STATISTICAL AND MACHINE LEARNING METHODS FOR MULTIVARIATE PROBLEMS IN MARKETING

Other Titles
Abstract

My dissertation investigates multivariate response problems in marketing applying both parametric and non-parametric approaches. In marketing research there are several empirical situations that examine multivariate responses. Examples include predicting customer’s inter-related decision making to transact on multiple channels, multiple firms’ inter-related decision to set prices or enter a market. In my first essay, I examine a multivariate problem on customers’ decision to transact with multiple channels of a firm using a parametric state-space model. I define a customer’s channel engagement as a latent semi-Markovian process conditional upon which she decides her multichannel activities. I model the channel activities of website visits, and online and offline purchases using a parametric specification. My research jointly predicts the customer’s online visitation behavior, and online and offline purchase propensity. Further, the framework recovers the customer’s underlying engagement state with each channel and the expected duration of each state. The second and third essays of my dissertation are motivated by the restrictions imposed by parametric methods. In particular, when the response vector is of higher order (> 3) it becomes difficult to parametrically specify the multivariate distribution. Further, parametric methods are ineffective when the dimensionality (or number of covariates) is large and there are more complex interactions, and the response outcomes are sparse. In my second essay, I use the non-parametric multivariate random forests to develop a variable selection procedure for high dimensional problems. I develop new variable importance measures for dimensionality reduction using a recursive feature elimination strategy. In my empirical application on an ecology dataset with sparse observations I find that the proposed measures have higher prediction accuracy than the extant ones. In my third essay I apply the proposed variable selection method for covariate extraction in a high dimensional marketing application. Here, I examine the inter-related price change decisions of multiple sellers on the Amazon marketplace. I model a series of multivariate regression models using the extracted covariates and compare their predictive performance against the embedded variable selection method of LASSO. I find that the generalized additive model trained on the extracted features outperform LASSO. Further, I provide interpretations of the underlying relationship between the predictors and the multivariate outcome.

Journal / Series
Volume & Issue
Description
Sponsorship
Date Issued
2019-05-30
Publisher
Keywords
Hidden semi-Markov models; Hierarchical Bayesian methods; Multivariate models; Multivariate random forests; Recursive feature elimination strategies; Variable importance measures; Marketing; Computer science; Statistics
Location
Effective Date
Expiration Date
Sector
Employer
Union
Union Local
NAICS
Number of Workers
Committee Chair
Kadiyali, Vrinda
Committee Co-Chair
Committee Member
Gupta, Sachin
Hooker, Giles J.
Degree Discipline
Management
Degree Name
Ph.D., Management
Degree Level
Doctor of Philosophy
Related Version
Related DOI
Related To
Related Part
Based on Related Item
Has Other Format(s)
Part of Related Item
Related To
Related Publication(s)
Link(s) to Related Publication(s)
References
Link(s) to Reference(s)
Previously Published As
Government Document
ISBN
ISMN
ISSN
Other Identifiers
Rights
Rights URI
Types
dissertation or thesis
Accessibility Feature
Accessibility Hazard
Accessibility Summary
Link(s) to Catalog Record