Robust multiplicity with a grain of naiveté

Rationalizability is a central concept in game theory. Since there may be many rationalizable strategies, applications commonly use refinements to obtain sharp predictions. In an important paper, Weinstein and Yildiz [2007] show that no refinement is robust to perturbations of high-order beliefs. We show that robust refinements do exist if we relax the assumption that all players are unlimited in their reasoning ability. In particular, for a class of models, every strict Bayesian-Nash equilibrium is robust. In these environments, a researcher interested in making sharp predictions can use refinements to select among the strict equilibria of the game, and these predictions will be robust.