應用灰關聯分析與類神經網路於歐元漲跌預測模式建立之研究

The Study on the Prediction Model of the Euro Exchange Rate by Using Gray Relational Analysis and Neural Network

林維謙(Wei-Cian Lin)

指導教授 ： 李樑堅 ； 黃永成

義守大學/管理學院/財務金融學系/碩士(2016年)

摘要

歐元(Euro)自1999年發行以來即有超越美元之情形，且歐元的發行對國際貨幣與金融體系造成重大之影響，洪德欽(2006)指出近年來歐元已成為國際主要通貨，在國際貿易、貨幣與金融市場取得重要國際地位。但歐盟成員國眾多且組成複雜，加上政治因素與債務危機，使得歐元匯率波動變化大且不易預測，因此預測歐元匯率與波動方向具有探討的價值和意義。
本研究目的在於使用倒傳遞類神經網路建立出能預測歐元匯率漲跌方向的模型，以降低投資人的風險及增加獲利的機會。其方式乃採用2004至2013年十年間的歐元匯率漲跌資料作為預測模型之建立，先利用灰關聯分析篩選出影響歐元匯率的因子，而後界定相關程度高的影響因子導入倒傳遞類神經網路來建立歐元匯率漲跌方向之預測模式；最後使用多元迴歸法來建構預測歐元匯率漲跌方向之計算式，並與本研究建立倒傳遞類神經網路的預測模式做比較，以界定何種為較佳模型，進而提供投資人作參考依據。
實證結果發現，先利用灰關聯分析篩選出影響歐元匯率漲跌高相關之影響因子，而後導入多元迴歸模型可提高其預測能力，表示灰關聯分析為良好的篩選工具；而倒傳遞類神經網路模型之模型均方誤差(MSE值)與實際樣本驗證之誤差皆小於多元迴歸模型，而預測漲跌準確率兩者相同且高達83.33%，綜合而言，類神經網路模型具有較佳的預測能力。

關鍵字

歐元匯率 ； 倒傳遞類神經網路 ； 灰關聯分析 ； 多元迴歸

並列摘要

Euro was launched in 1999, it faces currency issues beyond US dollar and its introduction represents important influences on international monetary and financial system. Hong (2006) pointed out that euro has become international major currency and get an important international status in international trade, financial and monetary markets. European Union member countries are numerous and constitution complicacy, coupled with political aspects and debt crisis affected the euro exchange rate volatility. Hence, case study of the prediction for euro exchange rate and its volatility are meaningful and valuable.
This paper investigate the prediction of euro exchange rate changes model using back-propagation neural network to reduce investment risk and improve profitability. The samples used in this paper are euro exchange rate changes data for the period between 2004 and 2013. First, we find out the impact factors and correlation of euro exchange rate using grey relational analysis. Then, the high correlation impact factors are derived into back-propagation neural network to construct the prediction of euro exchange rate changes model. We also calculate the prediction of euro exchange rate changes model using multiple regression method. Finally, the optimization model for investor decision making has chosen based on comparison between multiple regression method and back-propagation neural network model.
The empirical results show that high correlation impact factors are filtered by grey relational analysis and derived into multiple regression model can improve predication performance, it reveals that grey relational analysis is a good screening tool. The MSE value of back-propagation neural network and the error of actual sample verification are lower than multiple regression model, but the accuracy rate of prediction changes are consistent which is up to 83.33%. As a conclusion, neural network model has preferable predication performance.

並列關鍵字

exchange rate ； back-propagation neural network ； grey relational analysis ； multiple regression

參考文獻

朱旭 (2004)，非線性匯率預測模型之實證研究，國立臺灣大學國際企業研究所碩士論文。

余尚武、黃雅蘭 (2003)，台灣股價指數期貨套利之研究：類神經網路與灰色理論之應用，電子商務學報，5(2)，87-115。

吳靖東 (2012)，總體經濟因素對兩岸匯率變動的影響，東吳經濟商學學報，76期。

李俊毅、羅至美 (2010)，歐元歷史：分析層次的觀點，政治科學論叢，46期。

林敬凱 (2012)，類神經網路於財務危機預測模式之應用：時間預測變數的比較，淡江大學統計學系應用統計學碩士班碩士論文。