A new study indicates that artificial intelligence can predict a significant portion of active fund managers’ trading decisions, raising questions about the value of active management fees. The research, led by a Harvard Business School professor, found that a machine-learning algorithm could predict approximately 71 per cent of mutual fund trading decisions, such as whether a manager would buy, sell, or hold a particular stock. The model was trained using data from 1990 to 2023, considering factors like fund size, investor flows, and economic conditions.
The study revealed that the trades the algorithm failed to predict were more closely associated with outperformance. This suggests that the value lies in the investment activity outside of routine, detectable patterns. Lauren Cohen, a finance professor at Harvard and co-author of the paper, noted that if a substantial portion of decisions are predictable, justifying active management fees becomes challenging. The study highlights that the genuinely skilled component of fund management lies in the smaller share of unpredictable, non-routine decisions.
The findings arrive amidst growing concerns about AI disruption across various professional sectors. For active fund managers, who aim to achieve returns above market benchmarks, the critique is not new as investors have been shifting towards lower-cost index funds. This research further erodes the case for active management by demonstrating the extent to which trading decisions can be anticipated. The study, titled “Mimicking Finance,” used machine-learning models to capture how fund managers respond to flows and market signals.
While the model predicts the direction of trades rather than their size, the authors plan to address this limitation in future work. The research suggests that predicting manager behaviour is easier than predicting market movements, as professional habits tend to follow recognisable patterns. The policy implication, according to Cohen, is less about replacing managers entirely and more about repricing the value of their predictable versus unpredictable activity.
