The Handbook of News Analytics in Finance

5.4 CONCLUSION

Considering the relevance and impact of different company events is an important element when constructing market and industry sentiment indexes. Furthermore, applying event novelty as a filter is shown to bring significant value in predicting future returns of the S&P 500. Especially, I find that market-level sentiment strategies significantly outperform 1-month price momentum with information ratios of 1.75 vs. 0.40 and annualized returns of 26.5% vs. 6.8%. In addition, the market-level sentiment strategy delivers double-digit positive returns in four out of five years. In order to measure the impact of different company events, I consider RavenPack's Event Sentiment Score, which indicates how event categories are typically rated by financial experts as having positive or negative share price impact. Also, I use RavenPack's Event Novelty Score to measure how “new” or novel a news story is over a 24-hour time window. Rather than trading based on sentiment index levels, I find value in trading on deltas or monthly index changes. By aggregating news sentiment data over a trailing period of 3 months, similar news flow characteristics are represented in each window, addressing seasonality caused by quarterly earnings reporting. Beyond capturing market-level sentiment, I calculate industry-level sentiment indexes using multiple news classifiers which provide diversity and more sentiment data. I apply a slightly less restrictive company relevance criteria while still ...