Predicting Customer Churn in Non-life Insurance

Specialeforsvar: Natascha Vestergaard

Titel: Predicting Customer Churn in Non-life Insurance
A Multi-State Markov Model Approach

Abstract: Customer churn is the non-renewal of an insurance contract and is an expensive problem to most non-life insurance companies. This gives rise to the need of customer churn analysis in the non-life insurance industry, such that the insurance companies can determine the drivers of customer churn. Traditionally, this analysis has been conducted as a binary outcome analysis, considering only churn and not churn, and considering only a single insurance period. This thesis aims to expand this analysis to consider multiple states and multiple time periods. The policyholder behaviour is treated as a multi-state Markov chain, and the possibility of a higher-order Markov property is investigated. We will present different ways of modelling the transition intensities, including a parametrically specified multiplicative intensity model, kernel estimation and a semiparametric
Cox’s proportional hazards model. Finally, the use of a Cox’s proportional hazards model will be illustrated through an application of the model on a data set provided by Tryg Forsikring A/S. The application reveals how different factors affect the transition intensity between the states of the model, but also reveals the challenges of the model with a data set of the given size.

Vejleder: Munir Hiabu
Censor: Martin Møller Svensson,