Quantifying Market Efficiency: Information Dissemination Through Social Media

We examine stock market efficiency using Twitter as a proxy for the dissemination of public information. We use a dataset 8,221,848 tweets as a source of timestamped information and perform three different topic extraction methodologies to mine the information that they carry on different types of trading days. Using the same approach, we build a set of classifiers to predict the market movements based on the tweets of the previous day and validate it using an independent sample on five indices of the New York Stock Exchange. Our best classifier can accurately predict 55.99% (45.51%) of bull (bear) trading days, suggesting that the rest of the market movements are either based on private information or are due to market anomalies, thus pointing to semi‐efficient market. By executing our approach on subperiods corresponding to financial turbulence, we show that market efficiency increases during such periods, since public information as proxied by Twitter can explain a greater percentage of market movements. We confirm the findings using counterfactual analysis. Our results add to the discussion on market efficiency and show that Twitter can accurately proxy information propagation towards investors, suggesting a new methodological tool to test for the efficient market hypothesis.