本研究主要探討台灣股價指數波動度的特性,分別由ARCH、GARCH、GJR─GARCH、EGARCH與QGARCH等五種不同波動度模型中配適出較適合台股指數波動度的模型,以及由常態分配、t分配和GED分配等三種誤差分配下找出較符合台股指數波動度的分配。再者,本研究引入了realize range-based volatility代理資產的真實波動(true volatility)。而在探討預測績效方面,本研究使用MAE、MSE、MME和VaRE(VaR-based Error)等多種不同的損失函數,以及將預測之波動度帶入B-S model與市場價格比較,並且利用更具強健性的SPA test來檢定多種模型預測績效的比較。另外,除了日資料,更進一步使用週資料,探討不同資料頻率下對於資料模型的配置是否一致。結論顯示在日及週兩種資料頻率下,不對稱之模型以及誤差分配設定對於預測具有不對稱特性的台股波動度有較佳的績效,說明不對稱與誤差分配的設定對於波動性預測之重要性。 This study selects the appropriate model to match volatility of Taiwan stock market from ARCH, GARCH, GJR-GARCH, EGARCH and QGARCH models and find the appropriate distribution assumption from normal, t and GED distribution. In the meantime, we use realize range-based volatility to be the proxy of true volatility. This study not only uses many kinds of loss functions, including MAE, MSE, MME, VaRE and Black-Scholes equation, but also employ more robust SPA test to compare forecasting performance of models. Besides daily data, this paper uses weekly data to know whether different frequency data are consistent. The empirical result indicates that there are high performance to forecaste volatility of Taiwan stock market which is asymmetric when asymmetric models and correct distribution assumption be used. Therefor, alternative asymmetric and distribution assumption are important for volatility forecasting.