ChatGPT's $100 Stock Portfolio Gains 25% in First Month

An ambitious experiment by a Redditor has seen OpenAI’s advanced AI model, GPT-4o, generate notable returns in its first month of stock trading with a modest $100 investment.

Nathan Smith, the Redditor behind the project, described it as a “6-month experiment to see how a language model performs in picking small, under-covered stocks with only a $100 budget.” Using GPT-4o, the AI-managed portfolio increased in value by 25 percent over its initial month, turning the $100 into $125. This performance notably outpaced two key small-cap stock indexes, the Russell 2000 and XBI, which saw gains of approximately 3.9% and 3.5% respectively during the same period. For context, the broader S&P 500 index rose by less than 3 percent, suggesting ChatGPT made effective selections within its niche.

The concept of AI-driven stock selection is not entirely new. In December, researchers from Germany’s Duisburg–Essen University published findings in Finance Research Letters indicating that advanced OpenAI models appeared capable of selecting profitable stocks. However, expert opinions vary. Alejandro Lopez-Lira, an assistant finance professor at the University of Florida, stated in a June interview with Morningstar that years of simulating AI stock selection yielded less impressive results than what might be observed with larger, real-world investments. “Our results on paper are much more optimistic than what the performance in reality would be with a reasonable investment size,” Lopez-Lira commented.

Furthermore, a common logical challenge with AI stock picking is that widespread adoption of a truly superior AI model would fundamentally alter market dynamics, potentially negating the very advantages it initially offered.

Smith’s motivation for undertaking this project stemmed from his skepticism towards pervasive online advertisements promoting AI’s supposed ability to identify undervalued stocks and generate significant profits. He decided to test the premise himself.

Crucially, the experiment is not entirely autonomous. Smith provides ChatGPT with daily trading data for its portfolio and implements strict “stop-loss” rules, which mandate that the AI sell a stock once it reaches a predetermined price to limit potential losses. Smith himself acknowledges that this daily human input, committed until the end of December, means the AI is not yet “managing money without guidance,” despite the experiment’s initial stated purpose.

While the initial returns are compelling, particularly during a generally positive market month, the true test of the AI’s long-term stock-picking prowess will come as the experiment progresses and faces various market conditions. The ultimate question remains whether the bot’s portfolio will conclude its six-month run in positive territory.