以大數據探究財經新聞對台灣股票市場表現之影響

Abstract

一直以來，許多研究都以財經資訊為切口，探究新聞資訊對投資人乃至股票價格漲跌的影響。這些研究不僅來自金融學領域，也大量存在於資料科學與傳播學領域。然而，多數過往研究傾向從西文傳統媒體之資訊與國外股市之低頻交易資料中，探究二者之關聯，對國內中文財經資訊與股市資料的關注極少。同時，它們也少有關注財經新聞來源這一可能影響研究結果的重要變項。 本研究收集了《工商時報》電子版、《蘋果日報》網路版、《鉅亨網》、《自由時報》電子報、聯合新聞網五家網路媒體一年的全部財經資訊，輔以台灣股市每分鐘的高精度交易資料，並透過支援向量機這一機器學習模型，測量財經新聞對股價漲跌的預測準確率，以此探究財經新聞資訊對於台灣股票市場之表現的影響力。研究結果證明，台灣財經新聞對股票市場之表現存在顯著影響，且該影響在不同媒體、不同類別之股票下不盡相同。研究不僅在台灣財經新聞與台灣股市上，印證了二者之關聯，填補了該項空白，還針對中文語料之特點，比較了不同的文字探勘技術，對後續研究有一定的借鑒意義。

In recent decades, many researchers from different background, such as finance, data science and communication, have focused on finding the proof of financial news influence on stock market. However, most of them just concentrated on relationship between news from traditional media (newspaper, magazines, TV program, etc.) and daily stock price fluctuations. Only few researches tried to analyze reports from Taiwanese online media and intraday stock price data, which was more precise than before. Here we report a result providing some strong evidence of financial news influence on stock market in Taiwan. We collected financial news from five local Taiwanese online media and minute-level stock price data. By utilizing support vector machine algorithm, we found the accuracy of prediction for stock price fluctuations is always more than 50%, which means strong relationship between financial news in Chinese and stock market in Taiwan. More precisely, that influence varies with different media and various kinds of stocks. We anticipate that this result will prompt other researchers to find out the underlying reason of that influence in various. What is more, comparison between two feature extraction methods in our experiments will offer some suggestions for following related researches.