被诅咒的理性化
Cursed Rationalizability
- The Hebrew University of Jerusalem（耶路撒冷希伯来大学）
- Harvard University（哈佛大学）
机构由 AI 辅助整理，请以论文原文为准。
AI总结：
提出被诅咒理性化解概念，分离偏差与均衡，在战略互补下给出行为界限，并在实验室博弈中提供精确预测。
AI中文摘要：
被诅咒的玩家理解对手行为的分布以及与其自身信息相关的收益状态分布，但他们的反应却仿佛这两者是相互独立的。被诅咒的均衡额外要求关于行为分布的统计上正确的信念。我们提出了一种将偏差与均衡分离的解概念：被诅咒的理性化迭代地删除每一个不是对对手幸存行为猜测的被诅咒反应的行动。被诅咒理性化集合是一个虚拟博弈的中间相关理性化集合，在该虚拟博弈中，收益在状态上取平均。在战略互补下，被诅咒的理性化产生与被诅咒均衡相同的行为界限。一个谨慎的细化，即可容许性的类比，在记录了赢家诅咒的实验室博弈中提供了精确的预测。
英文摘要：
Cursed players understand the distribution of opponents' actions and the distribution of the payoff-relevant state, conditional on their own information, but respond as if the two were independent. Cursed equilibrium additionally requires statistically correct beliefs about the distribution of play. We propose a solution concept that separates the bias from equilibrium: cursed rationalizability iteratively deletes every action that is not a cursed response to a conjecture about opponents' surviving play. The cursed-rationalizable set is the interim correlated rationalizable set of a virtual game, in which payoffs are averaged over the state. Under strategic complements, cursed rationalizability yields the same bounds on behavior as cursed equilibrium. A cautious refinement, the analog of admissibility, delivers sharp predictions in the laboratory games that documented the winner's curse.
