At some point, artificial intelligence (AI) will make it unnecessary for households to pick stocks. And that is a good thing, because people are awful at it.
Dalbar’s 2025 Quantitative Analysis of Investor Behavior (QAIB) Report is out, and it is another doozy. Since 1994, Dalbar has measured the effects of investors’ decisions on buying, selling, and the timing of mutual fund transactions. I compared that report to Hendrick Bessembinder’s publication, “Which U.S. Stocks Generated the Highest Long-Term Returns?,” and, spoiler alert, investors (professional and amateur alike) would have been better served by sitting tight in low-cost index funds than buying and selling individual stocks.
No, really, that was a genuine spoiler alert, because many people won’t believe that pertains to them. However, thousands of people have told me that they consistently beat the stock market, despite all the data to the contrary. It is similar to when you ask people if they are above-average drivers and nine out of 10 people say they are. (If you ask them, something like nine out of 10 people will tell you that they are better than average at things like being outgoing, communicating, cooking, parenting, avoiding drama, giving advice, their job, or having better-than-average common sense.) People aren’t as smart as they believe they are. But is AI better at some of those things than humans? More specifically, can AI pick stocks better than the rest of us? Probably. After all, like I said, humans are awful at it, no matter how good they think they are.
Experiments with AI-driven stock picking have produced some eye-opening results, but not always. For instance, earlier this year, a project by Finder.com tested ChatGPT’s ability to select stocks. Surprisingly, its portfolio gained 4.9 percent over a period where the average human-led investment fund lost 0.8 percent. Similarly, an experiment conducted by a high school student using ChatGPT to invest in small-cap stocks generated returns of 23 percent compared to the Russell 2000’s 3.9 percent increase.
On the surface, these experiments suggest AI might be onto something. But let’s pause and take a deeper look.
First, let’s consider the allure of stock picking. It is exciting, it is appealing, and it is part of human nature to believe we can beat the market. Yet the reality, backed by decades of research, is that picking individual stocks is far more challenging than most people appreciate.
According to research by Hendrik Bessembinder at Arizona State University, analyzing nearly a century of stock market data from 1926 to 2023, more than half (51.6 percent) of the 29,078 publicly listed common stocks had negative cumulative returns, including dividends, over the time they were in the Center for Research in Security Prices (CRSP) database.
Seventeen of those 29,079 stocks delivered returns greater than 5 million percent. Altria Group led those 17 stocks with an unimaginable 265 million percent cumulative return (the S&P 500’s return has been about 10 times higher than Altria’s over the last decade). However, even more tellingly, the median return across all the other thousands of stocks was a negative 7.41 percent. Negative. Not even a minor gain.
Bessembinder also measured the returns of over 64,000 global stocks from 1991 to 2000 and found that 55.2 percent of U.S. stocks and 57.4 percent of non-U.S. stocks underperformed one-month Treasury bills. During that period, a mere 2.4 percent of stocks accounted for 100 percent of the global stock market value creation.
Now, juxtapose that against AI-generated portfolios. AI strategies generally tend to follow established patterns. A recent experiment, known as the “Comet Portfolio,” quickly constructed a conventional, tech-heavy portfolio featuring stocks like Amazon, Nvidia, and Microsoft. In other words, AI may simply be aggregating and reflecting prevailing market wisdom rather than generating innovative insights, which isn’t necessarily bad; it is essentially momentum investing. And that can be a good thing if it mitigates two common human decision-making factors: fear and greed.
Moreover, Dalbar’s “Quantitative Analysis of Investor Behavior” (QAIB) 2025 Report adds another critical dimension: human psychology. The average equity fund investor earned 8.5 percent less than the S&P 500 in 2024. The QAIB study identifies human psychology as the culprit behind that investor gap. A typical example is that investors might choose to raise cash when the market is down.
Dalbar’s report highlights that the biggest enemy of investors isn’t necessarily poor stock selection; it is their own psychological biases. Loss aversion, regret, media responses, and herd behavior consistently sabotage investor returns.
Can AI truly pick stocks better than humans?
Could AI counteract these emotional and psychological human pitfalls? Perhaps. But even AI is only as good as the instructions provided, and if those instructions mimic human biases, the result is unlikely to outperform the market in the long term. And let’s not overlook the obvious: Even if the AI trader is the best in the world, humans might still “fire” the AI trader when things are “going wrong.” And, if you define “going wrong” as price corrections, things always go wrong in the market.
AI’s capabilities are undoubtedly impressive (at times) when it comes to trading stocks. But its achievements thus far largely reflect well-trodden paths and conventional wisdom. AI-driven experiments have delivered interesting but ultimately short-term results. However, not all AI trading goes well at all. A recent academic study found that timing-based investment strategies using large language models (LLMs) underperform passive benchmarks. The models tend to be too conservative in bull markets and too aggressive in bear markets, resulting in substantial losses. However, I suspect it is only a matter of months, not years, until AI can protect investors from losses.
AI has not yet provided evidence that it can reliably outperform the wisdom of buying the entire haystack rather than searching for individual needles. And neither have humans.
Professional stock pickers and AI haven’t consistently proven superior to simple index strategies. So, if your broker insists they can outperform by picking individual stocks and charges a hefty fee for the privilege, it may be worth asking whether they are selling investment prowess or just hope and hype. And, perhaps, in the future, you can fire your stock pickers and rely on AI. However, until it demonstrates consistent long-term market-beating abilities, investors would be wise to temper their enthusiasm for AI investment advice with healthy skepticism.
Bonus: Key insights and statistics from Dalbar’s 2025 QAIB study
- Asset Allocation: Average asset allocation was 74.45 percent equities and 25.25 percent bonds by the end of 2024, reflecting a slight shift toward risk assets compared to the previous year (73.72 percent equities at the start of 2024).
- Guess Right Ratio (Market Timing Success): In 2024, investors guessed the market direction correctly only 25 percent of the time, tying 2019 for the lowest success rate recorded.
- Historical Context: Historically, 73 percent of negative S&P years rebounded immediately with positive returns the following year. Sixty-four percent of negative years were followed by gains exceeding 20 percent.
- Behavioral Highlights: Investor returns consistently lag due to emotional trading behaviors, particularly selling at market lows and chasing returns after rallies. Common behavioral pitfalls identified include loss aversion, anchoring, regret, herding, and excessive optimism.
Allen Harris is an owner of Berkshire Money Management in Great Barrington and Dalton, managing more than $1 billion of investments. Unless specifically identified as original research or data gathering, some or all of the data cited is attributable to third-party sources. Unless stated otherwise, any mention of specific securities or investments is for illustrative purposes only. Advisor’s clients may or may not hold the securities discussed in their portfolios. Advisor makes no representation that any of the securities discussed have been or will be profitable. Full disclosures here. Direct inquiries to Allen at AHarris@BerkshireMM.com.
