CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: A Bayesian structural uncertainty model to target rebates to consumers with correlated preferences Authors:  Bikram Karmakar - University of Florida (United States)
Ohjin Kwon - Central Connecticut State University (United States)
Gourab Mukherjee - University of Southern California (United States)
Sivaramakrishnan Siddarth - Marshall School of Business, USC (United States) [presenting]
Abstract: A spatial autoregressive multinomial probit model is proposed and estimated, in which consumers product preferences are correlated based upon their close they are to each other. The proposed model uses a Bayesian structural uncertainty approach to combine multiple sources of such contiguity information and also incorporates consumer response heterogeneity. The model is applied to the unique problem of improving the efficacy of promotional programs that offer targeted conquesting and loyalty discounts to consumers, which is common in the auto industry but unstudied in the marketing literature. Model calibration on automobile transaction data from the Los Angeles market confirms that previous purchases made by consumers are predictive of the future purchases of other consumers. Targeted discounts derived from the proposed model for conquesting and loyalty promotional programs substantially increase manufacturer profits. We demonstrate that the extant method of using a linear combination of the individual weight matrices provides an inferior fit and lower incremental profits than the proposed Bayesian structural uncertainty approach to information assimilation.