Incomplete information: Perfect Bayesian equilibrium Lectures in Game Theory Spring 2011, Part 7 10.12.2010 G.B. Asheim, ECON4240-7 A static Bayesian game U 1 A ( 12 ) Nature B ( 12 ) D U 1 D Equivalent representation: 2 L R 2 L R 2 L R 2 L 1 L 2 A ( 12 ) Nature B ( 12 ) R L 2 R R 1 U D 1 U D 1 U D 1 U D What if the uninformed gets to What if the informed gets to observe the informed choice? observe the uninformed choice? 10.12.2010 2 G.B. Asheim, ECON4240-7 A dynamic Bayesian game: Screening A ( 12 ) Nature 1 2 L B( ) 2 R A ( 12 ) Nature B ( 12 ) Equivalent representation: The informed gets to observe the uninformed choice. 1 U 1 U D 1 U D 1 U D 1 U L 2 A ( 12 ) Nature B ( 12 ) R L 2 D R D 1 U D 1 U D 1 U D Use subgame perfect Nash equilibrium! 10.12.2010 G.B. Asheim, ECON4240-7 3 1 A dynamic Bayesian game: Signaling The uninformed gets to observe the informed choice. 2 L U 1 A ( 12 ) Nature B ( 12 ) Equivalent representation: L 2 U1 R 2 L D R D R Nature B ( 12 ) L R 2 L 1 L A ( 12 ) R 2 L U D 2 U 1 R L D R R Requires a new equilibrium concept: Perfect Bayesian equilibrium 10.12.2010 Why? 4 G.B. Asheim, ECON4240-7 Gift game, version 1 1 G 0, 0 N F ( 12 ) F F 2 A 2, 2 R - 2, 0 Nature E ( 12 ) 0, 0 NE 1 GE 1 F 2 G G E R A 2, 1 - 2, - 1 G F N E 1, 1 - 1, 0 1 0 - 1, 1 -1 N FG E 1, A 2, 0 N F N E 0, 0 0, 0 R - 2, - 2 Are both Nash equilibria reasonable? Note that subgame perfection does not help. Why? 10.12.2010 5 G.B. Asheim, ECON4240-7 Remedy 1: Conditional beliefs about types F1 G F 2 A 2, 2 0, 0 N ( q ) R - 2, 0 F ( 12 ) Nature E ( 12 ) 0, 0 (1 q ) A 2,, 0 N 1 GE R - 2, - 2 Let player 2 assign probabilities to the two types player 1: the belief of player 2. E What will the players choose? Remedy 2: Sequential rationality Each player chooses rationally at all information sets, given his belief and the opponent’s strategy. 10.12.2010 G.B. Asheim, ECON4240-7 6 2 Gift game, version 2 F1 GF 2 A 0, 0 N (q ) R F ( 12 ) R A 0, 0 0, 0 G G 2, 0 G F N E 1, 1 1, 0 Nature F E 1 - 1 - 1, 1 0 N FG E - 1, E ( 12 ) 0, 0 2 2, 2 1 (1 q ) A - 2, - 2 N F N E 0, 0 0, 0 N 1 GE R - 2, 0 E If q ½, then player 2 will choose R. If so, the outcome is not a Nash equil. outcome. 10.12.2010 7 G.B. Asheim, ECON4240-7 Remedy 3: Consistency of beliefs F1 GF 2 A 0, 0 N (1) R F ( 12 ) 2, 2 2, 0 Nature E ( 12 ) 0, 0 Player 2 should find his belief by means of Bayes’ rule, when-ever possible. 2 -2 (0) A - 2, Pr[ G | F] Pr[ F] E q Pr[ F | G ] N 1 G R - 2, 0 Pr[ G ] E An example of a separating equilibrium. An equilibrium is separating if the types of a player behave differently. 10.12.2010 8 G.B. Asheim, ECON4240-7 Gift game, version 3 If q ½, then player 2 will choose R. Bayes’ rule cannot be used. F1 G F 2 A 2, 2 0, 0 N ( q ) R - 2, 0 F ( 12 ) Nature E ( 12 ) 0, 0 2 R A G FG E 2, 0 - 2, 0 1 G F N E 1, 1 - 1, 0 1 - 1 - 1, 1 0 N FG E 1, (1 q ) A 2, - 2 N F N E 0, 0 0, 0 N 1 GE R - 2, 0 E An example of a pooling equilibrium. An equilibrium is pooling if the types behave the same. 10.12.2010 G.B. Asheim, ECON4240-7 9 3 Perfect Bayesian equilibrium Definition: Consider a strategy profile for the players, as well as beliefs over the nodes at all information sets. These are called a perfect Bayesian equilibrium (PBE) if: (a) (b) each player’s strategy specifies optimal actions given his beliefs and the strategies of the other players. the beliefs are consistent with Bayes’ rule whenever possible. 10.12.2010 10 G.B. Asheim, ECON4240-7 Algorithm for finding perfect Bayesian equilibria in a signaling game: posit a strategy for player 1 (either pooling or separating), calculate restrictions on conditional beliefs, beliefs calculate optimal actions for player 2 given his beliefs, check whether player 1’s strategy is a best response to player 2’s strategy. 10.12.2010 11 G.B. Asheim, ECON4240-7 Applying the algorithm in a signaling game Player 1 has four pure strategies. 1, 3 U 2 L 1 PBE w/(LL)? YES 4, 0 D [(LL), (DU), q , r 1 / 2] where q 2 / 3. R 2 U 2, 1 (q ) D 0, 0 A ( 12 ) Nature B ( 12 ) 2 4 U (1 r ) 2, 1 0 (1 q ) U 1, R L 1 D 1, 2 0, 1 D PBE w/(RR)? NO PBE w/(LR)? NO [(RL), (UU), q 1, r 0] PBE w/(RL)? YES [(RL), (UU), q 1, r 0] 10.12.2010 (r ) is a separating equilibrium. [(LL), (DU), q , r 1 / 2 ] is a pooling equilibrium. G.B. Asheim, ECON4240-7 12 4