本研究以主要工業國（美元、歐元、日圓、加拿大幣與英鎊）對新台幣每日匯率收盤價為研究對象，利用GARCH與APARCH模型在常態分配、t分配與偏態t分配的假設下，進行90%、95%、99%及99.5%信賴水準的多空頭部位風險值之估計。最後將所有風險值模型進行其績效衡量，並利用Kupiec（1995）的概似比檢定(Likelihood Ratio Test)對風險值模型進行比較，找出計算風險值較佳模型。研究結果發現： 1. 就多頭部位而言，GARCH-n模型在低顯著水準(10%及5%顯著水準)下的風險值績效有不錯的表現，但GARCH-n模型在高顯著水準(1%及0.5%顯著水準)下的風險值績效卻未能通過回顧測試，表示常態分配無法確切地捕捉厚尾的性質。t分配與偏態t分配皆能改善在尾部分佈下的風險值績效，並能緩和常態分配無法確切地捕捉厚尾的問題。然而，偏態t分配相對於t分配的尾端分佈風險值績效的改善程度卻有限。
2. 就空頭部位而言，普遍在常態分配下的GARCH與APARCH模型的風險值績效有較佳的表現。普遍在t分配下的GARCH與APARCH模型的風險值績效表現較不佳，因估算出的實際穿透個數往往低於理論穿透個數甚多，導致無法通過回顧測試的檢定，表示t分配假設潛在引發風險值估計過於保守的缺失。普遍在偏態t分配下的GARCH與APARCH模型可改善t分配假設潛在引發風險值估計過於保守的缺失。 The study is to compute VaR in the long and short trading position by using GARCH and APARCH models with normal distribution, student t distribution and skewed student t distribution at four significant levels. The sample daily data covers New Taiwan Dollar to US Dollar, European Euros, Japanese Yen, Canadian Dollar and British Pound exchange rate. Moreover, the study is to evaluate the validity of different models by using backtesting method based on likelihood ratio test proposed by Kupiec (1995). The empirical results are as follows:
1. For the long trading positions, the GARCH model with normal distribution has poor validity in high significant levels due to the fat-tail problem. The fat-tail problem could be improved by the student t distribution and skewed student t distribution. However, compared with student t distribution the improved VaR performance of the skewed t distribution is not significant. 2. For the short trading positions, the GARCH and APARCH models with normal distribution has good validity. The GARCH and APARCH models with student t distribution induced the over-conservative problem of VaR estimation, but the GARCH and APARCH models with skewed t distribution could improved it.
