Curve-fitting explained: how to spot an over-optimized strategy

Curve-fitting, also called over-optimisation or overfitting, happens when a strategy is tuned so closely to past data that it captures random noise instead of a repeatable market behaviour. The result is a backtest that looks excellent and a live performance that doesn't.

How it happens

Suppose you test a moving-average crossover. The 20/50 version is fine. You try 17/52, then 18/47 with an RSI filter, then add a time-of-day rule, then a volatility filter. Each change improves the backtest a little. After enough changes, you've found the settings that happened to work best on that specific stretch of history, which is not the same as finding an edge.

Warning signs

• Lots of settings and rules. The more knobs, the easier it is to fit noise.
• An unrealistically smooth equity curve. Real strategies have rough patches.
• Performance falls off a cliff when you nudge a setting. If 20 periods works but 18 and 22 don't, the result is probably luck.
• Few trades. A great result on a small sample is easy to get by chance.
• It only works on one market or one period. Real edges usually show up, at least weakly, in related markets and other years.

Simple tests that catch it

1. Out-of-sample testing. Tune on one part of the data and test on a later part you didn't touch.
2. Parameter stability. Check that nearby settings give similar results. You want a broad plateau, not a sharp peak.
3. Walk-forward testing. Re-tune on a rolling window and test on the next period, again and again.
4. Forward testing on paper. Run the final rules on live data you couldn't possibly have tuned on.

PaperShieldAi strategies can only see each bar after it closes, they must pass a historical replay gate, and then they have to keep winning live, on paper, before they can take the crown.

Educational content only, not investment advice. Trading involves substantial risk of loss. See our Risk Disclosure.