Loading Events

François HU (Milliman) “Obtaining Fair Insurance Premiums with Multiple Sensitive Attributes”

May 13 @ 10:00 am - May 20 @ 11:00 pm

Finance-Insurance
Time: 10.00 am
Date:13th of May  2025
Room 3001

François HU (Milliman) “Obtaining Fair Insurance Premiums with Multiple Sensitive Attributes”

Abstract : In the context of Algorithmic Fairness, the goal is to ensure that sensitive attributes have no influence on decision-making outcomes. This objective has driven the development of various fairness definitions and tools, which have been successfully applied across multiple domains, including insurance. However, the challenge becomes more complex when dealing with multiple sensitive attributes. In the absence of intentional discrimination, predictive models used in insurance – for tasks such as pricing, fraud detection, or claims estimation – can inadvertently result in biased decisions, such as ageism, racism, or sexism. As emphasized by Kearns et al. (2019), machine learning models do not inherently provide fairness unless explicitly designed to do so. To address this challenge, we propose a novel sequential framework that progressively achieves fairness across multiple sensitive attributes in insurance applications. By leveraging multi-marginal Wasserstein barycenters, we extend the concept of Strong Demographic Parity to ensure independence between multiple sensitive features and decision outcomes. Our approach provides a closed-form solution for an optimal, sequentially fair predictor, while maintaining interpretability in terms of how the sensitive features interact. Additionally, our framework accommodates the trade-offs between fairness and risk, enabling targeted prioritization of fairness improvements for specific attributes based on context-specific requirements. We demonstrate the practical applicability of our framework through a data-driven estimation procedure applied to both synthetic and real-world datasets, including an insurance dataset. Our numerical experiments show that this post-processing method not only enhances fairness in decision-making but also ensures transparency and maintains statistical guarantees, offering a robust solution for achieving fairer insurance premiums.

Organizers:  Jean-David FERMANIAN