| Many factors lead to volatility of stock market. For example, overall economy, inflation, trading strategies, market sentiment, and firm itself. Actually, all the information related to financial has an impact on stock market. We can classify the information into quantitative information and qualitative information. Quantitative information is actual observed data that we can get immediately, such as stock price, book-to-market, profit. But qualitative information is what we can’t depict exactly, such as business environment, cultural level, technical advantages, war, economic policy of government, natural disasters and so on. Online financial news involves large number of this information.It’s the fact that news has an impact on stock market. With the popularity of Internet, online financial news is the main way of people getting financial information. So we can deduce that there must be some relationship between Internet financial news and volatility of stock market. Many researchers work for the specific relationship, including computer science and economics.The researchers of computer science focus on mining relationship between news content and stock price. However there are so many factors affecting the volatility of stock price that prediction accuracy can’t be improved very much, which makes the research has little practice use.The researchers of economics focus on how the financial news affects the volatility of stock market, but news they research here only means news title or news number simply rather than news text content. News texts contain so much useful information. The reason why they don’t mining information from news text is that this is exactly the knowledge of computer science.In a short, the relationship of news content and stock market has not been penetrated into by the academe because of the limitations of domain knowledge, and there are some blank. This paper further investigates the relationship between news text and volatility of stock market by combining text mining technology in computer science and multivariate linear regression model in econometrics. The proposal technique is as follows:(1) We use text mining technique to quantify the impact of news text on stock market and the quantitative result is used as an effective factor of stock market. In this stage, vector space model, TFIDF, Chinese Word Segmentation, support vector regression and dimension reduction are adopted. To choose the best technique is especially important in this stage.(2) The quantitative result of news content on stock market and the main technical indicator in stock market are both used as explanatory variables, and the cumulative abnormal return is used as dependent variable. Then we adopt statistical theories of hypothesis tests to find how the news text content affects volatility of stock market exactly.Through the experiment, we find that news has an impact on volatility of stock market of both in Shanghai Stock Exchange and Shenzhen Stock Exchange, however, the impact in Shanghai Stock Exchange is less significant than that of Shenzhen, and the impact time in Shenzhen Stock Exchange is longer than that of Shanghai. Meanwhile, we find that the smaller scale of company is, the bigger affect strength of news on stock market is.The major innovation of the paper boils down to three parts as follows:First of all, in research methodology, this paper combine text mining in computer science and multivariate linear regression model in econometrics, which resolves an interdisciplinary problem.Secondly, in angles of study, most research about news and stock market focus on prediction, however, this paper works for how news content affects volatility of stock market in detail.Finally, it has a technical contribution. Based on the characteristic of stock market, we firstly establish stock features library, and then establish thesaurus library. The former is used to convert text document to vector and make word segmentation more accurate, and the later is used to lower the dimension of vector. Eventually, the vector space model is better than normal vector space model.The whole paper is divided into five parts. The first part introduces the background of the research content related with this paper. The second part describes the theoretical basis and technique used in this paper in detail. The third part introduces the data preparations. The fourth part introduces the key part that is experiment. The last part is the summary and expectation of this thesis. |
PDF Full Text Request