Robust quantile estimation under bivariate extreme value models (Q2303024)

In some applications such as financial risk and flood frequency analyses, it is important to find the smallest and largest \((1-p)\)-th quantile of a bivariate random vector, provided their true dependence stays within a certain distance around a reference dependence. In this paper, an efficient and robust method is developed when the bivariate random vector follows a generalized extreme value distribution. Bisection algorithms are presented to find the robust bivariate extreme quantiles. Numerical experiments are conducted to evaluate the performance of the proposed method and algorithms, which are also applied to analyze a real dataset.