Ask Claude how well it can pick stocks, and it will tell you (or at least it told us), “Honestly, not very.” That turns out to be true.
What happened: A slew of experiments to test the stock-picking ability of AI models found that even the best ones still usually lose money.
One contest, Alpha Arena, gave each of the frontier AI systems $10,000 and two weeks to make as much money as they could trading tech stocks. The portfolio lost a third of its value and 32 out of the 38 models tested lost money.
Why it’s happening: It’s not that surprising that off-the-shelf AI tools struggle with picking stocks. The professionals who do it successfully (and many don’t) depend on access to real-time, and sometimes proprietary, data that gives them an edge.
AI models, on the other hand, are trained on months-old data and are limited to searching the web for information that any Joe Schmo could find.
They’re also prone to the same mistakes humans make, per Bloomberg: “[AI systems] tend to mistime their trades, incorrectly size positions and buy and sell too often.”
Why it matters: 14% of Canadian investors have used AI to get investment advice, according to a Scotia Wealth Management survey, though only 7% say (or are willing to admit) they have made investment decisions solely on the basis of AI recommendations.
A study from last month found that interacting with AI chatbots tends to increase people’s confidence in their own investment ideas and encourage more stock trading — a characteristic that’s not likely to help people maximize their returns.
Yes, but: None of this is to say that AI can’t be helpful in some aspects of the investing process, like company research — at least that’s what the world’s biggest investment banks seem to think.
