Combining Aggregate and Plant-Level Data to Estimate a Discrete-Choice Demand Model

This paper builds on the methodology developed by Katayma, Lu and Tybout (2003), who use a nested logit demand model to estimate demand parameters from plant-level data that usually report only revenue and cost figures. I demonstrate how to extend their framework by including the extra information provided by commonly available data on aggregate physical output. Using data from the Colombian beer industry from 1977 to 1990, the model, estimated through Bayesian Monte Carlo Methods, shows a sizeable precision gain in the parameter estimates once the aggregate variable is included.