Social-media sentiment is a good predictor of monthly stock returns, provided traders tweeting about companies aren’t following too many of them, according to researchers in the U.K.


An analysis of tweets from users of social media platform StockTwits from 2010 to 2017 found that social-media sentiment “significantly” predicts stock returns, said University of Edinburgh associate professor Woon Sau Leung and Cardiff University researchers Woon Wong and Gabriel Wong, in a recent paper. StockTwits has attracted a large community of users who share trading ideas by tweeting, giving the authors a sample of 8,575 U.S. publicly-traded companies to evaluate.

They studied almost 30 million tweets using natural language processing, a machine-learning technology, to gauge social-media sentiment surrounding the companies. The researchers said they classified user posts into “bullish” and “bearish” tweets, removing bots from their analysis.


“Researching stocks requires significant amount of time and effort and typically involves gathering, analyzing, and interpreting financial information,” the authors wrote. “The loss in predictability due to increasing stock and industry coverage over future stock returns is only significant among firms that are harder-to-analyze.”

[II Deep Dive: Berkshire Hathaway Bet Big on Dialysis Giant DaVita. Jim Chanos Thinks It’s a Scam.] 


StockTwits users mainly tweet about stock trading, sharing ideas and predictions about stock prices through posts limited to 160 words, according to the paper. The authors studied more than 156,000 users for their finding that social-media sentiment significantly predicts stock returns.


Their research also found the predictability of stock returns declined when the number of stocks users tweeted about increased. While investors who track multiple stocks probably have “superior” knowledge of the broader market, those focused on few companies have an edge in understanding them in greater depth, according to the paper.


“Since human beings have limited cognitive resources over tasks, the attention spent on one task must reduce the attention available for other tasks,” the authors said. “The more specialized an individual user is, the better the quality of her stock analysis.”