Home › Glossary

What is overfitting in trading?

In one line: Overfitting is when a trading strategy is shaped so closely to past data that it learns random noise, so it looks great in tests and does poorly on new data.

Trading glossary · Updated 2026-10-21

The idea

Picture a student who memorizes last year's test answers. They score 100% on last year's test. On a new test, they're lost.

An overfit strategy is that student. It "knows" the past perfectly and has learned nothing that carries forward.

A simple example

You test a moving average rule on one year of data. You try every length from 5 to 100.

Length 37 looks best: +40R over the year in testing. Length 35 is +5R. Length 40 is −8R.

That's a warning sign. If tiny changes flip the result, 37 probably just matched that year's random wiggles.

You run length 37 on the next year. It comes out around 0R. A math example, not a promise.

The warning signs

• Many rules or settings for a small number of trades.
• Results that collapse when a setting moves a little.
• Great results on one period, poor on the next.
• A Win rate or a results line that looks too smooth to be true.

Why it matters

Overfitting is the main reason backtests look better than real trading. You think you have an edge. You have a memory of the past.

The fix is to test on data the strategy has never seen. See walk-forward testing and why backtests don't guarantee results.

A common mistake

Adding a new rule after every loss in your test. "Skip Mondays." "Skip if RSI is 63." Each rule removes a few losers from the past. Together they fit the past and nothing else. Keep rules few and simple. See how to backtest a trading strategy.

Questions