Website CV Paper
Fields
Industrial Organization, Econometrics
Job market paper
Testing the Substitution Predictions of Demand Models: Evidence from the Blue Bell Recall
Demand models are a workhorse of most empirical industrial organization analyses and can play an important role in the analysis of antitrust, marketing, and consumer protection questions, but evaluations of how well demand models perform when applied to real-world counterfactuals are limited. I use the 2015 Blue Bell ice cream recall, which removed a major regional brand from grocery shelves for at least four months, to test how a demand model's specification and identification strategy affect its prediction of where the brand's demand went. First, I systematically examine the out-of-sample fit of 64 specifications, from logit to the random-coefficients nested logit, by comparing their predictions with the observed changes in rival sales. Each specification is estimated on data from before the withdrawal. Second, I hold the specification fixed at the random-coefficients nested logit and examine how alternative identification strategies affect its out-of-sample fit. In the markets where Blue Bell had its largest share of sales, nests improve the prediction in every combination of the four components of consumer heterogeneity. Adding random coefficients to the nested logit improves predictions further, most clearly at the household level. With the BLP instruments augmented by macro instruments computed within each product group, the random-coefficients nested logit predicts rival recapture close to the observed level, and its out-of-sample fit is better than with the BLP or differentiation instruments. With this augmented set, household micro moments improve its prediction further.