Similarity‐based clustering for patterns of extreme values

Statistical modelling of the magnitude and frequency extreme observations is fundamental for a variety sciences. In this paper, we develop statistical methods similarity-based clustering heteroscedastic extremes, which allow us to group time series independent according their extreme-value index scedasis function (i.e., values, respectively). Clustering functions indices involves challenge grouping objects composed both (scedasis) scalar (extreme-value index), thus need partition product-space. Our analysis reveals an interesting mismatch between losses on London Stock Exchange corresponding economic sectors affected stocks. The further suggests that dynamics governing comovement in exchange contains information business cycle.