Bayesian Nash Equilibrium Static (or Simultaneous-Move) Games of Incomplete Information-Chapter 3 Bayesian Nash Equilibrium
Outline of Static Games of Incomplete Information Lecture 22 June 19, 2003 Outline of Static Games of Incomplete Information Introduction to static games of incomplete information Normal-form (or strategic-form) representation of static Bayesian games Bayesian Nash equilibrium Applications---Auction Game Theory-Chapter 3
a static game of incomplete information 什么是不完全信息静态博弈? 1. 不完全信息的囚徒困境 2. 不完全信息的Cournot双头垄断 3. 不完全信息的性别战 4. 首价密封拍卖(First-price, sealed-bid auction) Game Theory-Chapter 3
Static (or simultaneous-move) games of complete information 参与人集(至少两个参与人) 每个参与人的策略/行动集 每个参与人策略组合的收益,或每个参与人对策略组合的偏好?Payoffs received by each player for the combinations of the strategies, or for each player, preferences over the combinations of the strategies 所有这些是确定性的,且为每个参与人的共同知识(common knowledge ). Game Theory-Chapter 3
Static (or simultaneous-move) games of INCOMPLETE information 收益不再是确定的(不完全信息),尽管某些随机变量的分布仍为共同知识。 不完全信息意味着 至少有一个参与人不能准确的知道某个参数变量的值(如其他参与人的类型,type ) 不完全信息静态博弈也被称为静态贝叶斯博弈(static Bayesian games ) Game Theory-Chapter 3
Prisoners’ dilemma of complete information 两名犯罪嫌疑人被捕并受到指控,他们被关入不同的牢室。但是警方并无充足证据. 两名犯罪嫌疑人被告知以下政策: 如果两人都不坦白,将均被判为轻度犯罪,入狱一个月. 如果双方都坦白,都将被判入狱六个月. 如果一人招认而另一人拒不坦白,招认的一方将马上获释,而另一人将判入狱九个月. Prisoner 2 Mum Confess Prisoner 1 -1 , -1 -9 , 0 0 , -9 -6 , -6 Game Theory-Chapter 3
Prisoners’ dilemma of incomplete information Prisoner 1总是理性的(自利的,selfish ). Prisoner 2可能是理性的(自利的) ,也可能是利他的(altruistic ), 这取决于他是否高兴(happy). 如果他是利他的,那么他更偏好于mum,他认为 “confess” 等于额外“入狱四个月”. Prisoner 1不能确切的知道 prisoner 2是理性的还是利他的, 但是他推断(believes ) prisoner 2理性的概率为 0.8, 利他的概率为0.2. Payoffs if prisoner 2 is altruistic Prisoner 2 Mum Confess Prisoner 1 -1 , -1 -9 , -4 0 , -9 -6 , -10 Game Theory-Chapter 3
Prisoners’ dilemma of incomplete information (continued) 给定prisoner 1关于prisoner 2的推断( belief ), prison 1 应该选择什么策略? 如果prisoner 2 是理性或利他的,他应该分别选择什么策略? Payoffs if prisoner 2 is rational Prisoner 2 Mum Confess Prisoner 1 -1 , -1 -9 , 0 0 , -9 -6 , -6 Payoffs if prisoner 2 is altruistic Prisoner 2 Mum Confess Prisoner 1 -1 , -1 -9 , -4 0 , -9 -6 , -10 Game Theory-Chapter 3
Prisoners’ dilemma of incomplete information (continued) 解: Prisoner 1选择 confess, 给定他对prisoner 2 的推断 Prisoner 2如果是理性,选择 confess;如果是利他的,则选择 mum 这可以写成 (Confess, (Confess if rational, Mum if altruistic)) Confess 是 prisoner 1对prisoner 2的选择(Confess if rational, Mum if altruistic)的最优反应. (Confess if rational, Mum if altruistic) 是 prisoner 2对prisoner 1选择Confess的最优反应 这里的一个纳什均衡被称为贝叶斯纳什均衡(Bayesian Nash equilibrium ) Game Theory-Chapter 3
Cournot duopoly model of complete information 标准式表述: 参与人集: { Firm 1, Firm 2} 策略集: S1=[0, +∞), S2=[0, +∞) 收益函数: u1(q1, q2)=q1(a-(q1+q2)-c), u2(q1, q2)=q2(a-(q1+q2)-c) 所有这些信息是共同知识 Game Theory-Chapter 3
Cournot duopoly model of incomplete information 一种同质的产品仅仅由两家企业进行生产: firm 1 和firm 2. 产量分别用q1 和q2表示. 它们同时选择它们的产量. 市场价格: P(Q)=a-Q, 这里 a 是常数并且 Q=q1+q2. Firm 1的成本函数: C1(q1)=cq1. 以上均为共同知识 Game Theory-Chapter 3
