Twitter mood predicts the stock market

Motivated by behavioral economics, in which emotions strongly influence individual decisions, the paper asks whether entire societies experience collective mood states that shape their decision-making, and whether such public mood is correlated with, or even predictive of, economic indicators. The authors measure collective mood from large-scale daily Twitter feeds using two tools: OpinionFinder, which captures positive versus negative sentiment, and the Google-Profile of Mood States (GPOMS), which characterizes mood along six dimensions (Calm, Alert, Sure, Vital, Kind, and Happy). They cross-validate these mood time series by checking their response to two 2008 events, the presidential election and Thanksgiving.

To test predictiveness against the Dow Jones Industrial Average, the authors apply a Granger causality analysis and a Self-Organizing Fuzzy Neural Network to the mood series and DJIA closing values. They find that including certain public mood dimensions, but not others, significantly improves prediction: daily up-and-down movements of the DJIA are predicted with 87.6% accuracy, and the mean average percentage error is reduced by more than 6%. These results indicate that measurements of collective mood derived from large-scale social-media feeds can improve prediction of stock-market movements.