In this blog we examine how shifts in security sentiment on Twitter relate to subsequent daily price returns and how message volume influences this relationship.
Understanding the Metrics
Before diving into the results, it is worth clarifying the two metrics at the center of this analysis.
The S-Score Delta is the 24-hour sentiment change computed daily at 15:40 ET (20 minutes prior to market close). Rather than measuring the current sentiment, it captures the change in sentiment relative to a stock's own recent history. Securities are then ranked into quintiles based on their S-Score Delta for that day, and each quintile’s equally weighted portfolio is tracked on a close-to-close basis.
The SV-Score measures how abnormal a security's Twitter message volume is relative to its own historical average. An SV-Score value of 0 means message volume is at its historical norm, while SV-Scores of +1 or +2 indicate message volume is one or two standard deviations above average, respectively.
Quintile Analysis
The first chart displays the cumulative close-to-close returns of each sentiment quintile from January 2016 through March 2026. The results display a clean monotonic relationship between S-Score Delta and forward return: higher-quintile portfolios, containing securities with the most positive change in sentiment, consistently outperform lower-quintile ones over the full period.
Quintile 5, the most positive change in sentiment portfolio, delivered a cumulative return of 257.62% over the 10+ year period with an annualized return of 14.89% and a Sharpe ratio of 0.69. Quintile 1, representing the most negative sentiment, returned 123.45% on an annualized basis of 10.43% with a Sharpe of just 0.47. The spread between them is captured by the Long/Short portfolio (Q5 - Q1), which accumulated 55.69% cumulatively at a Sharpe of 0.96.
The average S-Score for each quintile confirms proper segmentation. Quintile 1 averages –2.217, while Quintile 5 averages +2.522. That means securities in the top and bottom quintiles sentiment on average shifted over 2 standard deviations either positive or negative from 24-hours prior. Average daily holdings are approximately 417 securities per quintile, providing robust statistical coverage throughout the sample.
Layering in Twitter Volume: SV-Score Thresholds
The baseline quintile results are compelling on their own, but the more interesting question is whether Twitter volume acts as a signal amplifier. The second chart answers this by decomposing Quintile 1 and Quintile 5 into four sub-portfolios. This includes the unfiltered universe, instances where SV-Score exceeded 0, SV-Score exceeded 1, and where SV-Score exceeded 2.
| Portfolio | Cumulative Return | Annualized Return | Volatility | Sharpe | Sortino | Avg. Score | Avg. Count |
|---|---|---|---|---|---|---|---|
| Quintile 1 | 123.45% | 10.43% | 22.39% | 0.47 | 0.73 | -2.217 | 417 |
| Quintile 2 | 167.77% | 12.10% | 21.89% | 0.55 | 0.86 | -0.530 | 417 |
| Quintile 3 | 165.89% | 12.02% | 21.87% | 0.55 | 0.86 | 0.025 | 416 |
| Quintile 4 | 225.81% | 14.06% | 22.05% | 0.64 | 1.00 | 0.704 | 416 |
| Quintile 5 | 257.62% | 14.89% | 21.65% | 0.69 | 1.08 | 2.522 | 417 |
| Q5-Q1 | 55.69% | 4.46% | 4.63% | 0.96 | 1.62 | — | — |
Positive Sentiment (Quintile 5)
For the high-sentiment portfolio, increasing the SV-Score threshold meaningfully improves performance. When restricted to securities where SV-Score exceeded 2 (roughly two standard deviations above a company's own historical average), Quintile 5 delivered a cumulative return of 423.88%, which is 166 percentage points higher than the unfiltered Q5 result of 257.62%. Annualized return jumps from 14.89% to 19.14%, and the Sharpe ratio rises from 0.69 to 0.80.
Even the more moderate SV-Score >0 filter adds value: filtering to above-average volume days pushes Q5 to 291.67% cumulatively. The monotonic improvement in average S-Score as SV-Score thresholds rise, from 2.522 to 2.920 to 3.452 to 3.859, suggests that high-volume, high-sentiment days carry genuinely stronger conviction in the underlying signal.
Negative Sentiment (Quintile 1)
The effect is equally pronounced for the low-sentiment portfolio. As the SV-Score threshold rises, Q1 performance deteriorates sharply. The unfiltered Q1 returned 123.45% over the period and applying an SV-Score >2 filter drops that to just 26.60%. That’s a cumulative return nearly 100 percentage points below the unfiltered baseline.
Key Takeaways
Taken together, the data support a clear conclusion: Twitter volume amplifies the directional signal embedded in sentiment scores. Days characterized by both extreme changes in sentiment and elevated social activity are not noisier—they are more informative. The effect is consistent on both sides of the distribution:
Twitter volume is not a source of noise that dilutes sentiment signals. It is a meaningful variable that, when elevated, indicates that market participants are paying unusually close attention. This combined with changes in Social Sentiment is a truly unique factor and alternative way of creating portfolios.
