Maybe AI models will be a reliable source of income for traders in the future. That day, however, has most certainly not come.
In reality, a recent research by DayTrading.com titled “AI Trading Error Rates: Accuracy, Risks, and Reliability” suggests otherwise. According to the study, top AI models fall short in several ways. For instance, they frequently have hallucinations and sometimes report “company financials that never existed.” Frequently, they provided “price data that was days or weeks old.”
In addition, the study found that during the time of August, none of the six top AI tools it looked at could “beat a good human swing trader.” On average, the six generated an 8.4% loss, while the S&P 500’s total-return metric saw a 2% gain.
Six prominent AI models among traders were assessed in the study. A series of factual questions were posed to each, and they were also asked to describe various financial principles, recognize distinct chart patterns, and suggest certain transactions. The models’ answers were scored based on how well they recognized risk and uncertainty in addition to their accuracy.
The study assigned a “danger index” to each of the AI models, which represented the overall risk a trader would encounter if he relied on the model without first carefully examining its accuracy and risk management. A higher score indicated a trader was in greater risk, as seen in the chart above that shows these models’ various danger indices. An alarmingly high 42.8 was the average for all six models.
The study discovered that the confidence of AI models, even when they are wholly incorrect, is one of their main weaknesses. The issue is that AI is a very powerful storyteller in addition to being a tool. Traders are also predisposed to trust stories, particularly when they are presented with polished charts and assurance-inspiring justifications. According to the results of the research, even losing AIs seemed confident about themselves.
Trading models using AI are not acceptable
These results shouldn’t come as a surprise because they align with earlier research on AI in the financial space. The success of the Eurekahedge AI Hedge Fund Index, which was created to offer a comprehensive gauge of the performance of underlying hedge-fund managers that apply machine learning and artificial intelligence to their trading procedures, was one notable outcome. When this index ended in January 2025, it had fallen well short of the overall return of the S&P 500, by 8 annualized percentage points for the trailing 10-year return and 11 annualized percentage points for the following five-year return.
Or look at the results of exchange-traded funds that use AI to choose equities on the American stock exchange. As determined by NerdWallet.com, the seven ETFs in this category are included in the above table. These include WisdomTree US AI Enhanced Value Fund and Amplify AI Powered Equity ETF. A wide variety of returns will be visible to you, some well ahead of the market and others considerably behind. A basic S&P 500 index fund outperforms the average AI-powered ETF across the following three and five-year periods.
And the bottom line? AI has the potential to be a very useful tool for traders. However, it should not be mindlessly trusted or obeyed because it is still only a tool.
