Does the method of reruning GARCH models every day (to update parameter values and improve out-of-sample forecasting performance) have a name?

Question

It is my understanding that normally GARCH models make forecasts for say T-K days ahead. Instead of doing that I would like to use the data for days 1, 2, ...,k in my dataset to fit a GARCH model to make a 1-day ahead forecast for day k+1. Then I would like to use days 1, 2, ..., k+1 to fit a new GARCH model and to make a 1-day ahead forecast for day k+2. I would like to keep repeating this process and end up doing it a total of T - k times, with the last time consisting of me using the days 1, 2, ..., T-1 to fit GARCH model to make a 1-day ahead forecast for day T. Does this approach have a name in the literature? Thank you!

Answer 1

I think this is simply known as expanding-window estimation. This can be contrasted to rolling-window estimation (also moving-window estimation) where you keep the window size fixed by discarding initial observations when you add new ones. (We have the tag moving-window for that.)

This is not specific to GARCH models but can be used with any time series model.

The forecasts obtained using the rolling-window method may be called rolling forecasts. (I suspect forecasts from expanding-window estimation might be called rolling forecasts, too.)