Robust Robustness

Abstract: The maxmin approach to distributional robustness evaluates each mechanism according to its payoff guarantee over all priors in an ambiguity set. We propose a refinement: the guarantee must be approximately satisfied at priors near the ambiguity set (in the weak topology). We call such a guarantee robust. The payoff guarantees from some maxmin-optimal mechanisms in the literature are not robust. We show, however, that over certain standard ambiguity sets (such as continuous moment sets), every mechanism's payoff guarantee is robust. We give a behavioral characterization of our refined robustness notion by imposing a new continuity axiom on maxmin preferences.