What is Curve Fitting?

Curve fitting is what happens when a strategy is tuned so tightly to past data that it stops describing a real market edge and starts describing noise. Also called over-optimisation, it produces a backtest that looks close to perfect and a live account that falls apart. The optimiser did its job too well: it found the exact settings that would have won on that specific history, and those settings tell you nothing about tomorrow.

What it is

Every market's history is a mix of real, repeatable behaviour and random noise. A good strategy captures the repeatable part. A curve-fitted one memorises the noise — the exact wick on a Tuesday in March that one parameter happened to catch. Run the optimiser hard enough, with enough parameters, and it will always find a combination that would have printed money on the data you gave it. That is not a discovery; it is software fitting a line through every dot.

The signs

• Too many parameters — the more knobs a strategy has, the easier it is to fit the past and the harder it is to trust.
• A flawless backtest — a near-straight equity curve with tiny drawdown almost never survives contact with a live market.
• A huge profit factor on few trades — a 3.0 profit factor across 40 trades is noise wearing a suit.
• Oddly specific values — a stop loss of 47 pips, or a filter that only fires at 09:14, is a number tuned to history rather than a reason.
• Fragile settings — if nudging one parameter slightly wrecks the backtest, the edge was never really there.

An everyday example

Imagine optimising an EA over three years of EUR/USD data. You let the software try thousands of combinations, and it settles on a take profit of 23 pips, a stop of 71, a moving average of 41 periods and a trade window of 08:00 to 10:30. The backtest looks superb. But none of those numbers came from a reason — they came from the software hunting for whatever happened to fit that exact stretch of history. Shift the date range by six months and a different "perfect" set appears. That is the tell: the settings describe the data, not the market, and the market is the only thing you will actually be trading.

Why it collapses live

A curve-fitted system is built for a market that has already happened. The moment live prices produce a pattern the optimiser never saw — which is immediately — the settings have nothing to grip. The equity curve that climbed in a straight line on historical data drifts sideways or drops in the first week of real trades. Nothing broke. There was never anything there to break.

How to avoid it

• Out-of-sample testing — optimise on one slice of history, then test on a separate slice the optimiser never touched. If it only works on the first slice, it is fitted.
• Walk-forward testing — repeatedly optimise on one window and test on the window that follows, rolling forward. It is the closest a backtest gets to honest.
• Fewer parameters — prefer a simple strategy you can explain over a complex one with a prettier curve.
• Forward testing — the strongest defence, because live data cannot be fitted after the fact.
• Distrust perfection — a slightly messy, believable curve beats a flawless one every time.

The perfect equity curve red flag

The strongest tell is the one that looks best: a backtest equity line that rises smoothly from corner to corner with barely a dip. Real strategies have losing streaks, flat months and drawdowns. A curve without them has usually been fitted until every loss was optimised away — which means every loss is still waiting in the live account, for the day the settings stop matching the market.

More in Grid & EA types →

Written and reviewed by the Karnek Research team. Last updated August 2026.

Educational content only - not financial advice. Past performance does not predict future results. Trading carries significant risk.