A rolling window model involves calculating a statistic on a fixed contiguous block of prior observations and using it as a forecast . It is much like the expanding window, but the window size remains fixed and counts backwards from the most recent observation.