Demand Estimation with Missingness of Product Availability
Open Access
Author:
Saram, Thilini
Graduate Program:
Statistics
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
March 18, 2021
Committee Members:
David Hunter, Thesis Advisor/Co-Advisor Prabhani Kuruppumullage Don, Committee Member Ephraim Mont Hanks, Program Head/Chair
Keywords:
Discrete choice models EM algorithms Missing variables
Abstract:
Discrete choice models predict the choices among two or more discrete alternatives. We discuss some existing models but focus on the Multinomial Choice Model (MNL) and explain Expectation-Maximization (EM) algorithms. We provide evidence that failing to account for product availability leads to bias in demand estimates and use an illustrative example to demonstrate this. We propose a new model accounting for product availability. To accomplish this, we use EM algorithms and direct optimization of observed data log-likelihood for estimating maximum likelihood estimates by introducing product availability as a missing variable. We use a simulation study to compare the models' prediction accuracy and fit the new model to the illustrative example.
