本研究衡量各種條件異質波動率模型(包括EGARCH、FIEGARCH、EGARCH-jump與EGARCH-skewed-t模型)在模型配置、波動率預測及VaR風險值預測上的表現。以單純的EGARCH模型為比較基礎,整體而言EGARCH-jump模型呈現最顯著的改善,幾乎在各方面的表現皆優於其他模型;但是,計算上繁複程度也隨之大量增加。如果計算簡便是一個重要的考量因素,則應用EGARCH-skewed-t模型於模型配置與風險值預測,及應用FIEGARCH模型於波動率預測皆是可行的次佳選擇,因為這兩個模型在這些方面的應用上亦有相當另人滿意的表現。此外,由於FIEGARCH模型只有在美國股價報酬率資料上表現得特別好,這顯示這種以不完整差分波動率模型所捕捉到的長記憶型態可能不是全球市場一致的現象。
This study evaluates the performance of alternative volatility models, including EGARCH, FIEGARCH, EGARCH-jump, and EGARCH-skewed-t models, on model fitting, volatility forecasting, and Value-at-Risk (VaR) prediction. As compared with the simple EGARCH model, the EGARCH-jump model demonstrates significant improvements, outperforming every other model in almost all aspects, but the computation load substantially increases by the inclusion of jump dynamics. If less expensive volatility models are preferred, then alternatives may include the use of the EGARCH-skewed-t model for model fitting and VaR prediction, and the FIEGARCH model for volatility forecasting, since these models also demonstrate fairly good performance for these specific purposes. Furthermore, since the FIEGARCH model demonstrates relatively good performance with regard to U.S. stock market returns only, this suggests that the long-memory pattern captured by the fractionally-integrated volatility model may not be a global stylized fact.