LLM-Trading-Lab

Why we included this project

Anyone who has wondered whether a large language model can make sound trading calls with real money at stake will find a documented answer here. The repo is a forward-only record of a six-month experiment in which ChatGPT ran a micro-cap portfolio under strict rules, with daily CSV accounting, automatic stop-loss enforcement, and every chat and decision logged so anyone can audit what happened. A 40-page evaluation paper plus analytics scripts covering CAPM, Sharpe, Sortino, drawdown, and comparisons to the S&P 500 and Russell 2000 turn the raw trades into numbers you can judge. For teams or researchers building their own AI-managed trading experiment, the repository also works as a reusable template, showing how an LLM behaves as a portfolio decision-maker when mistakes cost real money.