Last updated 17 July 2026. Every performance figure below was verified against Amplify’s SEC shareholder report for AIEQ and the fund’s own live performance page — not against secondary coverage, which turned out to be inconsistent. Sources and capture dates are linked inline.
Yes, AI can predict the stock market — at least as well as most human analysts, and there is peer-reviewed evidence for it. It still isn’t enough. The longest-running AI-driven equity fund available to ordinary investors turned $10,000 into $20,732 between October 2017 and September 2025. The S&P 500, dividends reinvested, made $29,829 over the same stretch. The fund’s 0.75% fee explains about 15% of that gap. The other 85% is the strategy itself. Forecasting is the part the machine is good at. Turning forecasts into a portfolio is where the money went.
Can AI predict the stock market better than a human analyst?
On the narrow question, the answer is yes, and it is not close to controversial. In From Man vs. Machine to Man + Machine: The Art and AI of Stock Analyses (Journal of Financial Economics, vol. 160, 2024), Cao, Jiang, Wang and Yang trained an AI analyst on corporate disclosures, industry trends and macroeconomic indicators. It beat most human analysts at predicting stock returns.
The paper is more interesting than the headline, though. Humans held their edge where institutional knowledge mattered — intangible assets, companies in financial distress. The machine won where information was transparent but voluminous, which is exactly what you would expect: it is a reading-speed advantage, not an insight advantage. And the combination of the two beat either alone, mostly by cutting the extreme errors.
So the interesting question isn’t whether AI can forecast. It’s why that forecasting skill has been so hard to convert into money.
What happened when an AI actually ran a fund for eight years
The AI Powered Equity ETF (AIEQ) launched on 17 October 2017. It runs on a quantitative model built by EquBot on the IBM Watson platform, screening thousands of U.S. companies across news, social media, analyst reports and financial statements. It is not a thought experiment. It is real money, real fees, real execution, reported to the SEC.
Here is what its own annual shareholder report says, for the year ending 30 September 2025:
| Annual average total return (NAV) | 1 year | 5 year | Since inception (10/17/2017) |
|---|---|---|---|
| AIEQ (the AI fund) | 19.68% | 8.71% | 9.60% |
| S&P 500 Total Return | 17.60% | 16.47% | 14.73% |
Source: Amplify AI Powered Equity ETF Annual Shareholder Report, period 1 October 2024 – 30 September 2025. Captured 17 July 2026.
Run the compounding over the 7.95 years from inception to the report date:
- $10,000 in AIEQ at 9.60% a year → $20,732
- $10,000 in the S&P 500 at 14.73% a year → $29,829
- Gap: $9,097
The AI delivered 69.5% of what the index delivered. Not 69.5% of the gain — 69.5% of the ending balance. An investor who wanted exposure to the smartest available stock-picking machine, and got it, ended up with roughly two-thirds of what someone who bought a plain index fund and ignored it entirely walked away with.
The five-year window is worse, not better: 8.71% against 16.47%, a shortfall of 7.76 points a year. On $10,000 that is $15,183 against $21,432.
(One honest caveat on the comparison: the S&P 500 Total Return is an index, and you cannot buy an index directly. A real index fund charges a few basis points, so the investable version of that $29,829 is slightly lower. It does not move the conclusion.)
Why the fee isn’t the explanation
The reflexive answer is “well, it charges 0.75% and the index fund charges nothing.” That answer is wrong, and the arithmetic says so plainly.
The since-inception shortfall is 5.13 percentage points a year. The expense ratio is 0.75%. So the fee accounts for 14.6% of the gap. The remaining 85.4% is the model’s own decisions.
This matters because it kills the easy fix. If fees were the story, a cheaper AI fund would solve it. They aren’t, so it wouldn’t.
What is in the other 85%? The report gives a strong hint on the same page. Portfolio turnover: 804%.
Turnover of 804% means the fund replaced its entire portfolio roughly eight times over in a single year. A 161-stock portfolio, rebuilt eight times. Every one of those trades crosses a bid-ask spread and pays commissions — and neither of those costs appears in the 0.75% expense ratio. They come straight out of returns, silently. Amplify says as much in its own disclaimer: brokerage commissions will reduce returns.
This is the mechanism, and it is worth stating plainly because it generalises well beyond this one fund: a forecast has to be traded to become a return, and trading costs money. An edge of half a percent per prediction is real, publishable, academically defensible — and completely consumed by the spread if you act on it eight times a year. The model was never the bottleneck. The transaction was.
How the same fund can “beat the S&P 500” and lose by $9,097
This is the part worth learning, because you will see the trick again everywhere.
AIEQ’s own annual report leads its performance discussion by saying the fund produced positive returns that exceeded the S&P 500 Index. That statement is true. Over the twelve months to 30 September 2025, AIEQ returned 19.68% against the index’s 17.60% — a genuine win by 2.08 points.
