Game Theory

Part 4
Bayesian games

PowerPoint Slides prepared by:

Andreea CHIRITESCU
Eastern Illinois University
© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as
1
permitted in a license distributed with a certain product or service or otherwise on a password-protected website for classroom use.
 Incomplete Information
• Some players
– Have information about the game that
others do not
• Players that lack full information
– Will try to use what they do know to make
inferences about what they do not

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as
2
permitted in a license distributed with a certain product or service or otherwise on a password-protected website for classroom use.
 Simultaneous Bayesian Games
• A two-player
– Simultaneous-move game
– In which player 1 has private information
– But player 2 does not
• We can model private information
– By introducing player types

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as
3
permitted in a license distributed with a certain product or service or otherwise on a password-protected website for classroom use.
 Player Types and Beliefs
• Player 1
– Can be of a number of possible types, t
– Knows his own type
• Player 2
– Is uncertain about t
– Must decide strategy based on her beliefs
about t

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as
4
permitted in a license distributed with a certain product or service or otherwise on a password-protected website for classroom use.
 Player Types and Beliefs
• The game begins at a chance node
– At which a value of tk is drawn for player 1
from a set of possible types,
T = {t1,…,tk,…,tK}
– Pr(tk) = probability of drawing type tk
– Player 1 sees which type is drawn
– Player 2 does not see which type is drawn
• Only knows the probabilities

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as
5
permitted in a license distributed with a certain product or service or otherwise on a password-protected website for classroom use.
 Player Types and Beliefs
• Since player 1 observes t before moving
– His strategy can be conditioned on t
• Let s1(t) be 1’s strategy contingent on his type
• Player 2’s strategy is an unconditional one, s2

• Player 1’s type may affect player 2’s

payoff in two ways
– Directly
– Indirectly through player 1’s strategy

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as
6
permitted in a license distributed with a certain product or service or otherwise on a password-protected website for classroom use.
 8.14
Simple Game of Incomplete Information

t = 6 with probability 1/2 and t = 0 with probability 1/2.

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as permitted in a license distributed with a
certain product or service or otherwise on a password-protected website for classroom use.
7
 8.15
Extensive Form for Simple Game of Incomplete Information

This figure translates Figure

8.14 into an extensive-form
game. The initial chance node is
indicated by an open circle.
Player 2’s decision nodes are in
the same information set
because she does not observe
player 1’s type or action before
moving.

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as permitted in a license distributed with a
certain product or service or otherwise on a password-protected website for classroom use.
8
 Bayesian-Nash Equilibrium
• Equilibrium requires that
– 1’s strategy be the best response for each
and every one of his types
– 2’s strategy maximize an expected payoff
• The expectation is taken with respect to her
beliefs about 1’s type

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as
9
permitted in a license distributed with a certain product or service or otherwise on a password-protected website for classroom use.
 Computing expected value…

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as
10
permitted in a license distributed with a certain product or service or otherwise on a password-protected website for classroom use.
 Bayesian-Nash Equilibrium
• Bayesian-Nash equilibrium
– In a two-player simultaneous move game
in which player 1 has private information,
– Is a strategy profile (s*1(t),s*2) such that
u1(s*1(t),s*2,t)  u1(s’1,s*2,t) for all s’1  S1
and
 Pr t  u  s , s t  , t    Pr t  u  s , s t  , t 
tk T
k 2
*
2
*
1 k k
t k T
k 2
'
2
*
1 k k

'
for all s  S 2 2

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as
11
permitted in a license distributed with a certain product or service or otherwise on a password-protected website for classroom use.
 8.6 Bayesian–Nash Equilibrium of Game
in Figure 8.15
• Two possible candidates for an equilibrium in
pure strategies
• 1 plays (U|t=6, D|t=0) and 2 plays L
• Not an equilibrium
• 1 plays (D|t=6, D|t=0) and 2 plays R
• A Bayesian-Nash equilibrium

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as permitted in a license distributed with a
certain product or service or otherwise on a password-protected website for classroom use.
12
 8.7 Tragedy of the Commons as a
Bayesian Game
• Herder 1
• Has private information regarding his value of
grazing per sheep, v1(q1,q2,t)=t-(q1+q2)
• His type is
• t=130 (the ‘‘high’’ type) with probability 2/3
• t=100 (the ‘‘low’’ type) with probability 1/3
• Value-maximization problem:
max{q1v1 (q1 , q2 , t )}  max{q1 (t  q1  q2 )}
q1 q1

• First-order condition: t-2q1-q2=0

• So, q1H=65-q2/2 and q1L=50-q2/2
© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as permitted in a license distributed with a
certain product or service or otherwise on a password-protected website for classroom use.
13
 8.7 Tragedy of the Commons as a
Bayesian Game
• Herder 2
• Expected payoff:
2 1
[q2 (120  q1H  q2 )]  [q2 (120  q1L  q2 )]  q2 (120  q1  q2 )
3 3
2 1
where q1  q1H  q1L
3 3
q1 q2
q2  60   30 
2 4
q2*  40, q1*H  45, q1*L  30

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as permitted in a license distributed with a
certain product or service or otherwise on a password-protected website for classroom use.
14
 8.16
Equilibrium of the Bayesian Tragedy of the Commons

Best responses for herder 2 and

both types of herder 1 are
drawn as thick solid lines; the
expected best response as
perceived by 2 is drawn as the
thick dashed line. The
Bayesian–Nash equilibrium of
the incomplete-information
game is given by points A and
C; Nash equilibria of the
corresponding full-information
games are given by points A’
and C’ .

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as permitted in a license distributed with a
certain product or service or otherwise on a password-protected website for classroom use.
15
 Experimental Games
• Experimental economics
– Explores how well economic theory
matches the behavior or experimental
subjects in a laboratory setting
– Play in the Prisoners’ Dilemma converged
to the Nash equilibrium as subjects gained
experience
– Results from the Ultimatum game
suggests that fairness matters

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as
16
permitted in a license distributed with a certain product or service or otherwise on a password-protected website for classroom use.
 Evolutionary Games And Learning
• Evolutionary model
– Players do not make rational decisions
• They play the way they are genetically
programmed
– The more successful a player’s strategy in
the population
• The more fit is the player
• The more likely will the player survive to pass
his or her genes on to future generations

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as permitted in a license distributed with a
certain product or service or otherwise on a password-protected website for classroom use.
17
 Evolutionary Games And Learning
• Learning model, players
– Are matched at random with others from a
large population
– Use their experiences of payoffs from past
play
– Are assumed to have a degree of
rationality
– Are not fully rational

© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as permitted in a license distributed with a
certain product or service or otherwise on a password-protected website for classroom use.
18