Exploring the role of asymmetry in interval forecasting from a tail risk perspective

Compared to evaluating the overall interval forecasting performance, this study proposes to assess interval forecasting performance from a tail risk perspective. The interval forecasting performance of tail data below the 0.05 quantile and above the 0.95 quantile is evaluated separately. GARCH models are employed for interval forecasting of three major US stock price indices: GSPC, IXIC, and DJI. The interval forecasting performance is evaluated by the average Winkler score (AWS). The results of this study show that in the left tail of left-skewed data, asymmetric models have superior interval forecasting performance. Among these asymmetric models, the GJR-GARCH-ST model performs the best. Overall, this study emphasises the importance of asymmetric models in interval forecasting, especially for data exhibiting skewness.