Sieve BLP: A Semi-Nonparametric Model of Demand for Differentiated Products

Abstract

We develop a semi-nonparametric approach to identify and estimate the demand for differentiated products. The proposed method adopts a random coefficients discrete choice logit model (i.e., mixed logit model) in which the distribution of random coefficients is nonparametrically specified. Our method minimizes misspecification error in this distribution to which routinely used parametric approach is subject. In addition, it overcomes the practical challenge of dimensionality in the number of products that remains the main hurdle in the nonparametric estimation of demand functions. We propose a sieve estimation procedure (referred to as sieve BLP) that remains simple to implement. Extensive Monte Carlo simulations show its robust finite-sample performance under various data generating processes. We use our method to investigate the welfare implications of a sugar tax in the ready-to-eat cereal industry in the US. This application underscores the usefulness of sieve BLP due to its ability to allow for flexibly specified individual heterogeneity in demand, especially when the researcher aims to quantify the distributional effects of a policy change.