Twitter Sentiment for ETF Price Movement

At Context Analytics, we lead the way in transforming unstructured financial data into quantifiable market signals. Our flagship product, the S-Factor feed, uses social media data to generate signals across various asset classes, including global equities, futures, FX, crypto, and ETFs. In this research, we focus on using Twitter sentiment data to signal ETF price movements, specifically close-to-close changes.

We engineered a new feature using fields from our S-Factor feed. The standard S-Score measures sentiment from Twitter messages on a scale of -1.000 to 1.000, aggregates it over a 24-hour period, and compares it to a 20-day historical baseline.

Another feature, S-Dispersion, captures the number of unique accounts tweeting about the topic, divided by the total tweet volume. This value ranges from 0 to 1, where 1 indicates every tweet comes from a different account, and values closer to 0 suggest most tweets come from a few accounts.

We created a new metric by multiplying S-Score by S-Dispersion, giving more weight to sentiment scores when the tweet volume is spread across multiple accounts. This reduces the impact of a single account driving sentiment.

Using this new metric, we divided our ETF universe, which covers over 2,100 ETFs historically, into daily quintiles. Each day, before the market close, we assign ETFs into quintiles based on their S-Score * S-Dispersion value, and then hold them until the next day's close. We’ve accumulated returns for each quintile since the start of 2019, and tracked performance starting in 2024 to capture more recent trends.

Since 2019, we've observed a strong relationship between the S-Score * S-Dispersion quintiles and close-to-close returns. Notably, Quintile 1 significantly underperforms, with returns around 60% lower than Quintile 2. Meanwhile, Quintile 5 outperforms the others, offering the best risk-adjusted returns. The spread is almost monotonic, though Quintiles 3 and 4 show some variability.

From the beginning of 2024, this relationship has grown even stronger. Quintiles 5 and 4 have outperformed, while Quintiles 1 through 3 have underperformed. Quintile 5’s year-to-date returns are ahead of SPY, highlighting its exceptional performance.

Overall, incorporating S-Dispersion into the weighting of the Sentiment Score creates a powerful metric for identifying short-term price movements in ETFs. This is just one of the many ways to leverage our S-Factor feed to uncover stock market signals. For more information, visit www.contextanalytics-ai.com .