Distilling Social Media to Trade the Stock Market

Can aggregate, distilled daily stock sentiment and attention, as extracted from financial social media, usefully predict U.S. stock market returns? In their March 2025 paper entitled “Market Signals from Social Media”, Anthony Cookson, Runjing Lu, William Mullins and Marina Niessner construct purified daily sentiment and attention indexes from millions of posts on StockTwits, Twitter and Seeking Alpha. They focus on the 1,500 stocks most-discussed on StockTwits with at least 10 daily posts on StockTwits. For each firm and each day during 2013 through 2021, they:

• For each source of posts, distill sentiment and attention metrics by excluding firm-specific news events and subtracting slow-moving firm-level attention/sentiment averages.
• For each source of posts, compute market capitalization-weighted averages of the distilled sentiment and attention measurements.
• Apply principal component analysis to combine distilled average metrics into overall sentiment and attention indexes.

They then relate levels of these indexes to subsequent S&P 500 Index returns and aggregate turnover. They further test economic significance of this return predictability via a strategy that employs monthly rolling regressions to assign daily weights (from 100% short to 200% long) to the S&P 500 Index based on predicted return divided by variance of predicted returns over the last 20 days. Using daily posts on the selected platforms as specified above and daily S&P 500 Index returns during January 2013 through December 2021, they find that:

(more…)