A Moment Inequality Approach to Estimating Multinomial Choice Models with Unobserved Consideration Sets∗ (Job Market Paper)

†I am indebted to my advisors Amit Gandhi and Xiaoxia Shi for invaluable guidance and support on this project. I would also like to thank Ken Hendricks and Alan Sorensen for insightful discussions and detailed comments. Conversations with Jean-Pierre Dubé, Matt Gentry, Han Hong, Francesca Molinari, Yanqin Fan, Marc Rysman have improved the paper. I am grateful to the participants at the 2014 North American Summer Meeting of the Econometric Society, as well as the seminar participants at Wisconsin-Madison, Wisconsin-Whitewater for their many helpful comments and questions. 1 This paper proposes a moment inequality approach to estimating random utility models when consumers ’ consideration sets are unobservable to econometricians. I show that, without relying on a specific model of consideration set formation, the random utility model can be identified and estimated via a system of conditional moment in-equalities derived from the utility maximization assumption. Monte Carlo experiments demonstrate that the method can correct the bias caused by misspecified consideration set generation. I apply the moment inequality approach to study whether attention inertia can explain some of the observed persistence in consumers ’ brand choices, as opposed to alternative explanations in terms of preference, e.g., state-dependent utilities. The estimation results, obtained using household scanner data, show that up to 20 % of the observed persistence, in terms of re-purchase probability, can be attributed to the fact that previous purchase of a brand increases its present consideration probability