Tests a model only on data from after the period it was fitted on, which is the minimum bar for any market model.
What it asks you to do
Fit on an in-sample window, then test on the period immediately following it.
Roll both windows forward and repeat across the full history.
Report only out-of-sample results.
Where it struggles
Prevents look-ahead bias but not survivorship bias or overfitting through repeated retesting. Every extra variant you try on the same data quietly weakens the result.
Source: Time-series cross-validation methodology
Articles
No AI & Markets articles published yet — the frameworks above are the reference for now.