Bayesian Games Yasuhiro Kirihata University of Illinois at Chicago Outline - Game in strategic form - Bayesian Game - Bayesian Equilibrium - Examples - Battle of Sexes with incomplete information - Cournot Duopoly with incomplete information Game in strategic form What is the game? … - It is regarded as a multiagent decision problem - Several players has several strategies. - Performs strategies to maximize its payoff funciton Game in strategic form (N, {A i }iN , {u i }iN ) is given by three objects specified by (1)the set of players: N {1, 2, ..., n} (2)the action(strategy)space of players: Ai, i 1, 2, ..., n (3)the payoff functions of players: u i : Ai R, i 1, 2, ..., n - Note that ui is determined by the outcome (strategy profile). What is Bayesian Game? Game in strategic form - Complete information(each player has perfect information regarding the element of the game) - Iterated deletion of dominated strategy, Nash equilibrium: solutions of the game in strategic form Bayesian Game - A game with incomplete information - Each player has initial private information, type. - Bayesian equilibrium: solution of the Bayesian game Bayesian Game Definition (Bayesian Game) A Bayesian game is a strategic form game with incomplete information. It consists of: - A set of players N={1, …, n} for each i∈N - An action set Ai, (A i N Ai) - A type set Θi, (Θ i N Θi) - A probability function, pi : i ( i) - A payoff function, ui : A R - The function pi is what player i believes about the types of the other players - Payoff is determined by outcome A and type Bayesian Game Definition Bayesian game ( N ,{ Ai }iN ,{i }iN ,{ pi }iN ,{ui }iN ) is finite if N , Ai , and i are all finite Definition(pure strategy, mixed strategy) Given a Bayesian Game ( N ,{ Ai }iN ,{i }iN ,{ pi }iN ,{ui }iN ), A pure strategy for player i is a function which maps player i’s type into its action set ai : i Ai A mixed strategy for player i is i : i ( Ai ) : i i (. | i ) Bayesian Equilibrium Definition(Bayesian Equilibrium) A Bayesian equilibrium of a Bayesian game is a mixed strategy profile ( i ) iN , such that for every player i∈N and every type i i , we have i (. | i ) arg max ( Ai ) p ( i i i i | i ){ aA j (a j | j )} (ai )ui (a, ) jN \{i} - Bayesian equilibrium is one of the mixed strategy profiles which maximize the each players’ expected payoffs for each type. Bayesian Equilibrium Remark) - This equilibrium is the solution of the Bayesian game. This equilibrium means the best response to each player’s belief about the other player’s mixed strategy. -In the definition of Bayesian equilibrium, we need to specify strategies for each type of a player, even if in the actual game that is played all but one of these types are non-exist Examples ー Battle of Sexes Battle of Sexes with incomplete information (1) N={1,2}: player1 and player2(wife and husband) (2) A1=A2={B,S} (Ballet and Soccer) (3) 1 { x }, 2 { l , h } - Type x: player1 loves going out with player2 - Type l : player2 loves going out with player1 - Type h: player2 hates going out with player1 (4) p1 (l | x) p1 (h | x) 1/ 2, p2 ( x | l ) p2 ( x | h) 1 (5)u1 and u2 are given in the game matrix on the next slide Examples ー Battle of Sexes Game matrixes of BoS S B B 2,1 0,0 S 0,0 1,2 type l B B 2,0 S 0,2 S 0,1 1,0 type h - Since player1 has only type x, we omit the parameter x in the payoff functions ui, i = 1,2. - These matrixes define the payoff functions: u1(B,B,l) = 2, u2(B,B,l) = 