The Handbook of News Analytics in Finance

9.5 CONCLUSION

As the preceding results demonstrate, news sentiment can be used to obtain long-term positive excess stock returns. Our brief event study illustrates a method for picking stocks that emerge from a period of media pessimism with good forward 1-year excess return potential. There is, however, much room for refinement in the methodology. For example, if the minimum time period over which reversals occur were extended from one month to one year, perhaps the optimal net news sentiment measurement period would decrease from 17 months to only a few months, and the forward returns might improve. Maybe instead of focusing on relevance scores, what should be used to supplement the MCQ rankings is DJNA data related to earnings announcements or novelty scores.⁵ Another idea from Hafez (2009) is to control for the seasonality of news, although such an approach is more germane to shorter net news sentiment measurement periods. The possibilities are endless. On the other hand, the plethora of available variables suggest the very real danger of data-fitting. More is needed than the bare correlation of variables to be convinced of the presence of an anomaly. Sentiment reversals as defined here seem to be supported not only by good performance, but a firm, intuitive hypothesis with evidence of monotonicity in the relationship between the primary variable (net news sentiment) and forward returns.

Future work on sentiment reversals might concentrate on developing a more market-neutral ...