Man versus Machine: A Comparison of Robo-Analyst and Traditional Research Analyst Investment Recommendations

Advancements in financial technology (FinTech) are revolutionizing product offerings across the financial services industry. As of 2018, more than $50 billion had been invested in 2,500 companies that are redefining the way in which individuals participate in financial markets (Accenture, 2018). Innovations in FinTech also appear to benefit end users, with recent evidence indicating that FinTech is enhancing lending and brokerage activities (D’Acunto et al., 2019; Fuster et al., 2019; Tang, 2019; Vallee and Zeng, 2019). Despite its growing importance and relevance, our understanding of how FinTech affects the production of investment information and the role of sell-side research analysts remains relatively unexplored. This is an important issue as the combination of constrained research budgets coupled with the challenges associated with analyzing increasingly massive amounts of disclosure suggest that the traditional research model is ripe for disruption.

In a new study, we provide the first large-scale empirical investigation of the properties of the investment recommendations produced by “Robo-Analysts,” which are human analyst-assisted computer programs conducting automated research analysis. Robo-Analysts represent an important innovation in the research industry as they can potentially analyze large amounts of financial data and generate stock recommendations that are less subject to the limitations of those generated by human analysts, which include behavioral, cognitive, or incentive-driven biases (e.g., De Bondt and Thaler, 1990; Michaely and Womack, 1999). Our unique dataset tracks the activity, recommendation revision patterns, and investment value associated with approximately 75,000 reports issued by seven prominent Robo-Analyst firms over the past 15 years.

Robo-Analysts ostensibly provide a straightforward value proposition: research reports that are both comprehensive and unconflicted (New Constructs, 2019). We test and evaluate three distinct empirical predictions related to these benefits. First, we assess whether automating the research process can reduce the cognitive and economic incentive-driven biases that analysts typically face. If so, we expect Robo-Analysts to produce, on average, less optimistic recommendations than do traditional analysts. Second, we examine how automation changes the research production process. We expect an automated research process to facilitate the production of more frequent recommendation revisions and also lead Robo-Analysts to more easily incorporate information contained in large, voluminous corporate disclosures. Third, and perhaps most important, we assess the profitability of Robo-Analyst investment recommendations. To the extent that Robo-Analysts are able to provide more objective and comprehensive research, we expect their recommendations to ultimately be more profitable for investors.[1]

Our results indicate several interesting trends in Robo-Analyst reports. First, we find that Robo-Analyst recommendations are significantly less skewed towards buy recommendations. For a given stock in our sample, Robo-Analyst firms are 14 percent less likely to have an outstanding buy recommendation and 16 percent more likely to have an outstanding sell recommendation than are traditional analysts. Second, we document differences in the research processes that Robo-Analysts employ. We find that Robo-Analysts revise more frequently than traditional analysts, issuing about one additional recommendation revision per covered firm per year. Robo-Analysts also rely on different information than traditional analysts. They are less likely to revise following an earnings announcement but instead tend to revise following a periodic filing.

Perhaps most important, we document mixed evidence on the return reactions to Robo-Analysts. In short-run market tests, Robo-Analysts do not elicit significant market reactions, suggesting that Robo-Analyst recommendations may have lower investment value or Robo-Analyst research firms are less high-profile and investors have limited awareness of or aversion (i.e., aversion to algorithms) to Robo-Analyst reports. To further assess the investment value of Robo-Analyst recommendations, we conduct an implementable trading strategy that forms daily portfolios based on the buy and sell recommendations issued by Robo-Analysts versus traditional analysts and then compares the returns to the buy and sell portfolios across each contributor type (i.e., Robo-Analysts versus traditional analysts).

Our portfolio analyses indicate several striking trends. First, the portfolios based on the buy recommendations of Robo-Analysts earn abnormal returns that are statistically and economically significant (annualized returns range from 6.4 percent to 6.9 percent). In contrast, the returns of portfolios based on human analyst buy recommendations earn abnormal returns that are weaker in terms of statistical and economic significance (annualized returns range from 1.2 percent to 1.7 percent). The incremental difference between alpha yielded from Robo-Analysts’ buy portfolios relative to traditional analysts’ buy portfolios is also statistically significant. For sell recommendations, however, we find no evidence to indicate that Robo-Analysts’ recommendations are incrementally more profitable than human analysts. If anything, our results indicate that portfolios based on Robo-Analysts’ sell recommendations generate positive, instead of negative, abnormal returns.

Overall, our evidence paints a textured picture of the role of Robo-Analysts in capital markets. On the one hand, their reports appear to offer some value to traditional investors, as they are less biased and revised more frequently. In addition, our portfolio analyses suggest that their buy recommendations generate abnormal returns that are higher than those generated by traditional analysts. On the other hand, their sell recommendations do not appear to be profitable. In addition, we expect that traditional analysts still likely add significant value through their softer product offerings, which are unavailable to common investors. In sum, automation appears to lead to an improvement in the aggregate quality of research available to individual investors, but it is unlikely that this approach to research can meet all of the objectives of traditional brokerage house services.


[1] We also assess the short-run market reactions to Robo-Analyst reports. We do not necessarily expect to observe short-run market reactions to Robo-Analysts’ recommendation revisions as Robo-Analysts may be less prominent than traditional analysts. In addition, investors may suffer from a bias known as “algorithm aversion” (e.g., Onkal et al., 2009). We discuss this issue in more detail below.

This post comes to us from Braiden Coleman, a doctoral student at Indiana University’s Kelley School of Business, and Kenneth Merkley and Joseph Pacelli, professors at the school. It is based on their recent paper, “Man versus Machine: A Comparison of Robo-Analyst and Traditional Research Analyst Investment Recommendations,” available here.