A validation design that moves the forecast origin, estimation window and next evaluation segment through time. It is not definitive proof and must state rolling or expanding windows, step, refit, horizon and use of the results.
→ Time-series validation and walk-forward · out-of-sample
Source
- Leonard J. Tashman, Out-of-sample tests of forecasting accuracy: an analysis and review, 2000 — rolling-origin designs and evaluation on successive observations not used for estimation.