A semiparametric approach for bivariate extreme exceedances

TL;DR

This paper introduces a semiparametric method for analyzing bivariate extreme exceedances that relies solely on theoretical constraints, combining marginal exceedance modeling with flexible copulae to better understand extreme dependence patterns.

Contribution

The paper develops a novel semiparametric approach that integrates extreme value theory with flexible copula models, avoiding arbitrary constraints and improving inference on extremal dependence.

Findings

Method effectively captures various dependence patterns.

Robustness of the new probabilistic criterion is empirically validated.

Approach outperforms existing methods in simulations and real data applications.

Abstract

Inference over tails is performed by applying only the results of extreme value theory. Whilst such theory is well defined and flexible enough in the univariate case, multivariate inferential methods often require the imposition of arbitrary constraints not fully justifed by the underlying theory. In contrast, our approach uses only the constraints imposed by theory. We build on previous, theoretically justified work for marginal exceedances over a high, unknown threshold, by combining it with flexible, semiparametric copulae specifications to investigate extreme dependence. Whilst giving probabilistic judgements about the extreme regime of all marginal variables, our approach formally uses the full dataset and allows for a variety of patterns of dependence, be them extremal or not. A new probabilistic criterion quantifying the possibility that the data exhibits asymptotic independence is introduced and its robustness empirically studied. Estimation of functions of interest in extreme value analyses is performed via MCMC algorithms. Attention is also devoted to the prediction of new extreme observations. Our approach is evaluated through a series of simulations, applied to real data sets and assessed against competing approaches. Evidence demonstrates that the bulk of the data does not bias and improves the inferential process for the extremal dependence.