has zero value for it.
I Note that in both the cases it is “efficient” to trade the car.
Example: Market for Used Cars
I What is the Bayesian game?
I The nature informs the seller if the car is good or bad.
I If the car is good, then the following game will be played:
Sell Not Sell
Buy 6 − p, p 0, 5
Not Buy 0, 5 0, 5
I If the car is bad, then the following game will be played:
Sell Not Sell
Buy 4 − p, p 0, 0
Not Buy 0, 0 0, 0
I In the interim stage (just after the nature moves), the seller
will know which game will be played.
I The buyer will not know which game will be played.
Asymmetric Information
I In many game theoretic situations one player posses more
information than the other players.
I This situation is known as asymmetric information.
I Thus, manipulating the information will affect the equilibrium
outcome of the game.
I Therefore, such manipulation of asymmetric information itself
becomes a game of strategy.
I What would a better informed player like to do?
Usages of Information
The better informed player would like to
1. Conceal Information or reveal misleading information
I In mixing game of zero-sum game you do not like to reveal
what you have done.
I You bluff in poker to mislead other player.
2. Reveal selected information truthfully
I In strategic moves you want that others to see what you have
done.
The less informed player would like to either one of the followings:
1. Elicit information or filter truth from falsehood
2. Remain ignorant
I What should less informed player observe about the more
informed player, words or actions?
I However, if more informed person know that less informed
player will observe my action to get information, he could
manipulate his action in his best interest.
Usages of Information
The better informed player would like to
1. Conceal Information or reveal misleading information
I In mixing game of zero-sum game you do not like to reveal
what you have done.
I You bluff in poker to mislead other player.
2. Reveal selected information truthfully
I In strategic moves you want that others to see what you have
done.
The less informed player would like to either one of the followings:
1. Elicit information or filter truth from falsehood
2. Remain ignorant
I What should less informed player observe about the more
informed player, words or actions?
I However, if more informed person know that less informed
player will observe my action to get information, he could
manipulate his action in his best interest.
Signaling
I In strategic game, a player could have two types of
information:
I ‘Good’ information in a sense that its disclosure would cause
other player to act in way that would increase his payoff.
I A health insurance policy buyer is non-smoker.
I ‘Bad’ information in a sense that its disclosure would cause
other player to act in way such that would decrease his payoff.
I He cheated his way in a college.
I In case of good information the player will take actions that
will induce others to believe his information is good.
I Such actions are called signals and the strategy of using them
is called signaling.
I In case of bad information the player will take actions which
confuses others to know correct information.
I This strategy is called signal jamming.
Signaling
I In strategic game, a player could have two types of
information:
I ‘Good’ information in a sense that its disclosure would cause
other player to act in way that would increase his payoff.
I A health insurance policy buyer is non-smoker.
I ‘Bad’ information in a sense that its disclosure would cause
other player to act in way such that would decrease his payoff.
I He cheated his way in a college.
I In case of good information the player will take actions that
will induce others to believe his information is good.
I Such actions are called signals and the strategy of using them
is called signaling.
I In case of bad information the player will take actions which
confuses others to know correct information.
I This strategy is called signal jamming.
Screening
I The less informed player will always try to take actions so that
he can reduce his informational disadvantage.
I The strategy of making another player act so as to reveal his
information is called screening.
I Any specific method used for this purpose is known as
screening devices.
I Note that signaling is a strategy for more informed player and
screening is a strategy for less informed player.
Cheap Talk
I What is the simplest way to convey information to others or
elicit information from others?
I Just tell them what it is or simply ask them.
I However, the credibility of mere words is questionable.
I Therefore, there is a saying that talk is cheap.
I Any direct communication without any cost is called cheap
talk by game theorists.
I Cheap talk equilibrium is equilibrium which is achieved by
using direct communication.
Cheap Talk
I What is the simplest way to convey information to others or
elicit information from others?
I Just tell them what it is or simply ask them.
I However, the credibility of mere words is questionable.
I Therefore, there is a saying that talk is cheap.
I Any direct communication without any cost is called cheap
talk by game theorists.
I Cheap talk equilibrium is equilibrium which is achieved by
using direct communication.
When Does Direct Communication Work?
I Direct communication works when the preferences of players
are well aligned.
I Coordination game is a typical example.
I Recall the familiar game matrix:
1/2 X Y
X 1,1 0,0
Y 0,0 2,2
I Both the players would like to meet.
I Suppose Player 1 send a message that “I am going to place
Y ”.
I Thus, player 2 does believe Player 1.
When Does Direct Communication Work?
I We can analyze this situation more formally.
I There are two stages of the game. In the first stage, Player 1
sends the message and in the second stage both are playing
the simultaneous-move game.
I Use backward induction to analyze the game.
I Recall that in sequential move games players specify the
complete plans of action.
I Thus, if Player 1 sends the message “I am going to place Y ”
in the first stage, then in the second stage Player 2 chooses to
go place Y as well.
I Observe that it is a rollback equilibrium.
I However, there is another equilibrium where Player 1’s
message is disregarded.
I This is known as babbling equilibria.
Credibility of Direct Communication
I Credibility of direct communication depends on
I on the alignment of players’ interests.
I Observe that in extreme case where interests are perfectly
aligned direct communication works well.
I Similarly another extreme case where players’ interests are in
total conflict direct communication does not work.
I In the penalty shooting game
1/2 L R
L 50,50 80,20
R 90,10 20,80
I In this game whatever message is sent by Player 1 in the first
stage, it will be disregarded in the second stage.
I Only equilibrium is the original equilibrium of mixed strategy.
