Quantitative finance
Quantitative strategy without backtest overfitting
By the end you'll know why most published Sharpe ratios are inflated, what a real walk-forward backtest looks like, and the discipline required before betting capital on a signal.
4 steps · ~30 minutes of reading total
- 1
Premise: an in-sample Sharpe of 2.0 means almost nothing
MilestoneMost published trading strategies are over-fitted to the period they were tuned on. Out-of-sample performance is the only number that matters and it's usually much worse.
- 2
Bailey, Borwein, López de Prado, Zhu — "Pseudo-Mathematics and Financial Charlatanism"
AMS Notices ↗Foundational paper on why testing enough strategies guarantees finding a fake signal. Mandatory reading before any backtest.
- 3
Milestone: you can explain why a 5-year Sharpe of 1.5 is probably noise
MilestoneIf you can articulate this in terms of the multiple-testing problem, you're past the worst trap in retail quant.
- 4
López de Prado — Advances in Financial Machine Learning (sample chapters)
ResearchGate ↗The book's central contribution: combinatorial purged cross-validation. The honest way to backtest when your data has serial dependence.