Cournot duopoly model of incomplete information (continued) Firm 2的边际成本依赖于某个只有它自己知道的因素 (如技术水平).它的边际成本可能是 较高(HIGH): 成本函数: C2(q2)=cHq2. 较低(LOW): 成本函数: C2(q2)=cLq2. 在生产前, firm 2能够观察到这个因素并且准确知道它的边际成本处于什么水平. 但是, firm 1不能准确知道 firm 2的成本. 也就是说,它不能确定 firm 2的收益. Firm 1推断 firm 2的成本函数 以的概率为C2(q2)=cHq2 以1–的概率为C2(q2)=cLq2. 以上均为共同知识 Game Theory-Chapter 3
Cournot duopoly model of incomplete information (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version one) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version two) 一种同质的产品仅仅由两家企业进行生产: firm 1 和firm 2. 产量分别用q1 和q2表示. 它们同时选择它们的产量. 市场价格: P(Q)=a-Q, 这里 a 是常数并且 Q=q1+q2. 以上均为共同知识 Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version two) (continued) Firm 2的边际成本依赖于某个只有它自己知道的因素 (如技术水平).它的边际成本可能是 较高(HIGH): 成本函数: C2(q2)=cHq2. 较低(LOW):成本函数: C2(q2)=cLq2. 在生产前, firm 2能够观察到这个因素并且准确知道它的边际成本处于是高是低. 但是, firm 1不能准确知道 firm 2的成本. 也就是说,它不能确定 firm 2的收益. Firm 1推断 firm 2的成本函数 以的概率为C2(q2)=cHq2 以1–的概率为C2(q2)=cLq2. Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version two) (continued) Firm 1的边际成本也依赖于某个只有它自己知道的独立(independent )因素.它的边际成本可能是 较高(HIGH): 生产函数: C1(q1)=cHq1. 较低(LOW): 生产函数: C1(q1)=cLq1. 在生产前, firm 1能够观察到这个因素并且准确知道它的边际成本是高是低. 但是, firm 2不能准确知道 firm 1的成本. 也就是说,它不能确定 firm 1的收益. Firm 2推断 firm 1的生产函数 以的概率为C1(q1)=cHq1 以1–的概率为C1(q1)=cLq1. Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version two) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version two) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version two) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version two) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version two) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version two) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version two) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version two) (continued) Game Theory-Chapter 3
battle of the sexes 两个人都愿意在一起度过这个夜晚. 但是Chris更喜欢歌剧. Pat则更喜欢拳击. Prize Fight Opera 2 , 1 0 , 0 1 , 2 Game Theory-Chapter 3
Battle of the sexes with incomplete information (version one) 现在Pat的偏好依赖于他是否高兴(happy ). 如果他高兴,那么他的偏好仍然是拳击. 如果他不高兴,那么他宁愿晚上自己独处,他的偏好见下面的表格. Chris不知道 Pat 是否高兴. 但是Chris推断Pat以0.5的概率高兴,以0.5的概率不高兴 Payoffs if Pat is unhappy Pat Opera Prize Fight Chris 2 , 0 0 , 2 0 , 1 1 , 0 Game Theory-Chapter 3
Battle of the sexes with incomplete information (version one) (continued) 怎样找到解? Payoffs if Pat is happy with probability 0.5 Pat Opera Prize Fight Chris 2 , 1 0 , 0 1 , 2 Payoffs if Pat is unhappy with probability 0.5 Pat Opera Prize Fight Chris 2 , 0 0 , 2 0 , 1 1 , 0 Game Theory-Chapter 3
Battle of the sexes with incomplete information (version one) (continued) 最优反应 如果Chris选择 opera 那么Pat的最优反应: opera(如果他高兴), prize fight(如果他不高兴) 假设Pat高兴时选择 opera, 不高兴时选择prize fight. Chris的最优反应是什么? 考虑Chris选择 opera的情形: 如果此时Pat高兴,则Chris的收益为2 ;如果 Pat不高兴,则Chris的收益为0. 从而Chris的期望收益(expected payoff )是20.5+ 00.5=1 考虑Chris选择 prize fight的情形:如果此时Pat高兴,则Chris的收益为0,如果 Pat不高兴,则Chris的收益为1.从而Chris的期望收益是00.5+ 10.5=0.5 由于 1>0.5, Chris的最优反应是opera 一个贝叶斯纳什均衡: (opera, (opera if happy and prize fight if unhappy)) Game Theory-Chapter 3
