Simple Two-Stage Inference for A Class of Partially Identified Models

Abstract

This note proposes a new two-stage estimation and inference procedure for a class of partially identified models. The procedure can be considered an extension of classical minimum distance estimation procedures to accommodate inequality constraints and partial identification. It involves no tuning parameter, is nonconservative and is conceptually and computationally simple. The class of models includes models of interest to applied researchers, including the static entry game, a voting game with communication and a discrete mixture model.