The Growth and Performance of Artificial Intelligence in Asset Management
Open accessThe DOI matches its Crossref record; 58 of the 58 references it cites were found in a registry or an index; 9 of the 30 sources its claims cite were checked for support.
Abstract
This paper examines AI adoption in asset management and its investment implications. We document that AI-driven investing is concentrated among hedge funds, particularly those employing macro strategies. AI funds exhibit greater alpha comovement and are launched by investment advisers facing stronger performance incentives. These funds significantly outperformed non-AI hedge funds on a risk-adjusted basis, but their outperformance declined over time and disappeared with the growth of AI-driven investing. Nevertheless, AI funds continued to outperform sibling funds managed by the same advisers. Our findings highlight both the alpha-generating potential and the limitations of AI as a source of investment performance.
Overview
CoChat Summary
TL;DR
Using regulatory disclosures, job postings, and fund strategy descriptions, the paper documents that AI-driven investing has grown since the early 2010s and is concentrated among hedge funds, especially systematic diversified macro strategies. AI hedge funds outperformed non-AI funds in early years, but that outperformance declined and became statistically indistinguishable from zero after 2017.
Key findings
- AI-driven investing is a small but growing niche: among 7,896 U.S. hedge funds from 2006–2024, 89 funds (1.1%) are ever identified as AI funds; AI fund-months are about 1% of the sample, versus 23% for quant funds.
- AI hedge funds outperformed non-AI funds by about 6% annually on a benchmark-adjusted basis in earlier years, but the outperformance declined by 5.9 basis points per year and disappeared after 2017; before 2017, AI funds outperformed by 49.6 basis points per month using Fung-Hsieh alpha.
- Within the same adviser, AI funds outperformed sibling funds by 34.9 basis points per month (FH alpha) and 41.1 basis points per month (AQR alpha), indicating results are not solely driven by adviser-level heterogeneity.
- Contrary to homogeneity concerns, AI fund pairs had lower average pairwise alpha correlation than non-AI pairs: 0.021 versus 0.124 using FH alpha, a difference of -0.103; AI funds also did not receive inflows beyond what realized performance would predict.
Methods
- The authors combine SEC Form ADV Part 2A filings (116,276 brochures, 62,253 adviser-years, 2012–2024), Lightcast job postings (2010–2024), and archived HFR hedge fund snapshots (2005–2024) with fund strategy descriptions.
- They use OpenAI’s GPT-5 to classify AI usage in adviser disclosures and fund strategy descriptions, defining AI-driven investing as AI-Autonomous or AI-Augmented strategies; they also identify quant funds.
- Performance tests regress next-month benchmark-adjusted fund returns on prior-year-end AI classification, controls, and strategy-by-month fixed effects; sibling-fund tests add adviser-by-month fixed effects. Alphas are estimated using Fung-Hsieh and AQR factor models.
Limitations
- The paper states that text-based AI classifications capture disclosed AI adoption, so some advisers and funds using AI/ML may not be identified, potentially attenuating estimated AI fund performance.
- The paper states that labor-data tests use a matched adviser sample with limited coverage of smaller, specialized firms, and that HFR covers only a modest share of hedge fund assets reported in Form ADV.
- The paper states that the number of identified AI hedge funds is relatively small and AI technologies evolve rapidly, so extrapolating estimates to future periods warrants caution; the conservative classification also prioritizes precision over coverage, implying false negatives.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
5 Valid · 4 Not valid · 15 Not checked · 211 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“Overall, these comparisons show that survivorship bias is unlikely to materially affect our findings.”
What the sources say
- Disputedcites Baquero, Ter Horst, and Verbeek, 2005 matched to “Survival, Look-Ahead Bias, and Persistence in Hedge Fund Performance” (2005) sourceExperimental — accuracy not yet measured
The Introduction says fund survival can bias performance analysis. The Concluding remarks describe a typical 2% annual effect associated with survivorship bias and stress the importance of correcting for it. The paper therefore does not support treating survivorship bias as immaterial; it also cannot establish the effect on the citing authors’ particular findings. (LLM, conf=0.78)
Checked by a reasoning model — may be wrong.
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Disputed
Every source the paper cites for this claim was checked and supports it.
In the paper
“Offering shorter lockup and restriction periods imposes implicit incentives to deliver better performance (Agarwal, Daniel, and Naik, 2009).”
What the sources say
- Supportedcites Agarwal, Daniel, and Naik, 2009 matched to “Role of Managerial Incentives and Discretion in Hedge Fund Performance” (2009) source
The discussion accompanying Table II (p. 19) explicitly says shorter lockup and restriction periods provide implicit incentives to perform better because poor performance can prompt capital withdrawal. The claim describes that incentive mechanism, not the paper’s finding about the net effect of period length on returns. (LLM, conf=0.94)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Because firms rarely disclose their AI-specific capital expenditures, the literature often measures firm-level AI adoption using labor market data.”
What the sources say
- Partly supportedcites Acemoglu, Autor, Hazell, and Restrepo (2022) matched to “Artificial Intelligence and Jobs: Evidence from Online Vacancies” (2022) sourceExperimental — accuracy not yet measured
The Introduction (page not provided) says the paper identifies AI adoption from AI-related hiring and online vacancy postings, supporting the use of labor-market data as a measure. It does not say that firms rarely disclose AI-specific capital expenditures or give that as the reason for using those data; nor does it establish that the literature often uses this approach. (LLM, conf=0.86)
- Not matchedcites Babina, Fedyk, He, and Hodson (2024a)
More than one paper fits this citation.
