Segmentation of Financial and Marketing Data: Mixture Logit model and Hidden Markov Model
thesisposted on 2012-12-14, 00:00 authored by Ziqian Huang
Segmentation refers to the assignment of each consumer to a set of similar consumers. The formation of the sets and the assignments are done simultaneously in an algorithm. We focus on utilizing the Mixture Logit Model (MLM) and the Hidden Markov Model (HMM) to segment financial and marketing data. Both the MLM and the HMM originate from the Finite Mixture Model (FMM). The MLM is also called the Mixture Logistic Regression (MLR). Traditionally, cluster analysis, including the K means algorithm and hierarchical methods, have been used as segmentation methods. Cluster analysis is unsupervised learning, in that there is no target variable. In the marketing and finance areas, segmentation results are often considered as one of the important inputs for modeling a target variable, because of the existence of different underlying market segments, both in theory and in reality. Our proposed segmentation methods begin with modeling a target variable, instead of unsupervised learning, and then the existence of segments is evaluated through certain model selection criteria. The characteristics of each segment are shown from their parameter estimates, and further the segments can be profiled by other variables which are not used in the model. This research makes a contribution by illustrating how to segment financial and marketing data objectively and systematically, with regard to incorporating the segmentation into the supervised modeling. We apply the MLM on one marketing solicitation responder dataset, the HMM on the S&P 500 monthly return data, and on one charity donation dataset. All of the results demonstrate that the MLM and the HMM perform better than the benchmark models or methods.