Berk-Nash Rationalizability (2505.20708v2)

Abstract: Misspecified learning -- where agents rely on simplified or biased models -- offers a unifying framework for analyzing behavioral biases, cognitive constraints, and systematic misperceptions. We introduce Berk--Nash rationalizability, a new solution concept for such settings that parallels rationalizability in games. Our main result shows that, with probability one, every limit action -- any action played or approached infinitely often -- is Berk--Nash rationalizable. This holds regardless of whether behavior converges and offers a tractable way to bound long-run behavior without solving complex learning dynamics. We illustrate this advantage with a known example and identify general classes of environments where the rationalizable set can be easily characterized.