AI can mimic most mutual fund trades. Here's what managers need to do next
A new working paper from the National Bureau of Economic Research reports that artificial intelligence can predict the direction of about 71% of mutual fund managers' trades. For some managers, it's even higher-nearly all of their trades in a given quarter.
The researchers analyzed data from 1990 to 2023, factoring in fund size, macro indicators, and investor flows. They also found a pattern: managers with a longer trading history and those operating in less competitive categories were easier to mimic.
National Bureau of Economic Research
The upside and the warning
There's a twist that matters for your P&L. Managers with a larger personal ownership stake in the fund were less predictable. And the less predictable managers strongly outperformed, while the most predictable managers significantly underperformed.
Even within a single portfolio, the stocks that were harder to predict outperformed those that were easy to predict. Translation: repeatable, obvious trades are getting competed away. Distinct judgment still pays.
Why this matters for management
The U.S. asset management industry is estimated at roughly $54 trillion. If AI can clone most trade directions, fee pressure, product design, staffing, and process will follow. This isn't a headline to ignore-it's an operating decision.
What to do now
- Measure your predictability. Backtest your trade directions against simple AI/ML baselines. Segment by strategy, market regime, and liquidity. Know where you're obvious.
- Increase "skin in the game." Where policy allows, higher manager ownership correlates with less predictability. Incentives shape behavior.
- Blend discretionary with systematic. Add regime-switch logic, alternative data, and cross-asset signals so models don't map you 1:1. Make your edge conditional, not constant.
- Use AI as a copilot, not an autopilot. Let models surface ideas, detect crowding, and run risk checks. Keep human oversight on sizing, timing, and "why now."
- Cut rote trades. If a move is easy to predict, assume returns are thin. Enforce a brief "edge memo" before adding or scaling any position.
- Revisit your fee and product mix. Highly predictable strategies belong in lower-fee, higher-capacity wrappers. Save premium pricing for truly differentiated processes.
- Upgrade your data and governance. Build repeatable pipelines, feature stores, and audit trails. Institute model risk reviews and clear accountability.
- Invest in people. Pair PMs with ML engineers and data scientists. Upskill analysts on prompt craft, feature engineering, and validation. See: AI Learning Path for Finance Managers
- Track the right KPIs. Maintain a "predictability score" by strategy, hit rate by predictability bucket, alpha decay after crowding, and slippage vs. model.
- Tighten compliance early. Document data rights, model limits, and disclosures. Review best-execution policies when AI influences routing or timing.
What this doesn't say
71% isn't 100%. Markets shift. Regimes break. Models lean on history; black swans don't. Human judgment still sets context, pressure-tests signals, and adapts when structure changes.
Your edge lives where novel information, timing, and portfolio construction meet client needs. Keep that hard to copy. Automate the rest.
Bottom line for leaders
Treat AI as both a rival and a tool. If a model can replicate most of your trades, your fees and differentiation are at risk. Make your process less predictable where it earns, and systematize the repeatable parts to lower cost, improve speed, and free up attention for real edge.
Your membership also unlocks:
