Overfitting (curve fitting)
Every dataset contains patterns that happened by chance. If you adjust a strategy's parameters until the historical results look ideal, you inevitably absorb some of that noise — and noise does not repeat. The result is a strategy that looks brilliant in the backtest and mediocre or worse live.
Symptoms include: many finely-tuned parameters, performance that collapses when a parameter moves slightly, and results that differ wildly between adjacent time periods. Robust strategies tend to keep working across a range of settings.
Covered in depth in Lesson 04: Backtesting without fooling yourself.
Related terms
- Backtest — Running a strategy against historical price data to see how its rules would have behaved in the past.
- Forward test — Running a strategy on live or demo markets going forward, after development, to see if backtest behavior holds up.
Get new lessons by email
Occasional, plain-language lessons on automated trading — the same tone as everything on this site. No signals, no promises, unsubscribe anytime.
- One plain-language lesson at a time — no jargon, no hype.
- The full curriculum as a printable field guide.
- No signals, no performance promises. Unsubscribe anytime.
We store your address to send you educational content and nothing else. See the privacy policy.