This talk has been automatically captioned. You can remove these by pressing CC on the video toolbar.

Name: Luca Trapin

Talk Title: Modelling panels of extremes

Abstract: Estimation of the parameters of the generalized extreme value distribution is notoriously challenging due to the small number of observations that are usually available in applications. Regression techniques are commonly used to pool information and to capture cross-sectional heterogeneity or non-stationarity in the data. We propose the notion of an extreme value panel consisting of several groups of individuals or gauging stations. We propose a new algorithm that jointly assigns the individuals to the latent groups and estimates the model parameters of the regression models inside each group. An extensive simulation study shows that this method efficiently recovers the underlying group structure without prior information. We apply our approach to risk assessment for financial, hydrological and climate data. All of these applications can be formulated as a panel problem and we show that our method provides improved estimates of extreme quantiles and helps to answer important domain-specific questions.

This talk is a contributed talk at EVA 2021.