摘要 就貿易為重心的國家而言,匯率變動對企業及個人的投資與獲利影響極大;跨國企業因世界佈局的需要,在資本移動的管理上,避免匯率變動產生營運成本的增加,在匯率預測及走勢的掌握,以降低企業資金的成本與風險,提升產品在國際市場的競爭力。因此預測的準確性,成為國際貿易的重要課題﹔此外,歐元的興起國內投資人可藉購買歐洲基金,以調整及分散手中持有的外匯比重,能否準確地預測其匯率走勢,對資產的獲利及避險,顯得格外重要。 本研究以影響匯率變動因素,M2貨幣供給額,工業生產指數,短期利率,長期債券利率,消費物價指數,生產成本指數,歐元整合區域相關經濟變數,以類神經網路之倒傳遞網路,建構歐元匯率預測模型,並佐以迴歸模型作為相關研究參考。 本研究發現預測模式需以資料特性為依歸,在歐元匯率每日變動預測上,受限於在自變數均為月統計資料,所以採多變量預測每日歐元匯率並不合適,而是以時間序列預測有較為合理解釋。此外,倒傳遞類神經網路模式在歐元匯率誤差均方根、平均誤差百分比與預測準確率方面,相對於統計迴歸模式有較佳之表現。且就模式而言,歐元匯率及其他自變數資料隨時間而成長,因此資料超過二期以上的預測,必須將資料、模式作重新的推估與預測,可得出較準確結果。
For a country that focuses on international trade, the change in foreign exchange(FX)rate has a great influence on the profits of both businesses and individual investments. Most multinational corporations have their operations countries. As a result, in terms of capital management, multinational enterprises have to predict the change in FX rate and the trend of FX rate in order to avoid the increase of operation cost due to revaluation in FX rates . An accurate prediction in FX rates can help businesses to lower their costs and risk , while increase the competitive ability of their products in the international markets. Consequently, the accuracy of prediction has become an important issue in international trade. Moreover, the increasing popularity of Euro dollars enables Taiwanese investors to adjust and diversify the ratio of FX in their hands through the purchase of Euro dollars. Therefore, whether a corporation can precisely predict the trend of FX rate has become very essential. This study applies several factors into the Back-Propagation of Neural Networks to construct a predictive model for Euro dollars. Such related economic variables including money supply amount, industrial production index, short-term interest rate, long-term bond yield, consumer price index (CPI), production prices in Euro dollar integration which are the variables affecting FX rate. Additionally, regression model is provided as a reference. This work found that a predictive pattern should be based on the characteristics of collected data. In terms of the prediction of daily changes in Euro dollars, the accuracy is limited due to the fact that all independent variables are monthly statistic data. Therefore, it was not appropriate to employ multiple variables to predict daily exchange in Euro dollars. This investigation used time sequence for prediction instead. Moreover, in terms of Root Mean Square Error, Average Error Percentage and Prediction Accuracy, the Back- Propagation model of Neural Networks performed better than using the regression model. The results also indicate that the FX rate of Euro dollars and the other independent variables grew along with time period. Consequently, if the collected data exceeded two time-periods, it is required to re-evaluate the data and patterns in order to achieve a more accurate result.