CMStatistics 2018: Start Registration
View Submission - CFE
A0456
Title: Tail optimal combinations Authors:  Christos Argyropoulos - Lancaster University (United Kingdom) [presenting]
Ekaterini Panopoulou - University of Essex (United Kingdom)
Abstract: The utility of combining densities in improving the forecasting accuracy of risk measures is investigated. Specifically, we propose the Tail Quantile Score (TQS) rule which focuses directly on the tails of the distribution by taking into account the severity of the losses alongside the probability of events, across the tail of the distribution. Our simulation exercise suggests that TQS isolates efficiently the tail related performance of the methods when compared to competing tail related scoring rules. Furthermore, we develop time-varying weighting schemes in order to evaluate the benefits of combining the densities on the tail regions and the respective benefits on the accuracy of the Value at Risk and Expected Shortfall risk forecasts. Our results suggest that the optimal weights outperform the naive combination schemes with significant improvements in the forecasting accuracy of the respective risk measures.