本研究提出使用半參數迴歸方法,建構平滑係數部分線性模型藉由廣義經濟的變數預測台灣匯率,使其係數隨著變數GDP成長率的不同而改變。本研究使用從2003年9月到2014年11月的資料,並將資料分為訓練樣本與測試樣本兩個部分以估計參數與建構模型。本研究透過固定視窗、移動視窗以及累積移動視窗方法且藉由過去的經濟變數預測台灣匯率。以線性迴歸模型為比較基準,檢測是否本研究所提出的方法更適合應用於台灣匯率預測,在不同的條件下,透過預測精度選取出最適本研究之模型;半參數迴歸模型可以減少導致錯誤決策之風險。 This paper uses the smoothing coefficients partially linear model to forecast Taiwan''s exchange rate by marco-economic fundamentals with the level of GDP growth rate. Based on the exchange rate data ranging from September of 2003 to November of 2014, we separate exchange rate data into training and test data, which the training data is used to build our model and estimate the coefficients, and the test data is used to check the performance of our model. Here, we build the model by controlling lagged periods of exchange rate data to forecast the exchange rate of next period with the fixed window, the moving window and the cumulative moving window method. In the procedure of build our model, we need to classify the coefficients of variables is varying or fix first, then use the generalized likelihood ratio test to check the significant of the relationship between Taiwan’s exchange rate and marco-economic fundamentals with the level of GDP growing rate. Our model can omit parametric model could lead to misspecification and incorrect policy prescription would be made.