Harvard study finds AI can predict 71% of human fund trades
AI excels at routine trading decisions
"Alpha"-generating, convention-defying calls remain human territory
A study has found that artificial intelligence (AI) can predict fund managers' trading decisions with an accuracy of about 70%. However, the ability to break market conventions, discover new leading stocks, and generate returns above the market average - the so-called "alpha" - still remains in the domain of human fund managers.
A trader is at work on the trading floor of the New York Stock Exchange (NYSE). Photo by AFP and Yonhap News
According to a Bloomberg report on the 24th (local time), a research team led by Professor Lauren Cohen at Harvard Business School trained an AI model using actual fund trading records from 1990 to 2023, along with economic environment data from those periods. The team then released a recent paper evaluating how accurately the AI could predict key decisions by fund managers, such as buying and selling.
The study found that AI's average prediction accuracy reached 71%. In particular, it showed high accuracy in "routine" situations where there were no major market fluctuations or anomalies. The researchers explained, "For some managers, the model was able to correctly predict almost all of their trading decisions for specific quarters."
By contrast, the roughly 29% of cases where predictions were wrong occurred mainly in decisions related to stocks that "outperformed" the market average. Bloomberg reported that this suggests AI can largely imitate trading decisions that follow existing trends, but that human judgment remains crucial for investment decisions that deviate from established conventions and patterns.
In an email interview with Bloomberg, Professor Cohen said, "The investment decisions that are difficult for AI to foresee are precisely those where uniquely human expertise comes into play, and we confirmed that this domain does in fact exist." However, he added, "This type of work does not account for a large share of the overall tasks," and said, "The significance of this study lies less in whether to completely replace human managers with AI, and more in recalibrating the value of high-value work."
People sit near an electronic board displaying stock market indices at the Indonesia Stock Exchange (IDX) in Jakarta, Indonesia. Yonhap News Agency, Reuters
Bloomberg analyzed that this study could serve as an opportunity to shed new light on the distinctive role of human fund managers, at a time when concerns are growing on Wall Street that AI could replace various types of work.
It also pointed out that the findings could influence debates over fund fees. Many investors are questioning the high fees of active funds and shifting to low-cost passive funds, amid ongoing criticism that a substantial portion of fund performance may simply come from following market trends and patterns.
According to the study, AI's prediction accuracy also varied depending on the working environment of the fund. Large funds with many investment professionals and intense competition showed prediction accuracy below the average. In contrast, when a single manager handled multiple products or had a long tenure, the AI's accuracy rate was relatively higher. This suggests that the more consistent each manager's investment style is, the more effectively AI may be able to learn it.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
