本研究係使用GARCH模型再搭配常態、t、GED及偏態t等四種報酬率分配之參數法風險值模型對美國四個股價指數(DowJones, NASDAQ, S&P500及NYSE)在次級房貸期間之進行95%、99%及99.5%三種信賴水準一天期風險值之估計與預測。再使用三種準確性檢定來評估此風險值模型之準確性。由實證結果我們發現在多頭部位,不論以準確性檢定之通過總數或以AQLF值之大小來衡量,GARCH-ST模型性能最佳,其次為GARCH-T、GARCH-GED模型,而GARCH-N模型性能最差,由此可判斷出在多頭部位偏態效應之影響力遠大於厚尾效應。在空頭部位,雖然準確性檢定之通過總數四種模型相差不多,但以AQLF值之大小來衡量,GARCH-GED模型性能最佳、其次為GARCH-T模型,由此可判斷出在空頭部位厚尾效應之影響力遠大於偏態效應。
This investigation applies the parametric method, which use GARCH model with normal, student t, GED, and skewed t return innovations, to examine the one-day-ahead VaR forecasting performance of four stock indices, the DowJones, NASDAQ, S&P500 and NYSE in United States during the period in the secondary mortgage under 95%, 99% and 99.5% confidence levels. Then we use three kinds of accuracy tests to evaluate the accuracy of these VaR models. Empirical results show that, irrespectively measure with the total number of passing accuracy tests or the value of AQLF, the GARCH-ST model provides the most accurate VaR forecasts followed by GARCH-T and GARCH-GED models for long position. This indicates that with regard to long position, skewness effect is more crucial than fat-tails effect. On the contrary, with regard to short position, although total numbers of passing accuracy tests of the four models are almost the same but when using the value of AQLF the GARCH-GED model provides the most accurate VaR forecasts followed by GARCH-T model, indicating that fat-tails effect is more important than skewness effect for short position.