Social media is a major communication tool within society and its widespread use has lasting impact on individuals and companies in various industries. Companies have engaged in social listening for quite some time now. Some of them have dedicated social media experts to listen to customers’ conversations on Facebook, Twitter, Instagram, and Pinterest in order to capture customer sentiments, make product recommendations and improvements, or even come up with new service designs. Over the years, we have worked on research studies to demonstrate the economic value of social media information; for example, using Facebook information to predict company performance, using Twitter information to forecast music album sales, and using Google trends to predict macro-economic indicators such as the Consumer Price Index (nowcasting).
Lately, one particular research study related to the use of Twitter in making stock market predictions has received quite some media attention. In this research project, we address the question: if one could know all the conversations embedded in the digital traces people left on the Twitter platform, could an individual investor make better investment decisions and potentially ‘beat’ the market? With such a question in mind, we gathered 21 weeks’ worth of tweets – millions of tweets – about Standard & Poor’s 100 companies. We examined the relationships between their information content using computational linguistics and stock market performance, as well as the role of social influence. More specifically, we investigated the relationships between Twitter message features (message bullishness, message volume, and message disagreement) and stock market performance (stock returns, trading volume, and volatility) both on a daily basis, and on a 15-minute basis. We also studied the possible mechanisms of the efficient information aggregation by studying the extent to which good investment advice receives greater attention. We found that message bullishness is indeed associated with daily abnormal returns. New information, reflected in the Twitter messages, is incorporated into market prices quickly. Users that provide above average investment advice are given a greater share of voice through higher levels of retweets, as well as larger influence. Notably, following expert users amplifies the relationship between message bullishness and abnormal returns. Further, we simulated a set of trading strategies based on our sentiment analysis and the results suggest that it is possible to exploit market inefficiencies even with the inclusion of fixed and variable transaction costs, though the existence of transaction costs challenges investors’ opportunities to collect the gains.