An Extreme Value Bayesian Lasso for the Conditional Left and Right Tails

Abstract We introduce a novel regression model for the conditional left and right tail of possibly heavy-tailed response. The proposed can be used to learn effect covariates on an extreme value setting via Lasso-type specification based Lagrangian restriction. Our track if some are significant lower values, but not (right) tail—and vice versa; in addition this, bypasses need threshold selection theory framework. assess finite-sample performance methods through simulation study that reveals our method recovers true distribution over variety scenarios, along with being accurate variable selection. Rainfall data showcase how distinguish between key drivers moderate rainfall, against those rainfall. Supplementary materials accompanying this paper appear online.