運用文字探勘分析網路財經新聞預測美金外匯之研究

Using text-mining technique to analyze online financial news to predict US dollar foreign exchange

黃嘉彥(Jia-Yen Huang) ； 王婉菁(Wang-Ching Wang)

《管理學術研討會》 第十六屆 (2018/11) Pp. 792-801

https://doi.org/10.29893/NCUTMAN.201811.0081

摘要

網際網路技術日新月異，如今的外匯投資者不需要親自前往銀行或外匯機構，便可直接使用網路連線工具甚至手機App進行外匯交易。但同樣因網路的普及與發展，使得外匯市場波動瞬息萬變且資訊量龐大，漲跌幅度甚至遠大於股票市場，使得預測匯率走向極為困難，往往在接到消息面的當下，就已過了該訊息反應的時間點。因此，除了更快接收訊息並判斷影響來決策判斷走勢，來源消息面的正確性亦非常重要，故本研究針對網路新聞暴增時，是否影響外匯美元兌新台幣漲跌的預測準確度進行探討。從新聞網站擷取2018年06月至08月與美元或新台幣相關之財經新聞資料，共計約459筆新聞資料。本研究先運用中研院CKIP系統處理文件進行斷詞，並建立中文新聞詞組規則以萃取文章之情緒詞與程度詞。接著統計新聞預測外匯漲跌情況以驗證隔日實際狀況，藉此分析當日消息面新聞數量的多寡與實際隔日狀況之間的關係。當國際事件發生時，因各種聲浪討論立場不一、眾說紛紜，對於外匯走勢看法分歧，由本研究結果顯示，當新聞量暴增時，將使得外匯預測準確度下降。

關鍵字

文字探勘 ； 財經新聞 ； 外匯兌 ； 預測

並列摘要

As Internet technology improves every day, it is easy to process foreign exchange trading by Internet even mobile App instead of going to bank. The foreign exchange market is multiple changes and the scope of news is huge due to Internet getting popular. It is very difficult to forecast the exchange rate since the change range is more than stock market. It always pass the perfect timing of reaction when receive the news. Therefore it is very important to make decision for trend with immediate news, besides, the accuracy of source of news also important. This study uses the technique of text mining to investigate the effect on the accuracy of predictable rate for foreign currency as the Internet news in previous day is increased. In total, 459 piece of financial news related to US dollar or NT dollar are collected from the news website from June to August in 2018. We use Chinese word segmentation approach to extract Chinese sentences by CKIP and build Chinese phrase extraction rules. Accounting the news and predict the change range of foreign currency and prove it in next day. After that, analyzing the cause-effect relationship between amount of news and exchange rate. There is variety of opinion for trending of foreign exchange when International issue happened. According to the result, the accuracy of predictable rate for foreign currency is getting lower when amount of news or related but not positive news is too much.

並列關鍵字

Text mining ； Financial news ； Foreign currency ； Prediction