What is it?
Walk-forward analysis: Walk-forward analysis repeats a time-aware development and evaluation process across ordered windows. Rolling windows keep training length approximately fixed; anchored windows keep the initial start while training history grows.
What question does it help answer?
Does the research conclusion persist when the development/evaluation boundary moves through time?
What can it tell you?
- How a conclusion behaves across multiple chronological evaluation periods.
- Whether evidence is concentrated in one favorable interval or appears across several folds.
- How rolling and anchored training designs expose different history-length assumptions.
What can it not tell you?
- Repeated windows do not guarantee independence between folds.
- Window lengths, retraining rules, costs, and selection performed inside each fold still matter.
- A stable historical walk-forward result does not guarantee future stability.
See the shape before reading the formula.
The Research Lab displays seven synthetic folds over 100 time units. Each blue training block is followed immediately by a gold held-out evaluation block, and the reader can compare rolling with anchored geometry.
Open this concept in Research Lab →Use the idea without turning it into a shortcut.
Keep the context: use this concept to interpret evidence and assumptions. For TraderCockpit controls and workflow details, open Docs. The app is in development; these examples are educational.