USING TWITTER TO GAUGE NEWS EFFECT ON STOCK MARKET MOVES
Sam Paglia · 2013
The rapid rate of data creation, interconnected graph structure and variety of data captured make Twitter an ideal candidate for characterizing the complicated and rapidly changing investor sentiments that move financial markets. In their 2010 work on the subject, Bollen et al1 claimed that sentiment analysis on a broad Twitter corpus can lead to 87.6% accuracy in predicting daily changes in the Dow Jones Industrial Average. In this project, I use a di↵erent method. Specifically, I take a news-based approach and aim to examine the specific e↵ect of news releases via Twitter on market moves. Given that all major news outlets have twitter feeds that they run in parallel to their other distribution channels, this project aims to explore two key questions. The first is whether an aggregated body of news tweets can be used as the basis for a viable prediction strategy. The second is an analysis of what key phrases in tweets tend to be associated with directional market moves My results indicate that, while news-feed analysis lacks significant predictive power as a binary classification tool of market directional moves, it does show promise as a multinomial classifier of market moves into discretized buckets. In this case, appropriately tuned Multinomial Naive Bayes and Multiclass SVM models show promise as Twitter-news-based predictors of market moves