Estimating mixture models in consumer segmentation

Abstract

Mixture models are used in many fields to identify different sources of uncertainty. Market demand is an accumulation of each individuals' choice probabilities. Consumers with different preferences will be have different choice models. It is thus required to estimate the underlying choice models and the mixture proportion to accurately predict true market demand. Various attempts in literature struggled to find a balance between prediction accuracy and model interpretability. This paper is motivated by an algorithm in a recent paper which proposes a non-parametric estimation method based on the Frank-Wolfe algorithm to segment consumers and further apply the calibrated consumer segmentation to a price optimization problem, an important application in revenue management.