Validation method
Walk-forward analysis
Robust Walk-Forward ยท the exit check, rolled
Walk-forward analysis locks a rule on one window and checks that same rule on the next window.
What walk-forward analysis is
The usual procedure is a roll through time. A rule is chosen, or locked, on an earlier window. That same rule is then checked on the following window, which was not the window used to settle it. The start date moves forward and the check repeats. The question is whether the rule still behaves after the period used to choose it. A clean roll is not a promise that the next live period will match the last one.
How Alphavo runs it
Alphavo's Robust Walk-Forward rolls the exit check. The question on that book is the exit, carried forward as its own check. It does not fold calendar quarters into the same roll. In-sample calendar-quarter stability is T6, and T6 is a separate book, read after the specification is locked. Reading T6 as a rolled exit check would mix the two books. T6 is not walk-forward.
How to read the two books
Read the rolled exit check on the walk-forward book. Read each calendar quarter on the T6 book. When a strategy is shown, the star is 0 stars, 1 star, 2 stars, or 3 stars. A Candidate Strategy is still a candidate. A candidate is not Alpha.
Common questions
What is walk-forward analysis?
Walk-forward analysis chooses or locks a rule on an earlier window, then checks that same rule on the following window. The window then moves forward. It asks whether the rule still behaves after the period used to settle it.
Is in-sample calendar-quarter stability the same as walk-forward?
No. Alphavo keeps them in separate books. Robust Walk-Forward rolls the exit check. T6 is in-sample calendar-quarter stability after the specification is locked. T6 is not a walk-forward result.
Does a passed walk-forward test promise the next period?
No. A rolled check says how the locked rule behaved on windows that followed the choice. It does not promise that the next live period will match the last one. A Candidate Strategy is not Alpha.