It appears on the same page as the table showing 9.60% against 14.73% since inception.
Nothing about that is fraudulent. Every number is accurate and audited. It is window selection, and it is completely legal and nearly universal. The fund did not choose eight bad years and one good one on purpose; it chose which one to put in the first paragraph.
The defence is simple and mechanical. When a fund quotes a return, ask what window, and then ask for the since-inception number. The since-inception figure is the only one nobody gets to pick. It starts on the day the strategy started, and it includes every quarter the strategy would rather you forgot. If it isn’t offered, that itself is the answer.
What happens when the AI claim isn’t even real
There is a floor below underperformance, and the regulator has already found firms standing on it.
On 18 March 2024, the SEC settled its first “AI washing” enforcement actions against two investment advisers. Delphia (USA) Inc. had advertised that it used AI on client data to predict which companies were about to make it big and invest ahead of everyone else. Under examination, Delphia admitted it had not used any client data and had not built the algorithm at all. Global Predictions Inc. claimed to be the “first regulated AI financial advisor” and could not produce a single document to substantiate it. They paid $225,000 and $175,000 respectively.
The useful takeaway isn’t that some firms lie. It’s the ordering. AIEQ is the honest case — a real model, really running, really reporting — and it still lost to the index by five points a year. Set your expectations for the marketing material accordingly.
Where this argument breaks down
Every claim on this site names the conditions under which it fails, and this one has real ones.
The recent record is genuinely decent. AIEQ’s trailing three-year return was 16.86% annualised, and it beat the index outright in FY2025. Anyone who bought in 2023 has no complaint. If you only look at the last three years, this entire article looks like sour grapes.
One fund is one fund. AIEQ is a single strategy from a single provider. It is the longest live public track record of its kind, which is why it is worth this much attention, but n=1 is n=1 and I am not going to pretend otherwise.
The model has changed. Eight years is long enough that the AIEQ of 2018 and the AIEQ of 2026 are not the same system. Judging today’s model by 2019’s results is its own kind of window error, just pointed the other way.
Forecasting genuinely is improving. The Cao et al. result is real. Nothing here says a future model can’t clear its own trading costs. What the record says is that this one hasn’t, over eight years, with real money — and that the gap between “the model is good” and “the fund made money” is much wider than the sales page suggests.
So what should you actually do?
The honest recommendation, and its cost.
Use AI as an analyst, not an allocator. This is what the evidence supports on both ends: the machine reads faster than you and it will find things you missed, which is Cao et al.’s finding and it holds up. What it has not demonstrated is the ability to convert that reading advantage into a return after costs. Let it summarise the 10-K. Don’t let it size the position.
The cost of that advice: it’s boring, and it means the answer to “how do I use AI to beat the market” is mostly “you probably don’t, and the index is right there.” If you were hoping for a tool recommendation, this isn’t one. For most people the unglamorous route in investing for beginners will beat any of this, and the reason is arithmetic rather than pessimism.
If you do want to use an AI fund anyway — and there are defensible reasons, including that AIEQ’s max drawdown history differs from the index’s — then size it as the speculative sleeve it is, and check the since-inception number every year against the benchmark. Not the one on the sales page. The one in the annual report.
The related question of whether these tools replace the human in the loop entirely is covered separately in will AI replace financial advisors, which reaches a similar place from the other direction: the machine is good at the part that was never the hard part.
Frequently asked questions
Can AI predict the stock market accurately?
It can forecast returns better than most human analysts, per the 2024 Journal of Financial Economics study. “Accurately” is the wrong frame, though — a forecast only has to be right often enough to overcome what it costs to act on it, and that is the test the live funds have been failing.
Has any AI fund beaten the S&P 500?
Over short windows, repeatedly — AIEQ beat it by 2.08 points in the year to September 2025. Since its 2017 inception, it has returned 9.60% a year against the index’s 14.73%.
Why do AI trading tools underperform if the models are good?
Because prediction and portfolio construction are different jobs. AIEQ’s 804% annual turnover means it pays spreads and commissions eight times over on the whole book — costs that never appear in its 0.75% expense ratio and come directly out of returns.
Is “AI-powered” investing marketing just a label?
Sometimes literally. The SEC’s first AI-washing settlements in March 2024 involved a firm that had advertised an AI algorithm it had never built.
This article is for informational purposes only and is not investment advice. I am not a licensed financial advisor. AIEQ is named here as the longest-running public example of an AI-driven equity strategy, not as a recommendation to buy or avoid it. Past performance does not predict future results. Do your own research and consider speaking to a licensed professional about your own situation.
Figures verified 17 July 2026 against Amplify’s AIEQ annual shareholder report (FY ending 30 September 2025) and the fund’s live performance page. Fund performance data changes; the since-inception and trailing figures above will be restated in Amplify’s next annual report, expected around December 2026.
