Flexible demand estimation and zero market shares

This paper develops a flexible discrete-choice demand framework for aggregate data sets that extends Berry, Levinsohn, and Pakes (1995) and the Pure Characteristics Demand Model of Berry and Pakes (2007). I provide a simple, computationally tractable, asymptotically normal estimator based on two contributions: a globally-convergent algorithm to recover utilities from observed demand and a Quasi-Bayes approach that minimizes simulation variance. The framework accommodates zero market shares, which are a challenge for alternative approaches. I show that zeros in demand generate an endogenously censored model, which leads to moment inequalities. As an application, I study moving costs US internal migration data.