Sentiment indices and stock returns: Evidence from China

This paper constructs three monthly sentiment indices based on social media, traditional newspapers, and Internet news. The predictive power of the social media sentiment index and Internet news sentiment index in the full sample is excellent, far better than that of the macroeconomic predictors, whereas the index established based on the traditional newspaper is unsatisfactory. The forecasting results of the indices across different business cycles imply that the social media sentiment index has the best predictive power during expansion periods, and the Internet news sentiment index significantly predicts stock returns during recessions. The two combination sentiment indices obtained by principal component analysis and equal combination provide more accurate forecasts of stock returns and can generate great economic value for investors. These results are broadly consistent across robustness tests.