During the last decade, alternative sources of data, including those that can be used to measure social media sentiment, have grown. Social media sentiment, based on StockTwits and Twitter messages, exhibits a correlation with liquidity measures that cannot be explained by news sentiment.

To measure sentiment, the authors use data from RavenPack, a database provider of news events associated with equities, and assess each news event for relevance, novelty, and sentiment. They use the RavenPack’s Composite Sentiment Score as a measure of news sentiment, determined by analyzing words and phrases in news text.

• The authors construct a regression analysis, intraday event studies on abnormal social media sentiment, and an intraday trading strategy. Several measures of liquidity and volume, such as the log number of trades and log number of quotes, are regressed on social media sentiment indicators.
• For the intraday event studies, the authors use a universe of 500 high-cap, high-volume stocks and match intraday trading data with intraday message data from StockTwits and Twitter. They then categorize messages into positive and negative events. An abnormal positive (negative) event has a social media sentiment score at least three standard deviations above (below) the average social media sentiment for that stock.
• The authors test two strategies that trade every 30 minutes—a benchmark mean-reversion strategy and a social media–augmented strategy—using Quantopian and data from 1 January 2011 to 31 December 2014. The intraday mean-reversion trading strategy trades in 30-minute intervals and buys stocks that exhibited negative returns over the previous window and shorts stocks that exhibited positive returns. The social media–augmented strategy bets on mean reversion after a high volume of social media events by placing more positive or negative weight on equities that had a high StockTwits and Twitter message volume in the prior 30-minute window.

Similar to existing research in behavioral finance, the authors’ findings reinforce that negative social media sentiment has a more significant effect on liquidity measures than positive social media sentiment. News and social media information can be used to predict liquidity measures prior to the market opening by using pre-trading measures of sentiment.

• The authors find much more demand for liquidity when social media sentiment is negative than when it is positive. Social media activity prior to markets opening predicts a higher demand for liquidity during the trading day; positive social media sentiment predicts more mini flash-crashes and a lower supply of liquidity.
• Highly abnormal social media sentiment is preceded by very high momentum and followed by mean-reverting returns.
• A social media–augmented trading strategy that doubles down on stocks with high message volume consistently outperforms the benchmark mean-reversion strategy that ignores social media activity.

The feedback effect (causality) between social media and markets is not explored in this article: Sentiment on Twitter and StockTwits could be affected by price movements, and this sentiment may directly affect trading activity. Some Twitter and StockTwits users are likely more influential than others in the social network and could have a stronger effect on community sentiment. This dynamic could provide direction for future work.