Information fusion of stock prices and sentiment in social media using Granger causality

Granger causality is proposed to fuse stock prices and social media sentiment information for stock market prediction. Sentiment extraction is performed on the Twitter data from major stock companies. Analysis shows that authoritative user's sentiment affects the other users after an event with the lag of 3 days. The prediction is performed for Twitter and stock data from four companies. The sentiment analysis algorithm reflects factors of Twitter which are relevant to user's authority to calculate sentiment weight of each message. Information is fused with vector autoregressive models to predict the future stock prices of companies with large twitter support. The fused prediction is shown to have improved performance compared to conventional methods.