Anchored and Rolling determine how the in-sample period changes from one Walkforward segment to the next.
Both methods continue to place an out-of-sample period after the in-sample selection period.
Anchored Walkforward
Anchored keeps the beginning of the in-sample period fixed.
As the Walkforward moves forward, newer periods are added to the in-sample history.
Segment 1
In-sample: A + B
Out-of-sample: C
Segment 2
In-sample: A + B + C
Out-of-sample: D
Segment 3
In-sample: A + B + C + D
Out-of-sample: E
The starting point remains fixed while the in-sample history expands.
Rolling Walkforward
Rolling moves the in-sample window forward while keeping its configured length.
As newer periods enter the window, older periods drop out.
Segment 1
In-sample: A + B
Out-of-sample: C
Segment 2
In-sample: B + C
Out-of-sample: D
Segment 3
In-sample: C + D
Out-of-sample: E
The in-sample window moves through the dataset without expanding from one fixed starting point.
The difference in one view
Keeps the original in-sample start and adds newer history as segments progress.
Moves the full in-sample window forward and drops older history as newer history enters.
Both still follow the same sequence:
- evaluate candidate settings in-sample
- select one configuration
- apply that configuration to the later out-of-sample period
- move to the next eligible segment
Segment movement
The Walkforward configuration also defines how far the structure moves before the next segment begins.
Use Step Period to set this movement.
Where to go next
Continue with Best and Stable selection to see how one configuration is chosen from the in-sample candidates.