Glossary
Walk-forward analysis
A test that fits a rule on one window of history, trades it unchanged on the next window, then rolls forward and repeats, so every result comes from settings chosen earlier.
In a walk-forward test the data is cut into training and test windows. The rule is fitted on the training window and applied, with no changes, to the test window that follows. The test results of all windows are joined into one out-of-sample series.
The training window can roll (fixed length) or anchor (growing from a fixed start). The window lengths are themselves choices, so the method reduces overfitting without removing it.
Example: ten years of data, a 3-year training window and 1-year test windows give 7 test windows and 7 years of out-of-sample results.
Sources
- Advances in Financial Machine Learning. Wiley (Marcos López de Prado), 2018. Accessed 9 Nov 2026