Social media platforms like Twitter host thousands of investor opinions daily, often influencing short-term market movements. Context Analytics (CA) has been systematically collecting and analyzing financial tweets for over 10 years.
Through our proprietary natural language processing (NLP) pipeline, CA generates security-level sentiment metrics known as S-Factors. These are calculated over a rolling 24-hour window, updated continuously, and designed to capture real-time investor sentiment across the equity market.
What is an S-Score?
The S-Score is a core S-Factor that quantifies recent sentiment for each security:
- Exponential weightinggives more importance to newer messages
- Normalizationadjusts for each stock’s historical sentiment mean and standard deviation
- Scale:from -1.0000 (extremely negative) to +1.0000 (extremely positive)
This allows analysts and algorithms to compare sentiment across stocks while neutralizing the effects of message volume.
How Does S-Score Predict Returns?
The S-Score is used to evaluate the tone of sentiment over the previous 24 hours, which is expected to have a positive relationship to subsequent daily equity returns. Thus, securities with extremely positive S-Scores are expected to outperform securities with extremely negative S-Scores.
In the analysis below, we evaluated how S-Scores predicted next – day stock returns by grouping securities into sentiment quintiles based on the S-Score calculated at 3:40 PM ET, 20 minutes prior to market close.
Securities are required to have a price above $5 and are equally weighted within each quintile. The Q5-Q1 spread represents a hypothetical Long/Short strategy which is a combination of going long the highest sentiment quintile (Q5) and short the lowest (Q1).
The analysis reveals a strong monotonic relationship between Twitter sentiment and next-day stock performance:
- Securities with highly positive sentiment(Q5) consistently outperformed.
- Securities with extremely negative sentiment(Q1) consistently underperformed.
- Along/short strategy (Q5 minus Q1) produced a stable return profile:
- Sharpe ratio:21
- Annualized return:57%
- On average,over 1,800 securities were sorted into sentiment quintiles daily, providing strong cross-sectional coverage.
These results support the use of real-time Twitter sentiment as a reliable input for short-term return forecasting.
How Volume Filtering Improves Signals
To enhance signal quality, we filtered out securities with low social discussion volume and created a required minimum. To be included in the quintile plot, volume must be greater than the securities 20-day historical average (SV-Score > 0).
This reduces the influence of outlier opinions from one or two users or securities with abnormally low message volume, which may otherwise skew sentiment values.
The introduction of a message volume filter significantly enhanced the performance spread between the highest and lowest sentiment quintiles.
- Positive sentiment = outperformance
Stocks in the top sentiment quintile (Q5) consistently outperformed. - Negative sentiment = underperformance
Stocks in the bottom quintile (Q1) consistently underperformed. - Long/short strategy (Q5 – Q1):
- Cumulative return:+30.25%
- Annualized return:57%
- Sharpe ratio:21
- Coverage:
- Over 1,300 securitieswere sorted into sentiment quintiles daily
Although the number of securities per quintile dropped from 371 to 262 after filtering, the long/short strategy’s performance improved — adding over 2.5% in annualized returns in the 3.5 year period.
Dataset Overview and Use Cases
The S-Factor feed remains Context Analytics’ most mature product offering, backed by over a decade of out-of-sample data. Additional metrics, such as the S-Volume or S-Buzz, can serve as complementary or alternative filters to low message volume, providing further flexibility for research and model development.
The availability of sentiment data from a real-time social media source like Twitter enables robust back testing and strategy development grounded in behavioral finance principles. Researchers, data scientists, and quantitative analysts can leverage this dataset to assess alpha signals, sentiment risk factors, and more.
Explore the Data
To learn more about the S-Factor dataset or request access, visit
www.contextanalytics-ai.com or contact the team at ContactUs@ContextAnalytics-AI.com.
TL;DR:
Context Analytics analyzed over a decade of Twitter data using its proprietary S-Factor framework. Results show that securities with highly positive social sentiment significantly outperformed those with negative sentiment. A long/short portfolio using top vs. bottom S-Score quintiles returned +42.36% over 3 years with a Sharpe ratio of 1.37 when filtered by message volume.