April 21, 2022 — A morning “meeting with industry” of the Bézout Labex is organized on the site of Marne-la-Vallée, on the topic of numerical methods and artificial intelligence for insurance, with talks from members of SCOR and Milliman France.
Scientific coordinator: Benjamin Jourdain, CERMICS
The meeting will take place in Université Gustave Eiffel Bâtiment Copernic – 5 Boulevard Descartes 77420 Champs sur Marne – 2nd floor Room 2V070 & 2V080. Registration for the buffet lunch is mandatory, by April 12: register here.
Program
8:45 welcoming coffee
9:05 Presentation of Labex Bezout – Éric Colin de Verdière, Director of Labex Bézout
9:15 Modelling risks in insurance – Alexandre Boumezoued, Milliman
10:00 Les enjeux de recherches d’Intelligence Artificielle appliqués à l’assurance et aux sciences actuarielles – Antoine Ly, SCOR
10:45 Coffee break
11:15 Fair regression via Wasserstein barycenters – Mohamed Hébiri, LAMA
12:00 A synthetic model for Asset Liability Management in life insurance and numerical methods for Solvency Capital Requirement computation – Aurélien Alfonsi, Cermics
12:45 buffet lunch (registration is mandatory: register here by April 12)
Alexandre Boumezoued, Milliman France : Modélisation des risques en assurance
Abstract : In this talk, we will provide an overview of some developments for risk modelling that are driven by incentives specific to the insurance industry: – Business value arising from the precision of calculations and predictions, for better decision making in a competing market and varying environment, – Cost saving and process efficiency, by accelerating and automating the calculation chain, – Increasing awareness of regulators combined with growing regulatory guidance that prescribe specific calculations. The presentation will cover in particular: – Improving mortality data & forecasts using neural networks, – Modelling cyber attacks with Hawkes processes, – Accelerating the calibration of interest rate models with Edgeworth expansions and polynomial models.
Antoine Ly, SCOR : Les enjeux de recherches d’Intelligence Artificielle appliqués à l’assurance et aux sciences actuarielles
Abstract : Since the concept of “Big Data” reached the insurance industry a decade ago, the sector has been evolving and facing challenges to adapt to different evolutions. Amplified with the booming of academic research in “Artificial Intelligence” (concept we will define in regard to actual applications in the insurance industry) and releases of different major regulations like GDPR, innovation has changed gears. During this talk, we will try to emphasis those new challenges and how academics and companies could work together to face them.”
Mohamed Hebiri, LAMA : Fair regression via Wasserstein barycenters
Abstract : Powerful AI algorithms are used to make crucial decisions for various real world applications. In the insurance sector, such applications include the accurate pricing of contracts, the estimation of actuarial risk of individuals as well as various tasks automization, such as the estimation of reparation costs from pictures of a damaged car. These algorithms are trained in order to achieve high prediction accuracy but also have to take into account ethical and regulatory principles of our society. Algorithmic fairness is a recent machine learning field devoted to the study of methods that are able to mitigate bias in the data. In this talk, we illustrate some applications that highlight the importance of fairness and then consider a specific problem of fairness. In particular, we focus on the problem of learning a real-valued function that satisfies the Demographic Parity constraint. It demands the distribution of the predicted output to be independent of the sensitive attribute. We consider the case that the sensitive attribute is available for prediction. We establish a connection between fair regression and optimal transport theory, based on which we derive a close form expression for the optimal fair predictor. Specifically, we show that the distribution of this optimum is the Wasserstein barycenter of the distributions induced by the standard regression function on the sensitive groups. This result offers an intuitive interpretation of the optimal fair prediction and suggests a simple post-processing algorithm to achieve fairness. We establish risk and distribution-free fairness guarantees for this procedure. Numerical experiments indicate that our method is very effective in learning fair models, with a relative increase in error rate that is inferior to the relative gain in fairness.
Aurélien Alfonsi, CERMICS : A synthetic model for Asset Liability Management in life insurance and numerical methods for Solvency Capital Requirement computation
Abstract : We introduce a synthetic ALM model that catches the main specificity of life insurance contracts. First, it keeps track of both market and book values to apply the regulatory profit sharing rule. Second, it introduces a determination of the crediting rate to policyholders that is close to the practice and is a trade-off between the regulatory rate, a competitor rate and the available profits. Third, it considers an investment in bonds that enables to match a part of the cash outflow due to surrenders, while avoiding to store the trading history. We use this model to evaluate the Solvency Capital Requirement (SCR) with the standard formula, and illustrate the importance of matching cash-flows.
Then, we focus on the problem of evaluating the SCR at future dates. For this purpose, we study the multilevel Monte-Carlo estimator for the expectation of a maximum of conditional expectations. We obtain theoretical convergence results that complements the recent work of Giles and Goda. We then apply the MLMC estimator to the calculation of the SCR at future dates and compare it with estimators obtained with Least Squares Monte-Carlo or Neural Networks. Last, we discuss the effect of the portfolio allocation on the SCR at future dates.