- Partly supportedcites Eisfeldt, Schubert, Taska, and Zhang (2026) matched to “Generative AI and Firm Values” (2026) sourceExperimental — accuracy not yet measured
Section I, Data and Measurement, uses occupational employment data to construct a firm-level measure of exposure to generative AI. It does not establish that the literature often measures AI adoption this way, or that firms' failure to disclose AI-specific capital expenditures is the reason. Exposure to AI is also not the same as adoption. (LLM, conf=0.86)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Although our sample includes both live and dead funds, the requirement of at least 12 months of returns for alpha estimation could introduce an implicit survivorship bias (Baquero, Ter Horst, and Verbeek, 2005).”
What the sources say
- Partly supportedcites Baquero, Ter Horst, and Verbeek, 2005 matched to “Survival, Look-Ahead Bias, and Persistence in Hedge Fund Performance” (2005) sourceExperimental — accuracy not yet measured
The Introduction (p. not available) says that conditioning on a fund’s survival over consecutive periods can bias performance analysis, even when returns from funds that later disappear are included. Section 4 describes a one-year return-history requirement that makes rankings conditional on survival. The paper calls this look-ahead bias, however, and does not examine a 12-month requirement for alpha estimation specifically. (LLM, conf=0.84)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“They also suggest that mentioning AI in strategy descriptions per se does not attract investor flows, whereas the strong historical performance delivered by AI funds does.”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
122 more claims
Create a free account to see all 235 claims and every source verdict.
- It may state two separate claims as one. 56
- Probably not a substantial claim. 39
- The quote only partly states this claim. 22
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ 5.4. Money flows to AI hedge funds · p. 28
“measured with a lagged 12-month rolling-window average FH alpha.”
§ Table 8: Comovement of Hedge Fund Performance · p. 51
“Performance is lagged 12-month rolling-window cumulative alpha based on Fung-Hsieh seven hedge fund factors.”
Citations
Cites
Mutual Fund Flows and Performance in Rational Markets
Berk, J. B., & Green, R. C. (2004) · Journal of Political Economy · doi:10.1086/424739 · Cited by 2,083
The Adaptive Markets Hypothesis
Lo, A. W. (2004) · The Journal of Portfolio Management · doi:10.3905/jpm.2004.442611 · Cited by 1,670
Artificial intelligence, firm growth, and product innovation
Babina, T., Fedyk, A., He, A., & Hodson, J. (2024) · Journal of Financial Economics · doi:10.1016/j.jfineco.2023.103745 · Cited by 1,460
Does Fund Size Erode Mutual Fund Performance? The Role of Liquidity and Organization
Chen, J., Hong, H., Huang, M., & Kubik, J. D. (2004) · American Economic Review · doi:10.1257/0002828043052277 · Cited by 1,430
Artificial Intelligence and Jobs: Evidence from Online Vacancies
Acemoglu, D., Autor, D., Hazell, J., & Restrepo, P. (2022) · Journal of Labor Economics · doi:10.1086/718327 · Cited by 1,193
The Performance of Hedge Funds: Risk, Return, and Incentives
Ackermann, C., McEnally, R., & Ravenscraft, D. (1999) · The Journal of Finance · doi:10.1111/0022-1082.00129 · Cited by 1,101
Hedge Fund Benchmarks: A Risk-Based Approach
Fung, W., & Hsieh, D. A. (2004) · Financial Analysts Journal · doi:10.2469/faj.v60.n5.2657 · Cited by 1,031
Predictably Unequal? The Effects of Machine Learning on Credit Markets
FUSTER, A., GOLDSMITH‐PINKHAM, P., RAMADORAI, T., & WALTHER, A. (2022) · The Journal of Finance · doi:10.1111/jofi.13090 · Cited by 872
Measuring skill in the mutual fund industry
Berk, J. B., & van Binsbergen, J. H. (2015) · Journal of Financial Economics · doi:10.1016/j.jfineco.2015.05.002 · Cited by 820
Predictably Unequal? The Effects of Machine Learning on Credit Markets
Fuster, A., Goldsmith-Pinkham, P. S., Ramadorai, T., & Walther, A. (2017) · doi:10.2139/ssrn.3072038 · Cited by 769
Research works printed but not linked: 1
- Ilmanen, A., R. Israel, R. Lee, T. J. Moskowitz, and A. Thapar (2021). How do factor premia vary over time? a century of evidence. Journal of Investment Management 19 (4), 15-57.
Metadata
- Authors
- Chen, S., Sialm, C., & Xu, D.
- Affiliations
- The University of Melbourne (AU)
- National Bureau of Economic Research (US)
- Southern Methodist University (US)
- Corresponding author
- Xu, David
- Venue
- National Bureau of Economic Research
- Publisher
- National Bureau of Economic Research
- Published
- 2026-05
- OpenAlex
- W4415673022
- Semantic Scholar
- c35fc39d6374f14bc360dd98b94efde80d3d5fa5
- Cited by
- 0
- References
- 36
- Type
- report
- Language
- en
- Open access
- green
- doi.org
- Locations
- National Bureau of Economic Research (publishedVersion)
- SSRN Electronic Journal (acceptedVersion)
- SMU Scholar (Southern Methodist University) (submittedVersion)
- Topics
- Social Sciences › Economics, Econometrics and Finance › Finance › Financial Markets and Investment Strategies
- Social Sciences › Business, Management and Accounting › Management Information Systems › FinTech, Crowdfunding, Digital Finance
- Social Sciences › Decision Sciences › Management Science and Operations Research › Stock Market Forecasting Methods
- Keywords
- artificial intelligence, asset management, hedge funds, macrostrategies, investment advisers, performance-based incentives, risk-adjusted performance, fund performance persistence
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.