Misspecified Models in Learning and Games

The model misspecification literature studies biases in anticipating and interpreting information using the following approach. First, fix an incorrect, or misspecified, model of the information environment (e.g., overprecise signals). Then, use this misspecified model to explore how these biases impact decisions, learning, and strategic interactions. This review provides an overview of this literature. We first explore how model misspecification impacts learning. We provide some insight into classic results on misspecified learning in statistics. We then highlight the new long-run outcomes and technical challenges that arise when extending these results to common learning settings in economics, that is, active and social learning where information endogenously depends on action choices. Next, we provide an overview of Berk–Nash equilibrium, an equilibrium concept for strategic interaction with misspecified agents. We show how agents’ biases interact to influence equilibrium actions and beliefs. In closing, we discuss applications of the framework as well as its shortcomings and potential avenues for future research.