Essays in the Dynamics Bayesian Models in Marketing

Abstract

In this dissertation, I explore the expanding area of interest in the field of business analytics: the extraction of meaningful association structures and their connection to applicable business theory. In particular I develop a series of models based on dynamic Bayesian methods to address some of the problems not yet addressed in the marketing literature in regards to the market basket analysis.

In the first essay, I propose a Bayesian network method to model the joint association structure between product categories in a multi-category setting and also incorporate the effects of marketing-mix variables, demographic heterogeneity and seasonal effects. I show that our approach is able to handle the large joint association structure arising from multiple product categories. I also illustrate the evidence of heterogeneous effect arising from various promotion strategies for different demographics and also find that the effects of time should not be modeled as an exogenous factor.

In the second essay, I extend our Bayes network based approach to tackle the dynamic problem. I discuss the benefits of defining the problem in a dynamic setting in detail and show how the dynamic Bayes network can be used to incorporate time into the system. I also show how promotion and purchase decisions in one time period affect future consumer purchase behavior.

These two essays together illustrate a comprehensive approach in analyzing different association structures in market basket data and to learn various effects of promotion. One can utilize the methods discussed in the first essay for general inference purposes and to identify product clusters for dynamic analysis, described in the second essay.