Market Efficiency in the Age of Machine Learning

Abstract

As machines replace humans in financial markets, how is informational efficiency impacted? We shed light on this issue by exploiting unique data that allow us to identify when machines access company information (8-K filings) versus when humans access the same information. We find that increased access by machines, particularly from cloud computing services, significantly improves informational efficiency, by reducing the price drift following information events. We address identification through a quasi-natural experiment, instrumental variables, and exogenous power outages. We show that machines are better able to handle linguistically complex filings and are less susceptible to bias from negative sentiment, whereas humans are better at combining incremental information.