Non-parametric Identification and Estimation of Demand and Preferences using Scanner Data

This thesis presents three essays on the partial non-parametric identification and estimation of demand and preferences using scanner data. In the first essay, I consider a critical real-world problem: the formal identification and estimation of food stamp fraud in the United States. It is shown that consumption choices can be used to bound fraud. The estimation procedure in this essay uses a standard assumption. In particular, it assumes that there exists a conditional quantile of consumption that coincides with an individual demand function. This assumption is strong, but useful for the structural estimation problem at hand. In the second and third essays, I propose flexible methods for estimating demand and preferences using scanner data that do not require this strong assumption. In the first proposed method, heterogeneity across consumers is introduced by assuming that the marginal rate of substitution is a random field. In such an environment, the theory of generalized functions can be used to test the integrability of expected demand at a parametric rate. If variation in preferences is small, then preferences can be recovered by approximating the relationship between preferences and demand with a first-order expansion and applying an analogue of the delta-method. In the second proposed method, individual-level heterogeneity is characterized by a distribution, and the heterogeneity across consumers is characterized by a Dirichlet process F over such distributions. Two frameworks for estimation are considered: a Bayesian framework in which F is known, and a hyperparametric (or empirical Bayesian) framework in which F is a member of a parametric family. Both methods are illustrated by applications to alcohol consumption.Ph.D