Scoring predictions at extreme quantiles

Prediction of quantiles at extreme tails is interest in numerous applications. Extreme value modelling provides various competing predictors for this point prediction problem. A common method assessment a set to evaluate their predictive performance given situation. However, due the nature inference problem, it can be possible that predicted are not seen historical records, particularly when sample size small. This situation poses problem validation with its realization. In article, we propose two non-parametric scoring approaches assess quantile mechanisms. The proposed methods based on predicting sequence equally different parts data. We then use function predictors. compared conventional and superiority former demonstrated simulation study. applied analyze cyber Netflow data from Los Alamos National Laboratory daily precipitation station California available Global Historical Climatology Network.