1, u1(B,B,h) = 2, …and so on Examples ー Battle of Sexes Calculate the Bayesian Equilibrium Player 2 of type l: Given player 1’s strategy 1 - Action B: 2 ( B | l ) 1 EP p2 ( x | l ){1 ( B) 2 ( B | l )u2 ( B, B, l ) 1 (S ) 2 ( B | l )u2 (S , B, l )} 1 ( B) - Action S: 2 (S | l ) 1 EP p2 ( x | l ){1 ( B) 2 (S | l )u2 ( B, S , l ) 1 (S ) 2 (S | l )u2 (S , S , l )} 2(1 1 ( B)) Best response is B if 1 ( B) 2 / 3 , S if 1 ( B) 2 / 3 Examples ー Battle of Sexes Player 2 of type h: Given player 1’s strategy 1 - Action B: 2 ( B | h) 1 EP p2 ( x | h){1 ( B) 2 ( B | h)u2 ( B, B, h) 1 (S ) 2 ( B | h)u2 (S , B, h)} (1 1 ( B)) - Action S: 2 (S | h) 1 EP p2 ( x | h){1 ( B) 2 (S | h)u2 ( B, S , h) 1 (S ) 2 (S | h)u2 (S , S , h)} 21 ( B) Best response is B if 1 ( B) 1 / 3, S if 1 ( B) 1 / 3 Examples ー Battle of Sexes Player 1: Given player 2’s strategy 2 (. | l ) and 2 (. | h) -Action B: 1 ( B) 1 EP p1 (l | x){1 ( B) 2 ( B | l )u1 ( B, B, l ) 1 ( B) 2 (S | l )u1 ( B, S , l )} p1 (h | x){1 ( B) 2 ( B | h)u1 ( B, B, h) 1 ( B) 2 (S | h)u1 ( B, S , h)} 2 ( B | l ) 2 ( B | h) -Action S: 1 (S ) 1 EP p1 (l | x){1 (S ) 2 ( B | l )u1 (S , B, l ) 1 (S ) 2 (S | l )u1 (S , S , l )} p1 (h | x){1 (S ) 2 ( B | h)u1 (S , B, h) 1 (S ) 2 (S | h)u1 (S , S , h)} 2 ( B | l ) 2 ( B | h) 1 2 Best response is B if 2 ( B | l ) 2 ( B | h) 2 / 3 Best response is S if 2 ( B | l ) 2 ( B | h) 2 / 3 Examples ー Battle of Sexes Bayesian equilibrium for pure strategy - Assume that both types of player 2’s strategies are pure strategy, and check the all combination of strategies pair. ( 2 ( B | l ), 2 ( B | h)) (1,1), (1, 0), (0,1), (0, 0) - Condition of Bayesian equilibrium is not satisfied by: ( 2 ( B | l ), 2 ( B | h)) (1,1), (0,1), (0, 0) - Bayesian equilibrium for pure strategy is given by: (1 ( B | x), 2 ( B | l ), 2 ( B | h)) (1,1, 0) Examples ー Battle of Sexes Bayesian equilibrium for mixed strategy - There is no equilibrium in which both types of player 2 mixes. (Because, if both type of player 2 mixes, 1 ( B) should be 2/3 and 1/3 in the equilibrium. This is contradiction!) - Suppose only type l mixes. Then, 1 ( B) 2 / 3. This implies that strategy of player 1 mixes, i.e. 2 ( B | l ) 2 ( B | h) 2 / 3. - Type h of player 2 does not mix and 2 ( B | l ) 0. 2 ( B | h) 0. - Bayesian equilibrium is given by: (1 ( B | x), 2 ( B | l ), 2 ( B | h)) (2 / 3, 2 / 3, 0) - Similarly, Bayesian equilibrium when type h mixes is given by: (1 ( B | x), 2 ( B | l ), 2 ( B | h)) (1/ 3, 0, 2 / 3) Examples ー Cournot Duopoly Cournot Duopoly model Players (2 firms): N {1, 2} Action set (outcome of firms): qi R , (i 1, 2) Type set: 1 {1}, 2 {3 / 4, 5 / 4} Probability function: p(2 3 / 4 | 1 ) 1/ 2, p(2 5 / 4 | 1 ) 1/ 2 (5) Profit function: (1) (2) (3) (4) u1 (q1 , q2 ,1 , 2 ) q1 (1 q1 q2 ) u2 (q1 , q2 ,1 , 2 ) q2 ( 2 q1 q2 ) Examples ー Cournot Duopoly Bayesian equilibrium for pure strategy - The Bayesian equilibrium is a maximal point of expected payoff of firm 2, EP2: EP2 * * (q1 , q2 ) 2 q1* 2q2* 0 EP2 u2 q2 q2* ( 2 ) ( 2 q1* ) / 2, ( 2 3 / 4,5 / 4) - The expected payoff of player 1, EP1, is given as follows: 1 1 EP1 q1 (1 q1 q2 (3 / 4)) q1 (1 q1 q2 (5 / 4)) 2 2 Examples ー Cournot Duopoly Bayesian equilibrium is also the maximal point of expected payoff EP1: EP1 * * 1 * * (q1 , q2 ) 1 2q1 {q2 (3 / 4) q2* (5 / 4)} 0 q1 2 2 q2* (3 / 4) q 2* (5 / 4) q 4 * 1 Solving above equations, we can get Bayesian equilibrium as follows: 1 11 * 5 q1* , q2* (3 / 4) , q2 (5 / 4) . 3 24 24