52 Pages Posted: 13 Feb 2017 Last revised: 20 Jan 2019
Date Written: January 17, 2019
Marketing research relies on individual-level estimates to understand the rich heterogeneity that exists in consumers, firms, and products. While much of the empirical marketing literature has focused on capturing static cross-sectional heterogeneity, little research has been done on modeling the temporal evolution of cross-sectional heterogeneity. In this work, we develop a framework that is capable of flexibly modeling dynamic heterogeneity through individual-level latent random functions, which are estimated with Bayesian nonparametric Gaussian processes. This novel specification, which we term Gaussian process dynamic heterogeneity (GPDH), flexibly captures both the evolution of population trends and, more importantly, individual-level departures from those trends over time. Similar to classical random effects or random coefficients specifications, GPDH allows for sharing of statistical information across individuals. However, GPDH also allows for information sharing within individuals over time, thus providing rich individual-level insights and statistically efficient inferences regarding dynamics. Theoretically, we show the mathematical links between GPDH and existing heterogeneity specifications, and illustrate the bias that can arise by not capturing the temporal evolution of heterogeneity. Substantively, we showcase the utility and versatility of GPDH through two applications: estimating individual-level preference dynamics for consumer packaged goods during the Great Recession, and estimating the product-level evolution of text in online reviews. Across both applications, we find robust evidence of dynamic heterogeneity, and illustrate the rich managerial insights that come from these individual-level estimates of dynamics, with applications to targeting, pricing, market structure analysis, and predicting product lifecycles.
Keywords: Dynamics, Heterogeneity, Bayesian nonparametrics, Gaussian processes, Choice models, Topic models
JEL Classification: C01, C11, C14, C23, M37
Suggested Citation: Suggested Citation