Battle of the sexes with incomplete information (version one) (continued) 最优反应 如果Chris选择 prize fight 那么Pat的最优反应: prize fight (如果他高兴), opera (如果他不高兴) 假设Pat高兴时选择 prize fight, 不高兴时选择opera. Chris的最优反应是什么? 考虑Chris选择 opera的情形: 如果此时Pat高兴,则Chris的收益为0 ;如果 Pat不高兴,则Chris的收益为2.从而Chris的期望收益是00.5+ 20.5=1 考虑Chris选择 prize fight的情形:如果此时Pat高兴,则Chris的收益为1 ,如果 Pat不高兴,则Chris的收益为0.从而Chris的期望收益是10.5+ 00.5=0.5 由于1>0.5, Chris的最优反应是opera (Prize fight, (prize fight if happy and opera if unhappy))不是一个贝叶斯纳什均衡. Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version three) (continued) Firm 2的成本依赖于某个只有它自己知道的因素 (如技术水平). 她的成本可能 较高(HIGH): 成本函数: C2(q2)=cHq2. 较低(LOW): 成本函数: C2(q2)=cLq2. Firm 1的成本也依赖于某个其他只有它自己知道的独立或不独立的 (independent or dependent)因素. 它的成本可能 较高(HIGH): 成本函数: C1(q1)=cHq1. 较低(LOW): 成本函数: C1(q1)=cLq1. Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version three) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version three) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version three) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version three) (continued) u1(q1, q2(cH); cH) u1(q1, q2(cL); cH) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version three) (continued) u1(q1, q2(cH); cL) u1(q1, q2(cL); cL) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version three) (continued) u2(q1(cH), q2; cH) u2(q1(cL), q2; cH) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version three) (continued) u2(q1(cH), q2; cL) u2(q1(cL), q2; cL) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version three) (continued) Game Theory-Chapter 3
Cournot duopoly model of incomplete information (version three) (continued) Game Theory-Chapter 3
Normal-form representation of static Bayesian games Game Theory-Chapter 3
Normal-form representation of static Bayesian games: payoffs Game Theory-Chapter 3
Normal-form representation of static Bayesian games: beliefs (probabilities) Game Theory-Chapter 3
海萨尼转换(the Harsanyi transformation)(p.116) 将不完全信息静态博弈转化为完全且不完美信息动态博弈. (1)引入虚拟的“自然”博弈方。自然赋予博弈各方的类型向量t=(t1, …,tn),其中ti属于可行集Ti; (2)自然告知参与人i自己的类型ti ,但不告诉其他参与人的类型; (3)参与人同时选择行动,每一参与人i从可行集Ai中选择ai; (4)除自然外,其余各方得到收益ui(a1, …,an; ti) Game Theory-Chapter 3
关于推断 (p.117) 自然根据先验的概率分布p(t)赋予各参与人类型向量t=(t1, …,tn),是共同知识 自然告知参与人i的类型ti时,他可以根据贝叶斯法则计算其他参与人类型的条件概率,得出推断 对Ti中的每一个ti ,都可计算出 Game Theory-Chapter 3
Strategy Game Theory-Chapter 3
Bayesian Nash equilibrium: 2-player Game Theory-Chapter 3
Bayesian Nash equilibrium: 2-player In the sense of expectation based on her belief player 2’s best response if her type is t2j player 1’s best response if her type is t1i In the sense of expectation based on her belief Game Theory-Chapter 3
battle of the sexes 两个人都愿意在一起度过这个夜晚. 但是Chris更喜欢歌剧. Pat则更喜欢拳击. Prize Fight Opera 2 , 1 0 , 0 1 , 2 Game Theory-Chapter 3
