Lecture-Rputation Games.docVIP

Lecture 2-Reputation Games Markus M. M?obius April 29, 2000 1 Introduction Some famous economist once said: ?All we know about reputation is that it takes a long time to build and that it is easily lost.? Economists started to model reputation in the 1980s. Agents develop reputation by playing repeatedly in the same way in the expectation that agents will finally expect him to play this way. In other words, agents build reputations in order to behave as if they are committed to an action. 一些著名的经济学家曾经说过：“我们都知道口碑是 这需要很长的时间来建立，并且它很容易丢失。“经济学家开始 声誉在上世纪80年代建模。代理开发的声誉打 反复在预期代理将最终希望以同样的方式 他玩这个样子。换句话说，代理商，以建立信誉 行为就好像他们正在致力于一个动作。 In a complete information world reputation building is not credible. Agents in the finitely repeated prisoner’s dilemma will always defect at any stage of the game in the unique SPE. Even after cooperating for many periods my opponent will still defect because SPE solves the game backwards. regardless of the history the players found themselves in. 在一个完整的信息世界声誉的建筑是不可信的。代理 在有限次重复囚徒困境总是会在任何阶段缺陷 游戏中的惟一的SPE。即使合作了很多时期我 对手仍然会背叛，因为SPE解决了比赛向后。无论 历史的玩家发现自己英寸 We therefore want a model which is not solvable through backward induction, i.e. a game of incomplete information. More precisely, we want some uncertainty about an agent’s type (such as some probability of being a ?crazy? cooperator, for example). A long-lived player can then take advantage of this small potential reputation (probability) of being ?crazy? to commit himself to play like the crazy type. In some ways the rational player can ?hide? behind the crazy type. 因此，我们希望有一个模型，是不可解通过逆向归纳法， 即一个游戏不完整的信息。更确切地说，我们需要一些 关于代理的类型（比如是一个“疯狂”的一些不确定性的概率 合作者，例如）。然后，长寿命的玩家可以利用这 小的潜在的被疯狂承诺自己的声誉（概率） 喜欢玩疯狂的类型。在某些方面，理性的玩家可以“隐藏”的背后 疯狂的类型。 In reputation models there are two paradigms: (a) one long-lived player who faces many short-lived opponents; and (b) many long-lived players. The latter case is exemplified by the Prisoner?s Dilemma game. The prototypical example for the first case is the chain-stor