一般化自我迴歸條件異質變異數模型可對條件分配做不同的假設，本研究比較在不同條件分配假設下，它們在模型配適、波動度預測、與價格分配預測上的表現。我們對模型假設了三種不同的條件分配：常態分配、偏斜 t 分配、與複合卜瓦松(跳躍)分配，以捕捉資產報酬的一般特性。實證分析建立在非線性不對稱一般化自我迴歸條件異質變異數模型的基礎上，並以S&P 500與FTSE 100指數為實證資料。實證結果顯示，在模型配適上的表現，跳躍模型與偏斜 t 模型較常態模型為優；但這樣的優勢不見於低波動度期間。在波動度預測上，跳躍模型表現最佳。而在價格分配預測上，雖然三者差異不多，但跳躍模型與偏斜 t 模型的預測仍比常態模型精確。 This study compares the performance of alternative GARCH models with different conditional distributions on model fitting, volatility forecasting, and density prediction. Three conditional distributions: normal, skewed-t, and compound Poisson, are assumed in order to model the stylized facts of returns in the stochastic innovation. Based on the NGARCH framework, parameters are estimated from the S&P 500 index and FTSE 100 index. The empirical results suggest that the NGARCH-jump model and the NGARCH-skewed-t model significantly raise performance in terms of model fitting, but the differences diminish when models are estimated in relatively low-volatility periods. In volatility forecasting, the NGARCH-jump model outperforms the others. Although the differences are not significant, the skewed-t model and the jump model provide more accurate estimated densities than the normal model.