Battle of the sexes with incomplete information (version two) Pat的偏好依赖于他是否高兴.如果他高兴,那么他的偏好仍然是拳击. 如果他不高兴,那么他宁愿晚上自己独处. Chris不知道 Pat 是不是高兴. 但是Chris推断Pat以0.5的概率高兴,以0.5的概率不高兴 Chris的偏好也依赖于她是否高兴.如果她高兴,那么她的偏好仍然是歌剧. 如果她不高兴,那么她宁愿晚上自己独处. Pat不知道Chris是不是高兴. 但是 Pat 推断Chris以 2/3的概率高兴,以1/3的概率不高兴. Game Theory-Chapter 3
Battle of the sexes with incomplete information (version two) (continued) Chris is happy Pat is happy Pat Opera Fight Chris 2 , 1 0 , 0 1 , 2 Chris is happy Pat is unhappy Pat Opera Fight Chris 2 , 0 0 , 2 0 , 1 1 , 0 Chris is unhappy Pat is happy Pat Opera Fight Chris 0 , 1 2 , 0 1 , 0 0 , 2 Chris is unhappy Pat is unhappy Pat Opera Fight Chris 0 , 0 2 , 2 1 , 1 检查 ((Opera if happy, Opera if unhappy), (Opera if happy, Fight if unhappy)) 是否是一个Bayesian NE Game Theory-Chapter 3
Battle of the sexes with incomplete information (version two) (continued) Game Theory-Chapter 3
Battle of the sexes with incomplete information (version two) (continued) Game Theory-Chapter 3
Battle of the sexes with incomplete information (version two) (continued) Game Theory-Chapter 3
Battle of the sexes with incomplete information (version two) (continued) Game Theory-Chapter 3
Battle of the sexes with incomplete information (version two) (continued) Chris推断 Pat以0.5的概率高兴,以0.5的概率不高兴 Chris is happy Pat (0.5, 0.5) (O,O) (O,F) (F,O) (F,F) Chris O 2 1 F 1/2 Chris is unhappy Pat (0.5, 0.5) (O,O) (O,F) (F,O) (F,F) Chris O 1 2 F 1/2 如果Chris高兴,当Pat选择(Opera if happy, Fight if unhappy), 而Chris选择Fight时Chris的期望收益 Game Theory-Chapter 3
Battle of the sexes with incomplete information (version two) (continued) Pat推断 Chris 以2/3 的概率高兴, 以1/3的概率不高兴 Pat is happy Pat O F Chris (2/3, 1/3) (O,O) 1 (O,F) 2/3 (F,O) 1/3 4/3 (F,F) 2 Pat is unhappy Pat O F Chris (2/3, 1/3) (O,O) 2 (O,F) 1/3 4/3 (F,O) 2/3 (F,F) 1 如果Pat 不高兴,当Chris 选择(Fight if happy, Fight if unhappy) ,而Pat’s选择Opera时Pat的期望收益 Game Theory-Chapter 3
Battle of the sexes with incomplete information (version two) (continued) 检查 ((Fight if happy, Opera if unhappy), (Fight if happy, Fight is unhappy)) 是否是一个贝叶斯纳什均衡. (Y) 检查 ((Opera if happy, Opera if unhappy), (Opera if happy, Fight is unhappy))是否是一个贝叶斯纳什均衡.(Y) 检查 ((Opera if happy, Fight if unhappy), (Fight if happy, Opera is unhappy是否是一个贝叶斯纳什均衡.(N) Game Theory-Chapter 3
3.2 Applications 3.2.A Mixed Strategies Revisited ---pp119-120. 3.2.B An Auction ----pp121-124 Appendix 3.2. B----pp124-125 3.2.C A Double Auction -----pp125-129. Game Theory-Chapter 3
First-price sealed-bid auction (3.2.B of Gibbons) Game Theory-Chapter 3
First-price sealed-bid auction (3.2.B of Gibbons) (continued) Game Theory-Chapter 3
First-price sealed-bid auction (3.2.B of Gibbons) (continued) Game Theory-Chapter 3
First-price sealed-bid auction (3.2.B of Gibbons) (continued) Game Theory-Chapter 3
First-price sealed-bid auction (3.2.B of Gibbons) (continued) Game Theory-Chapter 3
First-price sealed-bid auction (3.2.B of Gibbons) (continued) Game Theory-Chapter 3
3.3 The Revelation Principle 定理(显示原理)任何贝叶斯博弈的任何贝叶斯纳什均衡都可以表述为一个激励相容的直接机制(incentive-compatible direct mechanism ). -----pp129-132. Game Theory-Chapter 3