What is ai-quant-researcher?

AI-quant-lab is a Python framework for AI-assisted quantitative trading strategy research with a focus on statistical rigor. The core problem it addresses is this: if you use an AI like Claude to rapidly generate and backtest hundreds or thousands of trading strategies, the best-performing strategy in that pool will look good purely due to chance, the same way flipping a coin many times will eventually produce a long streak of heads. Without correcting for this "multiple-testing" problem, reported performance numbers are meaningless. Most similar frameworks skip this correction. The tool runs a research loop where Claude (via the Anthropic SDK) proposes and implements trading strategies, which are then backtested on historical price data. Each candidate must pass several statistical gates before being accepted: a deflated Sharpe ratio test that penalizes performance based on how many strategies were tested, a leakage detector that checks whether the strategy accidentally uses future data in its signals, a cross-correlation check to reject strategies that are just copies of already-accepted ones, and an adversarial critic that evaluates each strategy's reasoning. Walk-forward validation (testing on out-of-sample periods in sequence) is also included. Additional features include a cross-sectional portfolio engine for long-short strategies, realistic execution simulation with slippage and partial fills, and a production monitoring mode with automatic shutdown triggers. All research history is stored in SQLite so trial counts remain accurate across sessions. It requires Python 3.11 or newer, and examples 1 through 5 run without an Anthropic API key.