Trading term
What is Curve fitting?
Curve fitting, or over-optimisation, is tuning a strategy until it fits the historical data almost perfectly — including the random noise in it. The result looks superb in a backtest and fails immediately in live trading.
Every price series contains both signal and noise. Add enough parameters, or adjust the ones you have finely enough, and a strategy will start capturing the noise too — the particular accidents of that particular period. The backtest improves with every adjustment, which is exactly what makes the process so seductive.
The tell is the gap between in-sample and out-of-sample performance. A fitted system might return +45R over the data it was tuned on, at an implausible +1.50R per trade, and then lose money on the very next stretch of data it has never seen. Nothing about the market changed; the strategy was describing history rather than any repeatable behaviour.
The defences are all forms of restraint: use few parameters, prefer values that sit on a broad plateau of decent results rather than a lone spike, hold data back and test on it only once, and be suspicious of any result that looks too good. A strategy that works acceptably across a wide range of settings is far more likely to survive than one that works brilliantly at exactly one.
+1.50R per trade while it was being tuned; −0.55R the moment it meets unseen data. The backtest improved as the strategy got worse — that's the whole trap.
For example
A tuned system returns +45R in-sample with barely a drawdown — an eye-watering +1.50R per trade. Run on 20 unseen trades it loses 11R. The in-sample curve was a description of past noise, not evidence of an edge.
Go hands-on in Premium
That's Curve fitting in theory — it clicks when you read it on a live chart. Practise it hands-on in the TradeWize Premium Trading Psychology & Risk track.
Explore Premium →Why it matters to you
Curve fitting is how most retail systematic strategies die, and it does its damage silently: the backtest gets better as the strategy gets worse. Recognising the pattern — a spectacular in-sample result that collapses out-of-sample — is what stops you funding an account on the strength of a fantasy.
⚠ Re-tuning after a failed out-of-sample test burns the data
The single held-out test is only meaningful once. Traders who see poor out-of-sample results and go back to adjust parameters have quietly turned that data into more in-sample data, and the next test proves nothing. If the out-of-sample run fails, the honest conclusion is that the idea failed.