- 學位論文
匯率條件風險值之估計與比較
The Estimation and Comparisons of Conditional Value at Risk on Major Exchange Rates
摘要
風險值是一個因應衍生性商品蓬勃發展、金融市場波動的新興風險管理工具，它定義為持有某資產一段期間，在一定的信賴水準下，所可能遭受的最大損失。風險值擁有量化風險的優點，金融機構為了內部控管及調整，已經將風險值視為衡量市場風險的標準。近年來，更因財務市場上重大事件頻傳，漸漸喚起金融機構對有效風險管理的重視。 台灣對外貿易依存度高，而匯率在國際貿易活動中，又一值扮演著重要的角色，在未來市場面臨國際化、自由化，以及我國加入WTO的趨勢下，面對匯率風險乃是在所難免，所以本研究是以美元兌換國際間主要貨幣之即期匯率，包括有加拿大幣、英鎊、德國馬克、法國法郎、日圓和台幣為實證對象，評估不同模型估計出的風險值與預測績效。模型分為有母數模型與無母數模型兩大類。有母數模型假設資產報酬率為隨機、獨立之常態分配且序列不相關，分別包含了均等權數移動平均法、RiskMetric模型及含有異質變異的GARCH模型；無母數模型對於資產報酬並不假設任何分配，主要是透過歷史資料模擬而得，分別包含了歷史模擬法及蒙地卡羅模擬法。此外，本研究為了探討風險值間具有自我相關（autocorrelated）的特性，也更進一步的加入了Engle and Manganelli （2000） 提出的條件自我相關VaR模型 － CAViaR， CAViaR是將傳統衡量出風險值具有跳躍的情況予以平滑化（smoothly），亦即表示風險值間具有前後期的自我相關性。 本研究得到的結論為，以成功率來比較各模型，在短期平均而言，以蒙地卡羅模擬法最佳，中長期而言以RiskMetric較適；以均方差標準根來比較，RiskMetric所計算出的風險值相較於實際發生損失值間的差距較小，對於中長期風險值的估計，必須考量異質變異的情況下，CAViaR相較於其他傳統模型所估計出來的風險值之RMSE值較小，為值得考慮的較佳模型。本研究建議，投資人或金融機構要進行各外幣的短期投資以RiskMetric模型及5%的顯著水準來衡量風險值最佳，若進行中長期投資，在不考慮異質變異的情況下，建議以歷史模擬法為估計考量，發生異質變異之外幣，則以CAViaR模型較佳。就中央銀行或金融機構將VaR值作為風險控管的角度，在確定的資產持有期間下，本研究建議以歷史資料視窗長度一年（250天）及顯著水準5%的條件下，以Engle and Manganelli（2000）提出的CAViaR模型評估各外幣的風險值最為的恰當，因為CAViaR模型是預估出來的風險值和實際損失值較為接近的模型。
並列摘要
Value at Risk （VaR）is an emerging tool of risk management which is designed to meet the trend of various derivatives and more volatile financial markets. VaR is defined as the value that portfolio will lose with a given probability, over a certain time horizon. It is a better way of quantifying risk. It has become the standard measure of market risk employed by financial institutions for both internal and regulatory purposes. Recently, financial disasters have emphasized the importance of effective risk management for financial institutions. Foreign exchange plays an important role in the international trade for Taiwan. The purpose of this study uses the behavior of daily exchange rates. The data takes US Dollar to major currencies, such as Canadian Dollar, German deutschemark, French franc, British Pound, Japanese Yen and New Taiwan Dollar as our empirical data. It also compares the VaR and predicting effectiveness for different models. There are two types of models. One is parametric models. They are constructed under the assumptions that return is iid and normally distributed, but most return of financial assets is nonormally distributed. Parametric models include equally weighted moving average approaches, exponentially weighted moving average approaches and generalized autoregressive conditional heteroskedastic. The other is nonparametric models which require no assumptions on distribution of returns. It comes mainly from the simulation on historical data. Nonparametric models include historical simulation method and Monte Carlo simulation method. Besides, this study attempts to address a conditional autoregressive specification on VaR, which is called Conditional Autoregressive Value at Risk - CAViaR (Engle and Manganelli, 2000). CAViaR model which is different from traditional VaR model is the value of quantitative risk changes “smoothly” over time. The conclusions in my study are that Monte Carlo simulation method is the best model in short-holding periods as the performance test based on proportion of successes. In long-holding periods, RiskMetric model is relatively suitable. As the performance test based on root mean squared errors, VaR measured in RiskMetric is closed to the losses of real value. For long-holding periods, the heteroscedasticity must be considered to estimate the VaR. CAViaR is considered a superior model. When investors or financial institution are going to make a short-holding investment on varies foreign currencies, the VaR measured in 5% significant level of RiskMetric model is suggested by this study. In long-holding period, CAViaR model is suggested to use when heteroscedasticity is considered. When the Central Bank or financial institution use VaR to control and management in the risk, the measurement of VaR in varies foreign currencies is more appropriate to evaluated by the CAViaR model under the length of historical data window is 250 days and the significant level is 5%.