Nonparametric Analysis of Monotone Choice

Abstract

We develop a nonparametric approach to test for monotone behavior in optimizing agents and to make out-of-sample predictions. Our approach could be applied to simultaneous games with ordered actions, with agents playing pure strategy Nash equilibria or Bayesian Nash equilibria. We require no parametric assumptions on payoff functions nor distributional assumptions on the unobserved heterogeneity of agents. Multiplicity of optimal solutions (or equilibria) is not excluded, and we are agnostic about how they are selected. To illustrate how our approach works, we include an empirical application to an IO entry game.