本研究之目的為建立和比較五種單一變數預測方法，並尋找出適用於匯率升貶值最佳的預測方法，以便預測人民幣之每月匯率升貶值數據。研究之資料範圍為2007 年1 月至2011 年12 月。透過古典分解法、三角函數迴歸、季節性虛擬變數、灰預測、及混合型灰預測等五種預測模式，以產生人民幣之五種方法的預測值。藉由實證分析，利用平均絕對誤差(Mean Absolute Error, MAE)、平均絕對誤差百分比(Mean Absolute Percent Error, MAPE)及殘差均方根(Root Mean Squared Error, RMSE)等評估指標比較後，以驗證何者可提供預測最佳之精確度。實證結果發現，人民幣匯率升貶值以古典分解法平均絕對誤差百分比在1%以內最為準確，混合型灰預測平均絕對誤差百分比為1.1849%則排第二。就整體預測方法比較而言，古典分解法擁有最佳的人民幣匯率升貶值之短期預測準確度，顯示此種方法在短期預測中，對人民幣匯率升貶值具有一定程度的預測能力，其次為混合型灰預測。因臺灣為海島型國家高度仰賴進出口貿易經濟發展，易遭受匯率升貶波動、通貨膨脹的影響，故擇與我國經濟合作程度緊密且最重要的出口國中國人民幣匯率預測為主旨。

The purpose of this study is to establish and compare five univariate prediction methods to find out the most suitable way of forecasting the monthly RMB exchange rate. The research data were collected from Jan 2007 to Dec, 2011. Through Classical Decomposition, the Trigonometric Model, the Regression Model with Seasonal Dummy Variables, the Grey Forecast, and Hybrid Grey Forecast Model, it has the results of five forecasting models for the exchange rate of RMB. Mean Absolute Error (MAE), Root Mean Square Residual (RMSE) and Mean Absolute Percent Error (MAPE) are used to verify whichever provides the best prediction accuracy. Empirical results show that RMB exchange rate predicted by Classical Decomposition method was the most accurate as the MAPE is less than 1%. While the Hybrid Grey Forecast Method was the runner up as the MAPE is 1.1849%. In comparison with five prediction methods, the Classical Decomposition Method has the best short-term accuracy over RMB exchange rate, followed by Hybrid Grey Forecast method. Because Taiwan is an island country, it is highly dependent on import and export trade. Furthermore, Taiwan’s economic development is vulnerable to fluctuations in exchange rates and, inflation. Hence, the currency of our most important export country, RMB, was chosen as the studied subject.