What is Walk-forward backtest?
Testing the model only on data after each training window, the way it would have been used.
A walk-forward test moves through history in order: train on everything before a cutoff, predict the next slice, then step forward. Nothing is ever trained on data it is scored on.
This is the number the model card quotes and the site's gates use. A backtest that peeks at the future will always look better, and is worthless.
Related terms
- Model gateOur model
The rule that hides a forecast when the model could not beat a coin flip on it.
- Brier scoreOur model
Mean squared error of probability forecasts; 0.25 is the score of a coin flip.
- OverfittingOur model
Fitting noise instead of signal, so a model looks brilliant on old data and fails live.