Recovering Stockkeeping-Unit-Level Preferences and Response Sensitivities from Market Share Models Estimated on Item Aggregates
Published 5/1/2005
Author: David R. Bell, André Bonfrer, and Pradeep K. Chintagunta
View this content
Executive Summary
Category assortments often consist of several hundred stockkeeping units (SKUs), and this poses a considerable challenge for modelers of market response and for retail managers who decide category offerings. Studying consumer behavior across the full set of alternatives necessarily requires models that can be nonparsimonious and computationally intensive. Two existing options are available to the analyst who wishes to uncover true SKU-level parameters. The first is to estimate a model with a complete set of fixed effects, one for each SKU. Implementation with usual marketing data is difficult because the modeler quickly loses degrees of freedom. Furthermore, problems may arise when there is volatility in the underlying choice set due to entry and exit of SKUs in the category. A less serious issue is that the computational requirements for such a nonparsimonious model can be quite significant, especially when constructing multiple models over many categories or when running simulations that require repeated estimation of the models.
The second option is to use a pure characteristics–based approach of the type that Fader and Hardie introduced to the household panel data literature. Fader and Hardie exploit the notion that categories with several alternatives can usually be described according to a substantially smaller stable set of attributes (e.g., brand, size, flavor, form). Their method avoids the ad hoc aggregation of SKUs into choice composites, an approach still prevalent in individual-level choice models that populate the marketing literature. A third option (the one that the authors pursue in this article) is to develop analytical relationships between the complex model and simpler aggregated models and to exploit them to obtain the required parameters.
This research makes the following connection: An attribute-level model in which the unit of analysis is the market share for an alternative that is created by aggregation (e.g., Colgate toothpaste) is distinguished from a truly disaggregate SKU-level model, and an analytical relationship between parameters obtained from these two models is established. The authors show that SKU-level parameters can be recovered by calculation from estimated attribute-level parameters, circumventing the need for direct estimation of the more complex SKU-level model.
The authors calibrate the store data market share model using 98 weeks of data for ten brands and 168 SKUs of toothpaste. Instead of estimating 168 preference parameters (when there is an “outside” alternative in addition to the 168 “inside” ones), it is necessary to estimate only ten brand preference parameters from which the 168 parameters can be computed, as long as share and marketing-mix data are available at the SKU level. Covariate effects, such as marketing-mix response parameters, can be recovered in a similar fashion. Holdout tests demonstrate superior predictive performance, and the authors discuss implications for the derivation of elasticities for new SKU introductions. The authors intend to examine applications of this modeling approach in future work, particularly in the areas of assortment planning and product line trimming.
Biography
David R. Bell is Associate Professor at Wharton School, University of Pennsylvania. He holds a doctoral degree in business and a master’s in Statistics from Stanford University. His current research focuses on models of consumer response to marketing stimuli with applications to problems in retailing.
André Bonfrer is Assistant Professor at the Lee Kong Chian School of Business, Singapore Management University. He holds a doctoral degree in Business and an MBA from the Graduate School of Business, University of Chicago. His current research focuses on empirical modeling in retailing, competition, and customer valuation.
Pradeep K. Chintagunta is Robert Law Professor of Marketing at the Graduate School of Business, University of Chicago. He has a doctoral degree in marketing from Northwestern University. He is interested in studying packaged goods, technology, and pharmaceutical product markets.
J Marketing Research, Volume 42, Number 2, May 2005
View Table of Contents.