使用類神經網路於台灣貨幣市場匯率趨勢預測之研究

Abstract

臺灣是一個海島型經濟體系，國際貿易為經濟活動的主要動力。是故匯率 的變化也能反應本國與世界各國經濟的互動。所以有效的掌握匯率的變動 是做好外匯風險管理一項重要工作。本研究係透過匯率決定理論的探討， 歸納出影響匯率變動的總經因子，然後對總經因子進行資料前處理及落後 期的推導。最後以將所匯整出來的因素結合逆傳遞神經網路以建立神經網 路預測模型，並以此預測結果與傳統多元迴歸模型，進行實證研究分析。 本研究結果發現: 1.類神經網路應用於匯率預測優於多元迴歸模型。 2.日圓及馬克對美元匯率之趨勢與台幣對美元匯率有顯著相關。 3.經過 資料前處理及落後期推導，總經因子與匯率間趨勢更形顯著。 Since the trend of the exchange rate of Taiwan dollar to foreign currency can reflect the economic growth of Taiwan. A very effective control of the change of the exchange rate of Taiwan dollar to foreign currency is an important study to health the economic growth of Taiwan. Thus, in this study, How to make better foreign currency risk management by efficiently forecasting change of exchange rate is under study. Through the exploring of the exchange rate determination theories, the macroeconomics factors that affect exchange rate movement are summerized. Then, the macroeconomics factors, which have high correlation with the dynamic change of foreign exchange rate, are filtered. Besides, the time lag between the filtered macroeconomics factors with the foreign exchange rate, are calculated by using the statistical stochastic analysis. Finally, a back-porpagation neural network is built based on the time lag concerned macroeconomics factors data set. After the neural network processes, such as learning, testing are finished, an empirical study is setup, and some of the main conclusions are shown below : 1.Using the neural network to forecast the trend of exchange rate in the Taiwan monetary market seems better than it by using regression model. 2.There is high evident relationship that the exchange rate of Yen and Mark to US dollar influence the trend of the foreign exchange of Taiwan dollar to US dollar. 3.Through making the data pre- porcess regarding to the time lag effect. There is more evidence shown that neural network can help us to find the better relationship between the change of macroeconomics factors and the foreign exchange rate, if compared with the methodology of regression model with the same direction.