High School Student's ChatGPT Trading Bot Is Crushing the Russell 2000

In short

Nathan Smith’s ChatGPT-powered micro-cap bot is up 23.8% in 4 weeks—blowing previous the Russell 200 and professional inventory pickers.
In contrast to viral AI buying and selling claims, Smith’s experiment is absolutely documented, open-source, and prepared for anybody to strive.
Wall Avenue’s bots watch their backs: Smith’s journey from rural excessive schooler to finance prodigy is simply getting began.

A highschool child from rural Oklahoma simply did what Wall Avenue’s algorithms have not: he let ChatGPT run wild with $100 and watched it outperform the market by an enormous margin.

Nathan Smith’s experiment began 4 weeks in the past with a easy premise—give ChatGPT full management over a small portfolio of micro-cap shares and see what occurs.

The outcomes? A 23.8% return whereas the Russell 2000 and biotech ETF XBI crawled up simply 3.9% and three.5% respectively.

Picture: Nathan Smith

“After I was watching YouTube, I used to be continuously getting this advert about some AI inventory picker: ‘(the advert stated) We feed our buying and selling algorithm all shares on the NYSE blah blah blah…'” Smith instructed Decrypt. “I then began doing a little analysis and was stunned no person had tried a completely led LLM portfolio earlier than.”

The setup sounds deceptively easy. Smith gave ChatGPT a transparent mandate: construct the strongest potential inventory portfolio utilizing solely full-share positions in U.S.-listed micro-cap shares with market caps beneath $300 million. The target was easy—generate most returns from June 27 to December 27, 2025.

What units this aside from different buying and selling algorithms is the whole autonomy.

“The AI takes care of completely all the pieces. Place sizing, cease loss, and many others., are none of my selections,” Smith instructed Decrypt in an e-mail. The one human intervention comes when ChatGPT contradicts itself—a quirk he acknowledges as one of many system’s drawbacks.

Smith not too long ago calculated the danger metrics that skilled merchants obsess over.

His Sharpe ratio sits at 0.9413, indicating excessive danger, whereas the Sortino ratio of two.0021 suggests sturdy upside positive factors with restricted draw back. For context, a Sharpe ratio above 1.0 is usually thought-about good, whereas something beneath suggests the returns may not justify the danger.

The chatbot, which is absolutely open-sourced and is offered on Smith’s GitHub repo, has no diamond fingers and appears to be fairly goal, more often than not.

One in every of ChatGPT’s strikes concerned CADL, a inventory that generated roughly 50% of the portfolio’s earnings.

“It bought CADL and not using a second thought,” Smith stated. “I feel it correctly knew that in micro-cap territory, all positive factors may be worn out instantly. Not many hedge funds may make such a decisive transfer.”

Elegant in its complexity

Smith’s system has been working for under a month, which isn’t sufficient to backtest or assess with a excessive confidence degree, however the optimistic outcomes up to now are promising.

The technical infrastructure behind this experiment is a little more advanced than merely utilizing your typical chatbot, however it’s nonetheless easy to implement with a little bit little bit of dedication.

Smith constructed 5 primary features: guide shopping for and promoting for brand new picks, portfolio processing for monitoring trades, every day outcomes technology utilizing Yahoo Finance information, and graph technology for visualizing efficiency in opposition to the S&P 500.

“Truthfully, the setup is fairly easy,” he stated, describing a system that pulls benchmark costs from Yahoo Finance API into Panda’s information frames for evaluation.

ChatGPT chooses shares as soon as per week, and all the time stays beneath the $300 million market cap restrict, whereas Smith executes the trades and logs the outcomes.

{The teenager} from a tiny Oklahoma city found his ardour nearly by chance. “I’ve coded a little bit up to now (engaged on Harvard’s on-line CS course CS50), however utilizing C did not really feel invigorating (silly segmentation fault errors),” he instructed Decrypt. “Over the summer time, I found Quantitative Finance and the great thing about Python, and I fell in love.”

Oh, the issues excessive schoolers do these days.

With nearly 1,000 GitHub commits this previous yr and a superb quantity of followers on his newsletter, Smith has thrown himself into the world of quantitative finance. He plans to increase the experiment to a full yr as soon as it wraps in December, although he admits balancing it with ACT learning and self-studying for AP Psychology will likely be difficult.

“I actually assume I’ve discovered my ardour in life and hope to proceed this as an actual job in the future,” Smith stated.