A teenager from rural Oklahoma launched an experiment by handing over control of a small investment portfolio to ChatGPT and ended up outperforming the stock market. Over four weeks, the portfolio delivered a return of 23,8%, while the Russell 2000 index rose just 3,9% and the biotech ETF XBI gained 3,5%.
How the Algorithm Works
Seventeen-year-old Nathan Smith began with the idea of letting a language model fully manage $100 by investing in U.S. micro-cap stocks with market capitalizations under $300 million. The AI selects the assets, calculates position sizes, sets risk levels, and sells shares when needed. The only thing required from the human is to execute the trades and record the results.
The main feature of the system is its full autonomy. ChatGPT independently manages the investments without human interference - except in cases where the model contradicts itself, in which case Smith steps in.
Model on GitHub and Key Metrics
The experiment’s code is available on GitHub. ChatGPT updates the portfolio once a week, staying within the set market cap range. For performance tracking, Smith uses data from Yahoo Finance, processes it with Pandas, and visualizes the results on charts compared to the S&P 500 index.
In terms of risk management, the portfolio appears unstable. After traders took notice of the algorithm, Smith calculated its risk metrics. He found that the Sharpe ratio was 0.9413, indicating higher risk, but the Sortino ratio of 2.0021 shows a strong return-to-drawdown profile. For reference, a Sharpe ratio above 1.0 is generally considered acceptable.
