Build Production-Grade Backtesting Systems

This is a skill for building production-grade backtesting systems for trading strategies, aimed at producing reliable performance estimates while avoiding common backtesting pitfalls. It sets the workflow: define the hypothesis, universe, timeframe, and evaluation criteria; build point-in-time data pipelines paired with realistic cost models so results aren't inflated by look-ahead bias or unrealistic execution assumptions; implement event-driven simulation and execution logic; and validate with train/validation/test splits plus walk-forward testing rather than a single in-sample backtest. Detailed patterns and examples live in a companion resources/implementation-playbook.md file.

Use this skill when developing a trading-strategy backtest, building backtesting infrastructure, validating strategy robustness, or implementing walk-forward analysis. Don't use it for live trading execution or investment advice, when the historical data quality is unknown or incomplete, or when the task is only a quick performance summary rather than a rigorous validation. It carries an explicit safety constraint: never present a backtest as a guarantee of future performance, and avoid giving financial or investment advice.

Input is a trading strategy hypothesis, an asset universe, a timeframe, and historical price and cost data; output is a point-in-time-correct backtesting pipeline with event-driven simulation, split into train/validation/test periods and validated via walk-forward analysis, with detailed implementation patterns available in the companion playbook file.

Quantitative developers and researchers building or validating trading-strategy backtests who need to avoid common biases - look-ahead, survivorship, overfitting to a single historical period - rather than a quick, unvalidated performance summary.