Risk Prediction in Automobile Insurance

Abstract

The relationship between individual risks and the number of claims in automobile insurance has received growing attention over recent years among industry practitioners. The number of people who purchase cars grows exponentially. Risk managers of insurance companies need to deal with a variety of risks concerning vehicles and try to predict them as accurate as possible to avoid losses or to minimize them. For this reason, in this thesis I will examine different risk variables and models for these variables in the car insurance sector. The analyses are done based on the econometric approach. The main problem of the thesis concerns the risk prediction by count data models in the automobile insurance. First, the analysis will be done by Poisson regression model using maximum likelihood estimator. Then the Gamma heterogeneity will be taken into account and the negative-binomial regression model will be discussed. The latter provides a framework for the bonus-malus scheme. The extensions of Poisson regression will also be considered in this thesis. The models will be applied to the real data and the estimation results will be discussed at the end of the thesis.