Study Finds Robot Stock Pickers Outperform Humans

Chris Gaetano
Published Date:
Feb 11, 2020

In one of the first formal studies of its kind, researchers have found that, over the long term, robots outperform humans when it comes to choosing investments. They not only find good picks more often, they do so faster and more cheaply than people, according to Bloomberg. The study, from Indiana University, examined about 76,000 reports issued by seven different robo-analyst firms between 2003 and 2018, and compared them to human analysis from the Institutional Brokers Estimation System (IBES). 

What they found was that "buy" recommendations from  robo-analysts generated economically and significantly positive abnormal returns that went well beyond what humans could do. Portfolios based on the buy recommendations of robo-analysts earned positive abnormal annual returns (alpha) of 6.4 to 6.9 percent. In contrast, a portfolio based on human-made buy recommendations saw an alpha of only 1.2 to 1.7 percent. 

The reasons behind this disparity could come from fundamental differences in how humans versus machines evaluate stocks. For one, researchers found that the machines collectively produce a more balanced distribution of buy, hold, and sell recommendations than do human analysts, which suggests that they are less subject to behavioral biases and conflicts of interest. Second, consistent with automation facilitating a greater scale of research production, robo-analysts revise their reports more frequently than human analysts and also adopt different production processes. Their revisions rely less on earnings announcements, and more on the large, volumes of data released in firms’ annual reports. Third, robo-analysts’ reports exhibit weaker short-window return reactions, suggesting that investors do not trade on their signals.

The humans are not completely beaten yet, however. While the robots vastly outperform the humans on buy side recommendations, there seemed to be next to no performance difference on the sell side, the difference close enough to zero that the researchers just said it was zero. 

Click here to see more of the latest news from the NYSSCPA.