Yuxin Chen and Sha Yang
Executive Summary
The previous two decades have witnessed numerous applications of demand models that employ household panel data and aggregate sales data to understand consumer purchase behavior and response to price and promotion. Compared with aggregate data, household panel data are desirable for studying individual consumer choice behavior. However, they are often available only for a small number of product categories, stores, regions, and weeks, which limits the decision makers' ability to obtain relevant information.
Despite the lack of household panel data, store managers may still want to understand microlevel consumer behavior, such as purchase dynamics based on store-level aggregate information. This is challenging in the following two ways. First, a sales response model may not address this problem adequately, because it does not have a microlevel underpinning of modeling individual consumer behavior patterns. Second, it is computationally demanding to recover the purchase dynamics because consumer purchase history information is not directly observable from the aggregate information.
In this article, the authors develop a Bayesian method of modeling disaggregate consumer behavior with aggregate data. The basic idea is to augment the market share data to obtain the purchase history of a simulated panel of households, based on which micromodels are estimated. A key advantage of this method is the ability to model the impact of purchase history on consumer choice decisions with aggregate data. Because individual choices are not made available with existing estimation methods without using data augmentation, the aforementioned effect will be difficult to measure.
The proposed method is especially useful to firms that need a better understanding of microlevel consumer purchase dynamics and accurate estimates of brand price competition at the store level when high-quality panel data are not available. The approach developed in this article retains the benefits of discrete choice models, which provide a structural and parsimonious way of modeling consumer purchase behavior, and ensures that the estimates of own- and cross-elasticities are of the correct sign. As with all Bayesian methods, the proposed method facilitates exact, finite-sample inference and does not rely on asymptotic results.
To demonstrate the validity of the proposed method, the authors conduct a series of simulations, in which they generate the aggregate brand shares from individual consumers' choices given the prespecified response parameters and distribution of heterogeneity. Using the proposed method, they can recover the true model parameters. They then apply the method to an empirical application using store-level data of consumer purchases of refrigerated orange juice. They find a significant impact of last-period purchase on current period purchase.
Biography
Yuxin Chen is Associate Professor of Marketing in the Stern School of Business at New York University. He holds a BS in Physics from Fudan University, an MSBA from Washington University in St. Louis, and a PhD in Marketing from Washington University in St. Louis. Before becoming interested in Marketing, he studied computer science in the graduate school of Zhejiang University. His primary research areas include database marketing, Internet marketing, pricing, Bayesian econometric methods, and marketing research. He has previously published in Marketing Science, Management Science, and Quantitative Marketing and Economics. He won the Frank M. Bass Dissertation Paper Award and the John D.C. Little Award. He is on the editorial boards of Journal of Marketing, Journal of Marketing Research, and Marketing Science.
Sha Yang is Assistant Professor of Marketing in the Stern School of Business at New York University. She received her PhD in Marketing, MS in Statistics, and MA in Economics from Ohio State University. Her primary research focuses on modeling consumer behavior and market competition and developing Bayesian econometric methods in the analysis of customer- and firm-level data. Her research has appeared in various journals, including Marketing Science, Journal of Marketing Research, Quantitative Marketing and Economics, Journal of Retailing, Marketing Letters, and International Journal of Forecasting.
Journal of Marketing Research, Vol. XLIV, No. 4, November 2007
View Table of Contents.