Free sources can be incomplete, delayed, revised, or adjusted differently than institutional feeds. Yahoo Finance data may not include delisted securities, point-in-time constituents, borrow costs, accurate corporate actions for all instruments, or exchange-grade quote history.
Macroeconomic series are revised after initial publication. A rigorous point-in-time macro backtest should use vintage data. This repository uses current historical series, which can introduce revision bias.
Yahoo options chains are snapshots, not historical full-depth quote data. The implied volatility surface builder is useful for analytics and research structure, but production derivatives systems require exchange-quality quotes, no-arbitrage validation, dividend curves, rate curves, exercise style handling, and market microstructure checks.
The engine models daily close-to-close fills with linear costs. It does not model order-book depth, partial fills, borrow constraints, financing, margin calls, taxes, exchange fees, latency, intraday volatility, or market impact. Strategy results should be interpreted as research diagnostics, not live trading expectations.
Cointegration can be unstable. Macro regimes can be sensitive to feature selection and clustering assumptions. Momentum can be crowded and regime-dependent. Backtests can overfit through parameter selection. Every model should be stress-tested before being used for capital allocation.
This project is research software, not investment advice.