Contributions to life and health insurance provisioning and risk management.

Summary In the insurance sector, the latest regulatory developments and accounting standards are moving towards standardization of risk management within organizations. In this context, the main objective of my thesis is to propose different methodologies for risk assessment and analysis in this sector. The first part of this manuscript deals with the problem of individual provisioning in non-life. I propose adaptations of ensemblistic machine learning algorithms and of some performance metrics for estimating claim durations and ultimate claim expenses in the presence of right-censored data. The application of these methods to real data from loan contracts or group insurance contracts leads to better and more robust estimates of the considered parameters. The second part presents an approach for estimating a one-year shock on the Undertaking Specific Parameters of the life insurance module of pillar 1 of the Solvency II standard formula. The use of the American credibility (or limited variation credibility) allows the partial taking into account of the constraints of availability of experience data (volumetry and depth of history) when calibrating the shocks. As an illustration, I have applied this approach to the incidence and continuation (or reinstatement) risks of disability and out-of-work benefits in a portfolio of loan contracts. The results obtained show significant decreases in the solvency capital requirements (SCR) of the underwriting risk compared to the standard formula. The third part is a descriptive study of the calculations of the standard formula for the evaluation of the economic capital requirement of the dependency risk. It highlights the shortcomings of the standard and suggests ways to improve it in order to better take into account the specificities of this risk. Finally, in the last part of the manuscript, I propose a comparative study of the preferences of attitudes towards risk in the financial sector, in particular banking and insurance. This is an empirical analysis conducted in three geographical areas (America, Europe and Africa) in order to measure the links and differences between risk attitude profiles and certain socio-demographic variables.