TESTING · 6 MIN READ
Curve fitting, and how to recognise it
Curve fitting is tuning a strategy until it matches the noise in one stretch of history, so it shines in the backtest and fails on new data. It is rarely deliberate. It builds up from reasonable decisions made one at a time, and the equity curve it produces is beautiful and meaningless.
How curve fitting happens
Adjust a threshold, the backtest improves, keep the change. Each decision can be defended, and together they produce a system shaped entirely by the noise in one sample of history. No single step is the mistake.
Automated optimisation does not cause it, but it makes it efficient. Sweep thousands of parameter combinations and some are certain to look excellent by chance. Picking the best of them is a very good way to find the luckiest combination, which is seldom the best rule.
Signs a strategy has been curve fitted
- A big gap between in-sample and out-of-sample results. Some decline is normal; a strategy that was outstanding and turns break-even was fitted.
- Parameter instability: the best setting jumping around between neighbouring test windows.
- Results that collapse when a parameter is nudged slightly. A real edge sits on a plateau, where nearby settings do about as well; a fitted one is a spike.
- Profit factor above 2.0 on a short sample. It is possible, but assume fitting until shown otherwise.
- A big gap between how often a strategy trades in replay and how often it trades live. Our DOUBLE-TAP strategy fired 233 times in replay and never once live.
A curve-fitting example from our own bot
An earlier version scaled the reward-to-risk ratio by market regime: 1.6 in mixed conditions, higher in trends. It improved in-sample results, which is exactly what a fitted parameter does.
Testing it on an out-of-sample period showed the scaling was fitting noise. We removed the feature and used a single 2.0 ratio everywhere, and the system got better by getting simpler. Real findings often look like that: the change that helps is usually a deletion.
Another out-of-sample test went against us. We wrote the test down in advance and then ran the four strategies in the bot on their own, on gold from 2019 to 2024, years they were not tuned on. None of the six years was positive. We publish that, and the site never calls the bot's edge proven.
Keep reading
- Walk-forward analysis explainedWalk-forward testing repeatedly fits on one window and tests on the next, producing an equit…
- How to backtest a strategy without fooling yourselfA backtest is the weakest evidence worth having. Here is what makes one trustworthy, and the…
- Reading a verified track record properlyA verified badge proves the trades happened. It does not prove the account is live, the only…
- The 34 strategies and the 24 we cut
Further reading
Educational information only, not financial advice. Trading leveraged products carries substantial risk of loss. Last updated 2026-11-11.