本文旨在探討美國與台灣的總體經濟變數對於美元兌新台幣匯率的預測能力，並利用遠期匯率避險策略之表現，來評估模型的預測績效。 本文依據 Engel and Wu (2023) 所提出之匯率預測模型，以與匯率有關連的總體經濟變數建構預測模型，並進行樣本內與樣本外預測分析。在樣本內預測部分，採用單變量及多變量迴歸模型進行估計，結果顯示部分變數，如美國 CPI、台灣出口外匯收入金額與台灣銀行間隔夜拆款利率，皆對匯率變動具有顯著關聯性，且隨期間愈長，影響更顯著。樣本外預測則結合滾動迴歸 (rolling regression) 與Clark and West (2007) 所提出的Clark-West 檢定，評估各模型相較於歷史平均匯率的預測表現。結果顯示，當預測期間為3 個月及 6 個月時，部分經濟變數模型的遠匯避險績效優於遠期匯率報價，顯示遠期匯率可能未能完全反映所有市場資訊。 另外，本文以出口商角度，利用預測匯率與實際遠期匯率報價比較，作為是否進行避險之依據。實證結果發現，在單變量模型預測下，3及6個月的預測結果優於完全不避險，顯示各總體經濟變數對於匯率趨勢的影響，隨著期間拉長，有其預測能力；而在多變量模型預測，卻與即期匯率總和幾乎相符，可能是因特定期間匯率波動劇烈而削弱預測能力。 ;This study aims to examine the predictive power of U.S. and Taiwan macroeconomic variables on the exchange rate of the U.S. dollar against the Taiwan dollar (USD/TWD). The research framework is based on the exchange rate forecasting model proposed by Engel and Wu (2023), incorporating historical exchange rate data to conduct both in-sample and out-of-sample forecast evaluations. For the in-sample analysis, univariate and multivariate regression models are estimated. The results indicate that certain variables—such as the U.S. Consumer Price Index (CPI), Taiwan’s export amount, and Taiwan’s interbank overnight rate—exhibit significantly significant correlation with exchange rate movements, with their effects becoming more pronounced over longer forecast horizons. This thesis applies the analysis of rolling regression combined with the Clark-West test proposed by Clark and West (2007) to evaluate a specific model’s out-of-sample forecast performance relative to the forecast based on the historical average. The empirical results show that over three- and six-month forecast horizons, some macroeconomic models outperform the forward exchange rate, suggesting that the forward rate may not fully incorporate all available market information. Additionally, from the perspective of export firms, we formulate a hedging strategy based on the comparison of forecasted exchange rates and forward rates in which a firm sells USD forward if the predicted exchange rate forecast is lower than the forward rate. On the other hand, the firm does not hedge with forward if the predicted value is higher than the forward rate. The empirical findings reveal that under univariate models, forecasts for three- and six-month periods outperform the non-hedged scenario, implying that macroeconomic variables possess meaningful forecasting ability as the time horizon extends. In contrast, forecasts based on multivariate models approximate the spot exchange rate, possibly due to high exchange rate volatility during certain periods, which weakens forecast accuracy.
