Evaluation of point forecasts for extreme events using consistent scoring functions

We present a method for comparing point forecasts in region of interest, such as the tails or centre variable's range. This cannot be hedged, contrast to conditionally selecting events evaluate and then using scoring function that would have been consistent (or proper) prior event selection. Our also gives decompositions functions are mean particular quantile expectile. Each member each decomposition is itself emphasises performance over selected The score has natural interpretation rooted optimal decision theory. It weighted average economic regret user thresholds, where weight those thresholds corresponding interest.