Can AI tools beat a buy-and-hold strategy? A new study says investors chasing an edge won’t find one using a bot

Researchers from the University of Edinburgh, the University of California, Los Angeles (UCLA) and South Korea’s Sungkyunkwan University put a range of AI trading strategies through a backtest, a method that runs a trading strategy through years of real historical market data to see how it would have performed. They tested more than 20 years of data across more than 100 stocks.

In a paper not yet peer-reviewed, they found that most of those AI strategies still couldn’t beat buy-and-hold investing. The team plans to present the findings at the 2026 ACM SIGKDD Conference on Knowledge Discovery and Data Mining in August.

Why a short test made the bots look brilliant

Most earlier AI trading studies were run on a small group of winning stocks over a short period — sometimes only three months on as few as three stocks. Test any strategy on a single standout stock during its best stretch, and almost anything can look brilliant.

That kind of setup creates three common traps:

• Survivorship bias: The analysis only counts stocks that performed well, leaving out the ones that failed or were delisted
• Look-ahead bias: The model uses information that wouldn’t have been available at the time the trade was made
• Data-snooping: Running enough tests on the same data can make a strategy look like it’s working purely by chance

Mihai Cucuringu, a mathematics professor at UCLA and the University of Oxford, and one of the paper’s authors, says his team ran a much broader test, according to Moneywise. They included the 2008 financial crisis, the COVID-19 crash and the bull markets in between, and added delisted stocks back into the data so failures still counted. Cucuringu says the edge AI tools appeared to have in earlier research largely disappears once you test them over a longer period and a wider range of stocks.

Get a brokerage account that fits your needs*.* You can't control the market, but you can control fees, tools and how you invest. Find an online investing platform that helps you invest with confidence. Compare Canada's Best Brokerages

What 20 years and 100-plus stocks showed

Plain buy-and-hold turned out to be one of the strongest strategies in the study. When tested on steadier, less volatile stocks, the method delivered the best results once risk was taken into account — an annual return of about 7.9%, beating every AI tool tested.

The AI tools weren’t completely hopeless. In one test, an AI tool posted the highest raw annual return in the group at nearly 14%. But it also came with wild swings in value along the way, exactly the kind of risk a big headline return can hide.

A 55-year-old statistical model called ARIMA also beat the AI tools once risk was factored in. The two AI trading agents at the centre of the study — both built on commercially available large language models — showed no reliable edge over the market when tested fairly, and bigger models didn’t reliably beat smaller ones. “It’s a big misconception that better models automatically translate into better trading performance,” Cucuringu says.

The bots got the timing backward

The AI tools also got their timing wrong. They made cautious trades when markets were rising, missing out on gains, and aggressive trades when markets were falling, taking bigger losses. On a standard measure of return versus risk, buy-and-hold scored 0.61 in strong years and -0.28 in down years.

One AI tool managed -0.19 in up years and -0.97 in down years — a weak result even when markets were doing well. That same tool also traded far more often, and every extra trade comes with a cost that eats away at returns.

What this means for a Canadian portfolio

The lesson isn’t that AI is useless; it’s that using AI to beat the market by picking individual stocks isn’t its best use. In this study, the simplest approach — buying and holding — outperformed the most advanced computer tools over a span of 20 years.

For Canadians, “buy-and-hold” doesn’t require picking exotic U.S. tech stocks like the ones AI tools tend to chase. As of June 30, BlackRock reported a broad, low-cost Canadian index fund tracking the S&P/TSX Capped Composite Index has delivered an average annual return of roughly 8.62% since its 2001 inception, though returns over shorter recent periods have been considerably higher. Held inside a tax-sheltered account such as a Registered Retirement Savings Plan (RRSP) or a Tax-Free Savings Account (TFSA), that kind of steady, unglamorous exposure to the whole market has historically done more for a portfolio than chasing a hot tip — from a chatbot or anyone else.