Discrete Multivariate Generalized Pareto Distribution for Drought Risk Assessment

par Prof. Marie Kratz (ESSEC Cergy)

mardi 4 nov. 2025, 14:00 → 15:30 Europe/Paris

UTC GI

Droughts are among the most severe and growing risks associated with climate change, with major consequences for society. Managing these risks requires contributions from multiple disciplines, including climate sciences, probability and statistics. Our work contributes to this agenda by extending extreme value theory (EVT) to discrete multivariate settings, through the introduction of Multivariate Discrete Generalized Pareto Distributions (MDGPDs). These models bridge the gap between continuous EVT and discrete count data, offering a flexible approach to threshold exceedances for events such as dry spells. Rooted in Generalized Pareto theory, MDGPDs provide a principled framework for representing rare and compound events in a variety of applied contexts. We present the theoretical construction of MDGPDs, simulation methods, and likelihood-free inference techniques tailored to this discrete multivariate framework. A case study on European drought events illustrates the practical relevance of the model for climate-related risk assessment. The tools developed support decision-makers—such as insurers, policymakers, and climate risk analysts—in better understanding, anticipating, and pricing the impacts of extreme dry periods. This is a joint work with S. Aka (ESSEC CREAR & LSCE Saclay) and P. Naveau (LSCE Saclay).