Website CV Paper
Fields
Theory
Job market paper
Compatible Explanations
This paper studies which steady-state outcomes in misspecified games are compatible with learning from observations and reasoning about Bayesian rational opponents. It connects Berk–Nash equilibrium, a learning-based solution concept for misspecified games, with rationalizable conjectures. Under common knowledge of the game structure and subjective models, rationalizable Berk–Nash equilibrium requires each player to have a supporting belief that recursively justifies conjectured opponent behavior. Each conjectured strategy must be optimal under a belief supported on the opponent's best-fitting hypotheses for the observations she would receive if the player's hypothesis were correct. We characterize equilibrium through families of compatible explanations and provide an epistemic characterization in introspective type structures representing higher-order beliefs about opponents' inference and choice. Every Nash equilibrium of the objective game survives when both players' models contain the true state distribution. When players share such a model, the refinement coincides with the corresponding rationalizable conjectural equilibrium. With misspecified models, mistaken explanations can survive despite positive statistical losses, but equilibrium need not exist. Nonexistence implies an incompatibility between statistical inference and strategic reasoning under the given models, motivating model revision or relaxation of the requirement that conjectured opponent behavior be justified by Bayesian rationality at every order.