Modelling telecoms customers' spending patterns & churn

Abstract

The dissertation focuses on the use of Markov Chains in discrete and continuous version to model consumer spending patterns. In particular the survival model is applied to model churn as a two-state process with corresponding statistical measures captured through the hazard function and other related quantities. An enlarged continuous Markov Chain Model is developed to take into account purchase recency, monetary value and credit time-window. These stochastic processes have accompanying statistical inference problems related to estimation and goodness of-fit which are also considered in the text. Finally the dissertation applies the theory to billing data of pre-paid customers provided by Vodafone (Malta) Ltd. A number of observations are made relative to customer short-term and long-term spending behaviour from the results obtained. The model is also used to estimate the Customer Lifetime Value.