Applying Techniques of Text Mining on Trading Investment Strategy:an LDA Approach to Distinguish the Topics

105

Sentiment analysis has triggered a heated discussion in recent years, and it can be widely used in various kinds of fields. For example, It can be applied on the detection of network security, the prediction of the president election, the recommendation system on the shopping website, and so on. This thesis aims to apply the sentiment analysis on the trading investment strategy and make use of the articles of Federal Reserve to do the sentiment analysis to predict the return rate of stocks. Moreover, the thesis uses the topic model of latent dirichlet allocation to investigate the latent topics from the articles of Federal Reserve, and the goal is to distinguish the topics which influence the return rate of stock the most from the articles of Federal Reserve. Finally, my research expects to frame a lucrative trading investment strategy based on the research results. The thesis is inspired by the researches of Tetlock (2007) and Tetlock, Saar-Tsechansky, and MacSkassy (2008). First, I will use the topic model of latent dirichlet allocation to classify the words according to different topics. Second, I will eliminate the paragraph which is irrelevant to finance in order to assess the exact financial sentiment and to apply it on investment trading strategy. Last but not least, I will add the derivatives into the investment trading strategy so as to hedge the loss from the wrong prediction of sentiment, and then I will examine the performance of the investment trading strategy after the modification.