Overfitting in backtesting is tuning a strategy to capture the coincidental noise in a historical dataset as if it were a real, repeatable edge. The result performs well on the exact data it was tuned against and often fails anywhere else.
How it works
The risk grows with the number of adjustable parameters relative to the amount of independent data.
A strategy with many parameters on little data has far more room to match coincidence than one with few parameters on a large sample.
Why it matters
An overfit backtest overstates the strategy's real edge, which becomes expensive when the result fails live.
It is exposed by testing on periods the strategy was never tuned against and by checking whether the pattern holds across independent sub-periods.
A simple market example
A researcher adds a ninth parameter purely because it improves the historical fit, with no independent reason. The backtest looks better, but the improvement is the fit to noise, not a discovery.
Common mistakes
Judging a strategy by how good the historical fit looks instead of by independent-period consistency.
Adding parameters only to improve the fit, without independent justification.
Frequently asked questions
Is a good backtest proof of an edge?
Not by itself. It is evidence only if it survives checks for parameters, justification, sub-periods, and date sensitivity.
How many parameters are too many?
There is no fixed number. What matters is the count relative to the data and whether each parameter has an independent reason to exist.
How is overfitting exposed?
By testing on periods the strategy was never fitted to, and by splitting history into independent sub-periods.
Educational content only. Definitions describe common market usage and may vary by jurisdiction, instrument, or institution.