本研究運用修正後的歷史模擬法（Filtered Historical Simulation，FHS）、McNeil（2000）提出的GARCH-EVT模型和以Copula為基礎的FHS模型（Copula based FHS Model）三種方法，以評估涵蓋歐洲、美國和台灣三地股票市場的國際投資組合之風險值，並利用Kupiec（1995）提出的概似比檢定（Likelihood Ratio Test, LR test）和均方誤差法（Root Mean Squared Error, RMSE）評估風險值模型的準確性。 由實證結果可知次貸危機發生後，各國股價指數之間的關聯性結構具有顯著的變動，使得國際投資組合不再具有風險分散效果。由概似比檢定可知，無論是在金融危機前或是金融危機後，FHS模型皆可正確地預測風險值，但是，由RMSE可知，FHS模型較不具有資金運用效率；至於其他的風險值實證模型在金融危機期間皆無法正確地預測風險值。相較於傳統的線性結構，非線性關聯結構（Copula 函數）可以相對準確地預測風險值，此外，在金融危機前，若考慮越長的估計期間，則對於Copula函數配適股價指數之間的關聯性有顯著地正向影響，但是，在金融危機期間內，估計期間對於風險值實證模型的預測能力並不具有顯著地影響。 The study applies Filitered Historical Simulation, GARCH-EVT model which McNeil(2000) proposed, and Copula Based FHS Model to evaluate Value at Risk(hence VaR) for international portfolio which contained European, American, and Taiwanese stock market. On the other hand, the study applies Likelihood Ratio Test which Kupiec(1995) proposed and Root Mean Squared Error to evaluate the accuracy of VaR model. The empirical results demonstrate the relationship between international stock index has significant varying such that international portfolio didn’t have diversified effect of risk. By likelihood ratio test, in both before and after financial crisis, FHS model can correctly forecast VaR but other VaR model can’t correctly forecast VaR during financial crisis. However, FHS model didn’t have efficient of funding operation by RMSE. Compared with traditional linear structure, nonlinear structure are relatively correct on VaR forecasting. Otherwise, consider longer estimated period has significantly positive effect to Copula function on fitting the relationship between stock index before financial crisis but estimated period hasn’t significantly effect on VaR model’s forecasted ability after financial crisis.
