Balancing predictive performance and coherence in insurance pricing: From credibility to fair and autocalibrated models.

This evolution naturally leads to the following question: how can insurance pricing models be both more accurate—aligned with individual risk—and more ethically and socially responsible?This thesis addresses this question in two complementary parts. Part I revisits classical a posteriori pricing techniques, such as credibility models and Bonus-Malus Systems in the context of telematics. Part II explores modern predictive pricing, emphasizing fairness, autocalibration, and predictive performance. Together, these two parts aim to contribute both theoretical insights and practical tools for improving technical insurance pricing.