Learning and Experimentation in Games

近年来，非合作博弈论，特别是纳什均衡分析，已被应用于许多经济情况的研究。随着各种各样的应用而来的是持续不断的批评：为什么或何时平衡分析是合适的？均衡从何而来？人们如何在众多均衡中做出预测？这个项目使用两种不同的方法来解决这些问题。在第一种情况下，假设玩家遵循有限理性模型精神下的特定行为规则。第二步，在给定参与者对对手策略的外生先验信念的情况下，将参与者的预测和行为作为动态优化问题的解。这两部分的一个共同主题是，即使玩家寿命很长，学习也不一定会导致纳什均衡，除非他们也参与了足够多的实验。本课题提出的基于学习的均衡分析论证是一个非常重要的贡献。它为广泛使用的分析工具提供了更好的基础，确定了经济理论中新的和有前途的研究方向，并通过实验经济学对现有理论的检验来激励经济理论的研究。//

In recent years, noncooperative game theory, and especially Nash equilibrium analysis, has been used in the study of many economic situations. Along with the many and varied applications has come persistent criticism: Why or when is equilibrium analysis appropriate? Where do equilibria come from? How does one choose among many equilibria in making predictions? This project addresses these questions using two different approaches. In the first, players are assumed to follow ad-hoc behavioral rules in the spirit of models of bounded rationality. In the second, the players' forecasts and behavior are obtained as the solution to a dynamic optimization problem given their exogenous prior beliefs about the opponents' strategies. A common theme to both parts is that learning need not lead to a Nash equilibrium even when players are long-lived, unless they also engage in a sufficient amount of experimentation. The learning-based justification of equilibrium analysis developed in this project is a very significant contribution. It provides a better foundation for widely-used analytic tools, identifies new and promising lines of research in economic theory and motivates research in economic theory with the tests of existing theories in experimental economics.//