At Context Analytics, our quantitative news feed leverages and enhances proprietary Natural Language Processing (NLP) technology to provide predictive news sentiment, streamlining how financial news is consumed and analyzed for quantitative strategies and alpha generation.
Expanding Beyond Traditional Metrics
Previously, our research utilized raw calculated scores or standardized scores, focusing on average and summated sentiment. In this blog, we explore the use of additional metrics from our feed, specifically the Negative Count and Total Sentiment Count, which are calculated at the article level:
- Negative Count: The total number of negative words/phrases identified by our NLP in an article.
- Total Sentiment Count: The total number of sentiment-indicative phrases (positive, negative, or neutral) in an article.
Methodology
We apply these metrics in a trading setting by taking a 24-hour summation of the Negative Count and Total Sentiment Count for each company before market close. The ratio of these metrics, Total Negative Count / Total Sentiment Count, gives us the Negativity Ratio, indicating the percentage of negative financial language in the previous 24-hour news cycle for a specific company.
A high Negativity Ratio suggests a negative news day, while a low ratio indicates a relatively positive news day. All values lie between [0,1].
To test the predictive nature of this metric, we:
Results
Our historical analysis shows a clear monotonic spread across our universe:
- Quintile 5 (most negative news stocks) significantly underperforms other groups.
- Quintile 1 (least negative news stocks) outperforms other groups, including Quintile 5, by 26%.
The Negativity Ratio is designed to identify securities with bad news. The factor is highly predictive on the short side, but not as much on the long side. None of the portfolios outperform an equally weighted S&P 500 (RSP).
Across all five portfolios, there is an average of 1676 securities traded each day, over triple the number of constituents in the S&P. To find a portfolio using Negativity Ratio that outperforms RSP, we looked at smaller bucket sizes and more extreme Negativity Ratio values.
Exploring Extremes: Decile and Ventile Grouping
To assess the effect of more extreme scores, we analyzed decile (10 groups) and ventile (20 groups) groupings. We compared the top and bottom groups from each:
The change from quintiles to deciles has more effect on the short side than the long side. Decile 1 (top 10%) remains consistent with the Quintile 1 (top 20%), while Decile 10 (bottom 10%) underperforms Quintile 5 (bottom 20%) by over 6%, dropping to -33% cumulatively.
The difference between deciles and ventiles is significant. Ventile 1 (top 5%) outperforms all other groups, including the RSP baseline. The most extreme negative group (ventile 20) underperforms significantly, losing close to 43% while the market remained positive.
We observed changes in average Negativity Ratios across different groupings. For positive groups, the average Negativity Ratio drops from 9.6% (quintile 1) to 6% (decile 1) to 3% (ventile 1). For negative groups, the average Negativity Ratio increases from 44% (quintile 5) to 51% (decile 10) to 59% (ventile 20). These findings show that as the news sentiment becomes more positive, returns increase. Conversely, when news sentiment becomes more negative, returns decrease.
Conclusion
Our research indicates that the Negativity Ratio is a predictive metric for future returns. By extending the analysis to more extreme groupings (deciles and ventiles), we uncover more pronounced return differentials, with the top ventile group even outperforming the S&P 500. This demonstrates the potential for integrating new sentiment into risk and trading models, providing a valuable tool for quantitative strategies seeking to capitalize on news sentiment.
For more information on our quantitative news feed, visit www.contextanalytics-ai.com or click the button below.