Title: Modelling panels of extremes
Authors: Luca Trapin - University of Bologna (Italy) [presenting]
Debbie Dupuis - HEC Montreal (Canada)
Sebastian Engelke - Ecole Polytechnique Federale de Lausanne (Switzerland)
Abstract: Extreme value regression has been widely used over the last years to study the determinants of tail-risk events. Since such extreme events rarely occur, estimates of the model parameters are usually derived using a small number of observations, thus inducing high uncertainty. A class of panel regression models for the extremes is presented where the cross section of the data is pooled to obtain more reliable estimates of the regression coefficients. To account for possible unobserved heterogeneity in the data, we allow the extreme value panel regression model to have group-specific parameters, and design a likelihood-based clustering algorithm to recover the unknown group structure. A large simulation study assesses the finite sample properties of the panel maximum likelihood estimator and assesses the ability of the clustering algorithm to recover the unknown group structure. Finally, the usefulness of the new class of models is illustrated in several real-world examples.