Pricing strategies in automobile insurance: optimization and experimentation.

Summary The auto insurance industry is facing a number of regulatory, financial, behavioral and technological changes. In order to face the challenges resulting from these changes and maintain their profitability, insurers must innovate in terms of pricing. In this context, we develop in this thesis two themes related to automobile insurance pricing. The first theme is based on the optimization of pricing strategies, both in underwriting and renewal. The second theme is oriented towards the use of experiments in order to better understand the determinants of insurance demand. We illustrate how empirical demand models based on data available to the insurer can be used to optimize profitability and customer retention. We then extend the optimization framework by taking into account the inter-temporal dependencies between current rate decisions and profits generated in future periods. Thus, we introduce the Customer Value framework that allows the insurer to adapt its pricing strategy according to policyholders' behaviors over their customer lifetime while taking into account the market cycle. The empirical illustrations in the first two chapters are based on natural data observed by the insurer.In the second part of the thesis, we illustrate the contribution of field and laboratory experiments to the understanding of the demand for automobile insurance. A field experiment allows us to refine the measurement of customer price elasticity and to treat the pricing problem as a contextual bandit problem. Offline evaluation of several reinforcement learning strategies shows that those applying targeted fare experimentation achieve better financial performance compared to the myopic strategy, which excludes any possibility of experimentation. Finally, we present the results of a laboratory experiment whose objective was to measure the added value of private variables from decision models in risk. In particular, we analyze the role of risk aversion and risk perception in explaining car insurance choices. The same experimentation allowed us to analyze the external validity in experimental insurance, i.e. the similarity of individuals' behaviors in an experimental context and in the real economic context of the market.In addition to the experimentation-optimization duality in the field of insurance pricing, this thesis thus illustrates the duality between private data and public data, as well as the duality between empirical models of insurance demand and theoretical models.