Predictive modeling in Finance and Insurance.

The insurance business is characterized by an inverse production cycle, where the cost of a product is unknown at the time of selling.Heavy competition and anti-selection effects require insurance companies to constantly improve their risk classification process and accompanying tariff structures. Insurers therefore rely heavily on predictive models and historical data to accurately assess the underlying claim risk of policyholders in their portfolio. Recent technological advancements pose many opportunities to improve the pricing process with new modeling paradigms and data sources. Artificial intelligence (AI) and, more specifically, machine learning (ML) techniques allow to improve the current statistical model frameworks. Telematics technology makes new sources of information available regarding policyholder behavior. In this thesis we investigate how ML techniques and driving behavior can improve current state-of-the-art motor insurance pricing approaches. We put extra focus on the interpretability of the resulting premium structure. This is a key requirement in the insurance industry due to strict regulations and the need for explainability towards all stakeholders involved.