Artificial intelligence has transformed nearly every industry, and asset management is no exception, as the industry depends on rapid and precise information processing to gain a competitive edge and deliver superior performance. Xiaowen Hu, Maximilian Rohrer, and Hanjiang Zhang, authors of the October 2025 study “Active Machine Learning Based Trading and Mutual Fund Performance,” provided evidence of machine learning adoption in the U.S. mutual fund industry over the past two decades. Their findings reveal not just that funds are increasingly using AI, but that those doing it well are generating significant outperformance.
What the Researchers Studied
The research team tackled a challenging question: How can we identify which mutual funds are successfully using machine learning in their investment strategies, and does it improve performance?
To answer this, they developed a novel metric called Active Machine Learning Based Trading. It is an active share measure weighted by ML trading signals, which captures the degree by which a mutual fund aligns its investments with ML-based trading signals—a high AMLT means that the fund is actively engaged in an ML-based trading strategy. Their approach examined:
-
Thousands of actively managed U.S. equity mutual funds over the period 2002-2022.
-
235 different information inputs, including traditional financial metrics, textual data from financial filings, news articles, and earnings call transcripts.
-
Deep neural networks trained to predict stock returns based on this comprehensive information set.
Key Findings
1. ML Adoption Is Rising Rapidly
The study documents a clear upward trend in ML adoption across the mutual fund industry.
-
AMLT is a consistent investment strategy that some mutual funds follow.
-
Funds at the 80th percentile of ML usage more than doubled their adoption over the two-decade period.
-
The percentage of mutual fund employees with core AI skills (including machine learning, natural language processing, and computer vision) rose from 0.99% in the early 2000s to 1.93% by 2022.
2. AI-Powered Funds Delivered Significant Outperformance
-
Funds in the top 10% of ML adoption generated annual risk-adjusted returns of 1.1% to 1.4%.
-
These top performers beat bottom-tier funds by 2.4% to 3.0% per year after adjusting for risk.
-
This outperformance persisted over time, lasting multiple years rather than disappearing quickly.
-
They showed lower fund risk, regardless of the risk measures used (i.e., total, market beta or idiosyncratic fund risk).
-
Funds in all style groups were able to add value by utilizing AMLT strategy
3. Superior Performance Comes from Multiple Sources
The performance advantage comes from:
-
Better stock selection: ML models excel at identifying which stocks will outperform. Funds in the top decile actively deviated from their benchmark to align with ML signals in 23.3% of their portfolios, while those in the bottom decile misaligned by 36.5%.
-
Lower expenses: Interestingly, high-ML funds charge lower fees on average.
-
Efficient trading: Despite ML strategies typically requiring high turnover, successful funds manage trading costs effectively—their annual turnover was 14.6% lower than bottom-tier funds.
-
Performance During High-Volatility Regimes
4. The “Secret Sauce”: Information Processing Power
One of the most fascinating findings is what drives ML’s advantage.
-
Funds using simpler models (like linear regression) capture only about half the performance of those using sophisticated neural networks.
-
The advantage comes from ML’s ability to process vast amounts of diverse information and detect complex, nonlinear patterns.
-
Combining quantitative and textual data is crucial—using either alone produces only a fraction of the full performance.
While traditional quantitative models might analyze 50-100 financial metrics, these ML systems simultaneously process hundreds of variables from financial statements, news sentiment, earnings call tone, regulatory filings, and market data—finding patterns that would be impossible for humans or simpler models to detect.
5. Human Expertise Still Matters
Their research showed that ML success isn’t automatic. The best results came from funds that successfully combined human expertise with AI technology. High-performing funds demonstrated skill in:
-
Selecting stocks with richer information sets where ML has more to work with.
-
Managing the practical challenges of ML-based trading (high turnover, transaction costs, market liquidity constraints).
-
Maintaining performance across different market conditions—volatile and calm, expansionary and recessionary periods.
6. Smaller Funds
Small funds are good at following the AMLT strategy—smaller assets under management allow funds to better manage trading impact in forming ML-based portfolios.
7. Quant Funds
Unsurprisingly, quant funds exhibited higher AMLT measure than an average fund in the same style.
Investor Takeaways
-
Technology investment is becoming table stakes for investment managers. The competitive landscape is shifting. Funds not investing in AI capabilities may find themselves at an increasing disadvantage.
-
Information diversity matters. Simply having data isn’t enough—the research shows that combining traditional financial metrics with alternative data sources (news sentiment, textual analysis of filings) significantly enhances performance.
-
Implementation skill is critical. Having ML models isn’t sufficient; funds need the expertise to manage trading costs, select appropriate stocks, and integrate AI insights with human judgment.
The Bigger Picture
Artificial intelligence is genuinely transforming asset management, and the transformation is accelerating. The firms successfully implementing ML technology are building competitive advantages through superior information processing. However, the finding that human expertise remains essential is equally important. The best outcomes arise from combining AI’s computational power with human judgment, experience, and oversight—what some call “augmented intelligence” rather than artificial intelligence.
Looking Ahead
As AI technology continues to advance and more funds adopt these strategies, several questions remain:
-
Will the advantage persist? As ML adoption becomes more widespread, will the performance edge diminish?
-
What about newer AI technologies? This study focuses on machine learning through 2022. Technologies like large language models (ChatGPT and similar) have emerged since then—how will they impact fund performance?
-
How much capacity exists? If too much capital chases ML-driven strategies, market impact and crowding could erode returns—Andrew Lo’s Adaptive Markets Hypothesis.
Conclusion
This research advances our understanding of how artificial intelligence is reshaping investment management. For investors, the key takeaway is clear: machine learning isn’t just a buzzword in asset management—it’s a powerful tool that, when implemented skillfully, can generate meaningful outperformance.
The future of fund management likely belongs to funds that successfully harness AI’s information-processing capabilities while maintaining the human expertise needed to implement these strategies effectively. As an investor, understanding which funds are genuinely leveraging these technologies—versus merely marketing AI credentials—will become increasingly important.
The revolution in AI-powered investing isn’t coming. According to this research, it’s already here.