Harvard Study Shows AI Can Predict 71% of Fund Manager Trades, Raising Questions About Fees

AI Exposes Predictable Patterns in Active Fund Management

A Harvard study published by the National Bureau of Economic Research has revealed that AI can predict active-fund trades with 71% accuracy, raising fundamental questions about what mutual fund managers actually deliver for their fees 1

1

. The working paper titled "Mimicking Finance," co-authored by Harvard Business School finance professor Lauren Cohen alongside Yiwen Lu at the University of Pennsylvania and Quoc H. Nguyen at DePaul University, analyzed trading decisions from 1990 to 2023 using a machine learning algorithm called a neural network 1

1

.

The system could predict trade directions—whether a manager would buy, sell or hold a given stock over a quarter—by drawing on information such as fund size, investor flows, stock characteristics and broader economic conditions 1

1

. This capability suggests that much of what passes for human judgment in the $54 trillion asset management industry actually follows detectable investment patterns that machines can map and potentially reproduce at lower cost 3

3

.

Where Genuine Alpha Actually Lives

The most revealing aspect of the research may be what the AI stock trading system couldn't predict. The roughly 29% of trades that fell outside the algorithm's grasp were, on average, more closely associated with outperformance 1

1

. "If 71% of your decisions can be anticipated by an algorithm, it becomes very hard to justify active-management fees for that portion," Lauren Cohen explained. "Now, the non-routine trades, the ones our model can't predict, are where genuine alpha lives. But those account for a relatively smaller share of overall activity" 1

1

.

This finding reframes what active management consists of in practice. While some predictable buy-and-sell trades serve essential purposes like managing liquidity, adjusting risk or rebalancing portfolios, the bulk of that activity appears to be rules-based rather than requiring expensive discretion 1

1

. The implication isn't that machines have cracked markets, but rather that they've learned much of the industry's common playbook—how fund managers tend to react to flows, market trends and their peers.

Predictability Varies Across Manager Types

The research found that predictability differs significantly based on manager characteristics. Senior managers with longer trading histories, especially those operating in less competitive categories, proved easier for AI to mimic with 71% accuracy—and in some cases, nearly all their trades in a given quarter could be anticipated 2

2

. Larger funds, those charging higher fees, those run by bigger teams and those facing more competition tend to be less predictable on average 1

1

.

Interestingly, managers with larger ownership stakes in their funds were harder for automation tools to replicate, suggesting that having more skin in the game makes human decision-making more valuable and less algorithmic 3

3

. This finding offers a potential path forward for fund managers seeking to differentiate themselves in an era where AI could replace fund managers who follow predictable patterns.

Implications for the Asset Management Industry

The study arrives as fears of AI disruption have already been sending shares of wealth managers and financial services companies swinging sharply 1

1

. For years, investors have been shifting money out of stock-picking funds and into low-cost index products, and this research adds quantitative weight to concerns about whether high active-management fees can be justified when machines can predict trade directions for the majority of portfolio adjustments 1

1

.

What distinguishes this research from earlier work is its focus on anticipating trades themselves rather than merely dissecting returns after the fact. The authors argue that machine learning models, particularly neural network architectures, are better suited than traditional linear factor models to capture the complex ways managers respond to flows, market signals and one another . The model currently predicts the direction of trades rather than their size—a limitation the authors plan to address in future work . As AI capabilities continue to advance, the pressure on mutual fund managers to demonstrate value beyond algorithmic patterns will only intensify.