A Simple Method to Estimate Discrete-type Random Coefficients Logit Models

Abstract

This paper proposes a new method for estimating random coefficients logit models using aggregate data. The method is applicable for models with discrete-type heterogeneity in consumer tastes when additional data on the total sales for each consumer type are available. The type-level data do not have to be divided by product. The method analytically obtains the value of the econometric error term and thus does not require numerical calculations, such as the contraction mapping established by Berry, Levinsohn, and Pakes (1995). Consequently, the method no longer suffers from problems due to numerical errors in the contraction mapping, including lack of convergence and incorrect parameter estimates. Moreover, the computation time is drastically reduced.