Overfitting (curve-fitting) is the failure mode where a strategy's parameters are tuned so aggressively to the in-sample period that they end up describing noise instead of repeatable structure. The signature is great in-sample performance and poor out-of-sample performance.
Overfitting risk grows with the number of parameters tuned and the amount of search done over them. Six binary on/off filters yield 64 combinations; the best-of-64 will look great on any data set by chance alone. This is the Multiple-Testing problem applied to trading.
Mitigations: keep parameter count small, optimize on long histories, hold out untouched OOS data, walk-forward in production, and stress-test by deliberately perturbing parameters to confirm performance is not razor-thin around the chosen values.
Come Noon Barbari usa Overfitting
Ogni concetto qui è implementato nella piattaforma. Apri la documentazione o lo strumento corrispondente per vederlo all'opera.
Anti-overfit workflow →Termini correlati
- Backtesting
Walk-forward optimization
Optimize on a rolling in-sample window, validate on the next out-of-sample slice.
- Backtesting
Out-of-sample
Data the strategy was not allowed to see during parameter selection.
- Backtesting
In-sample
The portion of history used to fit the strategy's parameters.
- Backtesting
Backtest
Simulating a trading rule on historical data to estimate how it would have performed.