B0420
Title: Statistical modelling with multivariate generalized Pareto distributions
Authors: Anna Kiriliouk - Universite Catholique de Louvain (Belgium)
Holger Rootzen - Chalmers (Sweden)
Johan Segers - Universite catholique de Louvain (Belgium)
Jenny Wadsworth - Lancaster University (United Kingdom) [presenting]
Abstract: The multivariate generalized Pareto distribution (MGPD) arises as the limiting distribution of appropriately normalized random vectors, given that at least a single component is large. A conceptual advantage of the MGPD over other threshold-based multivariate extreme value methods is that it corresponds to a proper multivariate distribution, and does not necessitate marginal transformation. There is a growing probabilistic literature on the topic, but seemingly little thus far on statistical inference. This talk will cover aspects of MGPDs that are particularly relevant for statistical modelling: construction of models, likelihood-based inference, threshold selection and diagnostics.
