PASS  Momentum / Ranking

Momentum Ranking Model: Out-of-Sample Validation (MOM-E1)

Can a model trained only on decision-time information rank which small caps will move unusually large the next day?

The model behind the Day-0 watchlist was retrained on the correct decision-time universe and evaluated walk-forward: every prediction uses only information available before the session it predicts, and evaluation windows never overlap training data.

Result: strong out-of-sample discrimination, with the top of the ranking capturing next-day large movers at roughly 10× the universe base rate (the forward measurement of that lift is published as its own finding, Day-0 Watchlist). Full evaluation metrics are pending research-review release.

What this means: the ranking signal is real, out-of-sample, and survives honest evaluation.

What this does NOT mean: ranking skill is not entry-strategy profitability. Where we tested direct trading implementations of adjacent signals, most died (see the kills in this archive). The model earns its place as information, and only that claim is made.

← All findings · How we measure

This page describes a research verdict under our published methodology. It is not trading advice, and information claims are not profitability claims. Past results do not guarantee future outcomes.