CMStatistics 2018: Start Registration
View Submission - CMStatistics
B1594
Title: Weighted energy score Authors:  Xiaochun Meng - University of Sussex (United Kingdom) [presenting]
James Taylor - University of Oxford (United Kingdom)
Abstract: Multivariate probabilistic forecasting is particularly intriguing and challenging due to its inherently complex nature and computational difficulty. For some applications, such as financial market risk assessment, a specific region is often of more importance than the whole distribution. To emphasise the region of interest for multivariate distributions, we propose the weighted energy score by generalising the existing energy score via threshold and quantile weight functions. The proposed weighted energy score is proper and provides useful insight into the evaluation of multivariate probabilistic forecasts. We use financial data to provide empirical support for the proposed weighted energy score.