Prediction and Trading of Dow Jones from Twitter: A Boosting Text Mining Method with Relevant Tweets Identification

Previous studies claim that financial news influence the movements of stock prices almost instantaneously, however the poor foreseeability of news limits their possibility of predicting the stock price changes and trading actions. Recently complex sentiment analysis techniques have also showed that large amount of social network posts can predict the price movements of the Dow Jones Industrial Average (DJIA) within a less stringent timescale. From the idea that the contents of social posts can forecast the future stock trading actions, in this paper we present a simpler text mining method than the sentiment analysis approaches, which extracts the predictive knowledge of the DJIA movements from a large dataset of tweets, boosting also the prediction accuracy by identifying and filtering out irrelevant/noisy tweets. The noise detection technique we introduced improves the initial effectiveness of more than 10%. We tested our method on 10 millions twitter posts spanning one year, achieving an accuracy of 88.9% in the Dow Jones daily predictions, which, to the best of our knowledge, improves the best literature result based on social networks. Finally we have used the prediction method to drive the DJIA buy/sell actions of a trading protocol; the achieved return on investments (ROI) outperforms the state-of-the-art.