Essays on the Interaction of Product Characteristics, Consumer Heterogeneity, and Information

Abstract

This dissertation addresses three related questions. The second chapter models a monopolist with private information about her product, over which consumers have heterogeneous preferences. Consumers may incur an audit cost to acquire this information before purchase, and the monopolist may signal this information via price. Equilibrium analysis yields two surviving equilibria: a separating equilibrium with price signaling and a pooling equilibrium with consumer audit. Welfare analysis reveals that either equilibrium may be preferable, depending on the value of the model's parameters. The third chapter assess whether or not a proxy variable for individual consumers' concern for their health is valuable in the estimation of demand for products in which a brand's ``healthiness' may be of concern to consumers. A random-coefficients logit model is applied to scanner data across 28 U.S. cities, together with demographic data from the Current Population Survey (CPS) and the Behavioral Risk Factor Surveillance System (BRFSS), to estimate the coefficients of the interaction between health-related product characteristics and a proxy variable for health conscientiousness in the estimation of demand for ready-to-eat cereal. The results suggest that the interaction of tobacco use, as a proxy for a consumer health conscientiousness, may be a worthwhile addition to the commonly collected demographic information used in the estimation of demand. The final chapter addresses the problem of estimating the multiple sets of preferences that underlie the purchasing behavior of a household, and doing so in such a way as to retrieve actionable household-level information about preferences. A hierarchical Bayesian finite mixture model is proposed as a solution, wherein the individual purchases of each household are the result of one of finitely many underlying distributions for preferences. Results applying the model to simulated household panel datasets demonstrate the ability of the model to converge to the true underlying parameters of such data.