Multivariate Sparse Clustering for Extremes

Identifying directions where extreme events occur is a significant challenge in multivariate value analysis. In this article, we use the concept of sparse regular variation introduced by Meyer and Wintenberger to infer tail dependence random vector X. This approach relies on Euclidean projection onto simplex which better exhibits sparsity structure X than standard methods. Our procedure based rigorous methodology aims at capturing clusters extremal coordinates It also includes identification threshold above values taken are considered extreme. We provide an efficient scalable algorithm called MUSCLE apply it numerical examples highlight relevance our findings. Finally, illustrate with financial return data. Supplementary materials for article available online.