WG Risk :“On the simulation of extreme events with neural networks”

The Working Group on Risk - CREAR, with the support of the ESSEC IDO dpt/Ceressec, Institut des actuaires, Labex MME-DII (CY), and the group Risques AEF - SFdS, has the pleasure to invite you to the seminar by :

Dr. Stéphane GIRARD, INRIA Université Grenoble Alpes Grenoble, France

Stéphane Girard is Senior Research Scientist at INRIA, Université Grenoble Alpes. He is also a part-time researcher at École Polytechnique, as a member of the Chair Stress-Test in Risk Management and Financial Steering. Between 2015 and 2021, he led the Simerge team of LIRIMA (Laboratoire International de Recherche en Informatique et Mathématiques Appliquées), a collaboration between INRIA Grenoble Rhône-Alpes and Université Gaston Berger, Sénégal. Prior to that, he was successively Associate Professor at Université de Montpellier and Université de Grenoble. His research interests encompass extreme value theory, statistical learning, dimension reduction, and nonparametric statistics. Stéphane Girard has supervised numerous doctoral theses and postdoctoral researchers, as well as published many papers in scientific journals at the intersection of these research directions.

“On the simulation of extreme events with neural networks”

This work aims to investigate the use of generative methods based on neural networks to simulate extreme events. Although they are very popular, these methods are mainly invoked in empirical works. Therefore, providing theoretical guidelines for using such models in an extreme-value context is of utmost importance. To this end, we propose an overview of some generative methods dedicated to extremes, giving theoretical tips on their tail behaviour thanks to extreme-value theory. More specifically, we focus on a new parametrization for the generator of a Generative Adversarial Network (GAN) adapted to the heavy tail framework. An analysis of the uniform error between an extreme quantile and its GAN approximation is provided: We establish that the rate of convergence of the error is mainly driven by the second-order parameter of the data distribution. The above results are illustrated on simulated data and real financial data. This is joint work with Emmanuel Gobet (CMAP, Ecole Polytechnique) and Michaël Allouche (Kaiko, Paris).

Dual format:

ESSEC Paris La Défense (CNIT), Room TBA, and via Zoom, please click here .