Overfitting

What it is

Overfitting is when a rule is tuned so closely to past data that it captures noise instead of signal.

A rule with enough parameters can fit almost any history. The result looks excellent in backtest and fails out of sample — because it learned the specific past, not a repeatable pattern.

How to spot it

Sign	Why it suggests overfitting
Many parameters, short history	Enough freedom to fit anything
Performance collapses on unseen data	It memorised, it did not generalise
Sensitive to small changes	Nudge a parameter, the result swings wildly
No out-of-sample test	Only evidence of fit, not of generalisation

How to check

Ask whether the result was tested on data the rule was not tuned on. A backtest on the tuning data is not evidence — it is the thing being fitted.