A Backtesting Protocol in the Era of Machine Learning
Read the paperopens papers.ssrn.com in a new tab
What they found
Three prominent researchers offer a checklist for doing backtests honestly, aimed at the machine-learning era where the number of possible models is effectively unlimited. The protocol covers seven areas: having an economic rationale before testing, controlling for multiple testing, being careful with data (survivorship, look-ahead, outliers), cross-validation done correctly, model dynamics and regime shifts, complexity and its costs, and the research culture that rewards positive results. Each point comes with practical questions to ask of any strategy.
What you can use
- Start with a reason the strategy should work; a backtest without an economic story is a search for noise.
- The checklist is a usable audit for your own research: look-ahead bias, survivorship, tried-and-discarded variants, and regime changes.
- More complexity is not more edge; it is more parameters to overfit.
Caveats
A protocol and opinion piece, not new evidence. Practitioner journal. SSRN version linked.
Tags: backtesting, research-protocol, machine-learning, beginner-friendly
Summaries are our own reading of the paper, not the authors' words. Educational only, not advice. Discuss it in Book Club.