gameL9 - Perfect Bayesian Equilibrium For an important...

This preview shows pages 1–6. Sign up to view the full content.

Perfect Bayesian Equilibrium For an important class of extensive games, a solution concept is available that is simpler than sequential equi- librium, but with similar properties. In a Ba y e s iane x t en s i v egam ew i thob s e r vab l ea c - tions ,naturemoves f rst and independently selects a type for each player. Afterwards, the actions chosen by players are observed by all.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
De f nition 231.1: A Ba y e s iane x t en s i v egam ew i th observable actions is a tuple, h Γ , ( Θ i ) , ( p i ) , ( u i ) i where 1. Γ = h N,H,P i is an extensive game form with perfect information and (possibly) simultaneous moves 2. The set of types for player i , Θ i ,isa f nite set, and we denote Θ = × i N Θ i 3. p i is a probability measure on Θ i ,whe re p i ( θ i ) > 0 for all θ i Θ i , and these measures are independent 4. u i : Θ × Z R is a von Neumann-Morgenstern utility function.
The set of histories is { } ( Θ × H ) The information set for player i P ( h ) is of the form I ( θ i ,h )= { (( θ i 0 i ) ): θ 0 i Θ i } To solve the game, we will be looking for a pro f le of be- havioral strategies in Γ (for each player and each type) (( σ i ( θ i )) and a belief system µ i ( h ) that speci f es a com- mon belief, after the history h , held by all players other than i about player i ’s type. Let s be a pro f le of behavioral strategies in Γ .D e f ne O h ( s ) to be the probability measure on terminal histories of Γ generated by s ,g iventheh isto ry h . De f ne O ( σ i ,s i i | h ) to be the probability measure on terminal histories of Γ , given i uses s i , other players use type-dependent behavioral strategies σ i ,theh istory reached is h , and beliefs are given by µ i .

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
De f nition 232.1: Let h Γ , ( Θ i ) , ( p i ) , ( u i ) i be a Bayesian extensive game with observable actions, with Γ = h N,H,P i . Apair (( σ i ) , ( µ i )) = (( σ i ( θ i )) i N,θ i Θ i , ( µ i ( h )) i N,h H \ Z ) is a perfect Bayesian equilibrium if the following con- ditions are satis f ed 1. Sequential Rationality : For every non-terminal his- tory h H \ Z ,eve ryp laye r i P ( h ) , and every type θ i Θ i , the probability measure O ( σ i i ( θ i ) i | h ) is weakly preferred by type θ i to O ( σ i ,s i i | h ) for any strategy s i of player i in Γ . 2. Correct initial beliefs : µ i ( )= p i for each i N . 3. Action-determined beliefs :I f i/ P ( h ) and a A ( h ) ,then µ i ( h, a µ i ( h ) .I f i P ( h ) , a A ( h ) , a 0 A ( h ) ,and a i = a 0 i µ i ( h, a µ i ( h, a 0 ) .
4. Bayesian updating :I f i P ( h ) and a i is in the support of σ i ( θ i )( h ) for some θ i in the support of µ i ( h ) , then for any θ 0 i Θ i ,wehave µ i ( h, a )( θ 0 i )= σ i ( θ 0 i )( h )( a i )[ µ i ( h )( θ 0 i )] P θ i Θ i σ i ( θ i )( h )( a i )[ µ i ( h )( θ i )] . Interpreting conditions 3 and 4: Action-determined beliefs—since types are independent, beliefs about player i ’s type cannot be in f uenced by the actions of the other players. You can’t signal what you don’t know. If player i does not make a move after the history h , then the action pro F le after history h does not change µ i .I f p l a y e r i does move after the history h ,

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 07/17/2008 for the course ECON 817 taught by Professor Peck during the Fall '07 term at Ohio State.

Page1 / 28

gameL9 - Perfect Bayesian Equilibrium For an important...

This preview shows document pages 1 - 6. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online