B0394
Title: Generalized Pareto regression trees for extreme event analysis
Authors: Antoine Heranval - CREST - ENSAE Paris (France) [presenting]
Maud Thomas - Sorbonne University (France)
Olivier Lopez - Ensae IP Paris (France)
Abstract: Finite sample results are derived to assess the consistency of Generalized Pareto regression trees as tools to perform extreme value regression for heavy-tailed distributions. This procedure allows the constitution of classes of observations with similar tail behaviors depending on the value of the covariates, based on a recursive partition of the sample and simple model selection rules. The results provided are obtained from concentration inequalities and are valid for a finite sample size. A misspecification bias that arises from the use of a "peaks over threshold" approach is also taken into account. Moreover, the derived properties legitimate the pruning strategies, that is the model selection rules, used to select a proper tree that achieves a compromise between simplicity and goodness-of-fit. The methodology is illustrated through a real data application in insurance for natural disasters. A methodology is also discussed that aims at pricing extreme events, based on a combination of individual information and of collective data. The output of the Generalized Pareto regression trees is used as the prior distribution in Bayesian credibility theory.