蔡麗茹，周忠賢
摘要/Abstract本文針對三種匯率進行實證，共得到下述結論：1.混合常態且條件變數為異質的模型, 因為同時融合了GARCH(或指數移動平均)與混合常態，使得此方法在整體上表現較佳，對未來風險的掌握能力較好。2.等權移動法的方法雖然簡單，在央行常介入外匯市場穩定匯率時，仍不失為一值得參考的方法。但若央行對外匯市場的干預較少時，則容易使所得到的VaR產生偏誤。3.在預期台幣兌美元匯率十日資料方面,所有的估計方法都會有高估風險的情形。 關鍵字：風險值, GARCH模型,混合常態分配 Since financial asset returns often don't follow the hypothesis of normal distribution (such as high kurtosis and fat tails), this study measures VAR by constant variance under the normal distribution as well as by GARCH model. Moreover, we try to forecast VaR under the mixture of normal distribution because abnormal events often lead to large volatility of asset price. Our empirical studies also compare the previous methods' performances with the performances of historical simulation. Finally, we try to find an objective risk-disclosing way by comparing the performances of the previous methods. The empirical results are as follows: 1. In general, GMB (mixture of normal distributions with GARCH variance) has a good ability to forecast risk because it combines the properties of GARCH, and the properties of mixture of normal distributions. GMIX (mixture of normal distributions with EWM variance) also has a good result. 2. EQW (equal weighted moving average) is simple to calculate, however it could be a good risk measurement if central banks conduct foreign exchange interventions to influence exchange rates. 3. The forecasts of risk are often too high when there are thin tails at out-of-samples period, such as NT dollar exchange rate. Key words : Value at Risk (VaR) , GARCH model , mixture of normal Distributions |