Marketing Letters, 2015, 1-16.
Joseph Pancras (Marketing) and Dipak K. Dey (Statistics) Co-author: Xia Wang (University of Cincinnati)
Targeted marketing is increasingly popular among new media firms and accurate targeting requires well-calibrated statistical models which will identify customer preferences from their previous historical transactions so as to customize an offering to their needs. A typical example of such targeted marketing is customized pricing, where a price sensitive customer is given a coupon with a higher face value, while a less price sensitive or brand loyal customer may be given a lower face value or no coupon at all. This is called ‘price discrimination’ and statistical models used to identify preferences to implement price discrimination have so far ignored the role of stochasticity (or randomness) in customer behavior. This paper studies the impact of customer stochasticity on firm price discrimination strategies by developing a new model termed the Bayesian Mixture Scale Heterogeneity (BMSH) model. The BMSH model incorporates both parameter heterogeneity and customer stochasticity, and the robustness of the model has been established using extensive simulations. Application of the new model to consumer yogurt purchases finds that compared to the benchmark mixed logit and multinomial probit models, this model shows that markets are less price elastic, and that a majority of customers exhibit stochasticity in purchases. The model also obtains better prediction and more profitable targeting strategies.