運用文字探勘與資料採礦技術建立匯率預測模型－以人民幣兌新台幣為例

由於貿易自由化與國與國之間貿易的蓬勃發展，匯率的變動成為影響利潤的重要關鍵之一，而台灣為一個四面環海之島型國家，其大部分的資源都亟需靠國際貿易來補足，經濟發展更是如此，因此能夠精準的預測匯率的趨勢變得十分重要。而中國無論是與我國或其他各國在政治與經濟方面都扮演重要的角色，因此本研究以人民幣兌新台幣為例，並且利用文字探勘與資料採礦的技術建立預測模型。本研究收集從2012年10月到2013年3月之新聞文件做為文字探勘之資料庫，並且根據文字探勘所提供的訊息挑選出與匯率變化有關之可能變數，並透過相關分析與特徵選取等方法挑選出最終建模變數。透過複合式時間序列演算法建立短期與長期之時間序列模型，模型結果上可發現在短期預測上預測值與其本身匯率（人民幣兌新台幣）與加權股價指數前三期有著重要的關聯性，而在長期預測上模型為ARIMA(1,0,1)。且根據模型評估結果，本研究所提出之預測模型為一個高精確度之預測模型。

關鍵字

匯率；文字探勘；資料採礦

並列摘要

As trade liberalization with other countries and rapid development of trade between countries, exchange rate fluctuation became a important thing for profits. Taiwan was an island country and most of resources were needed to make up by international trade, especially in economic development. So, it was very important thing for predicting accurate exchange rate trends. China played a important role in political, economic and many aspects with Taiwan and other countries, therefore in this study took RMB against NTD for example, and used text mining and data mining technologies to establish predictive models.In this study, we collected news documents as text mining database form October 2012 to March 2013. According to information form text mining, we identified the possible variables that related to changes in the exchange rate. Use correlation analysis and feature selection methods to choose final modeling variables. By Composite Time Series algorithm established short-term and long-term time series model. According to the model results, we found significant relevance with its own exchange rate (RMB against NTU) and TAIEX at the past tree historical value in the short-term predictive value and ARIMA(1,0,1) is a long-term prediction model. In the model evaluation results, we provided a highly accurate model of prediction.