Does the RS score have edge?
Mixed evidenceA pre-registered, point-in-time test of whether a high RS rating today predicts a higher return tomorrow — measured without hindsight. We rebuilt the exact live RS score on 35 past month-ends using only data available then, ranked the ~2,776-name liquid universe into deciles, and measured the return that actually followed. The score, universe, and momentum math are the same ones the board above uses.
Stability by year — 1M top−bottom spread
| Year | Rebalances | Spread | Mean IC | IC hit-rate |
|---|---|---|---|---|
| 2023 | 9 | -1.0% | -0.03 | 44% |
| 2024 | 12 | +2.2% | 0.03 | 67% |
| 2025 | 12 | +2.9% | 0.04 | 58% |
| 2026 | 2 | +2.9% | 0.04 | 50% |
How hindsight is ruled out
At each past rebalance date T, the RS ranking is rebuilt from grouped-daily closes dated ≤ T only — never today’s universe or today’s prices. Because each day’s data is whatever actually traded, names that later delisted are present at past dates, so there is no formation survivorship bias.
The forward return is close(T+H) / close(T) − 1, with H counted in trading days after T. The window that built the RS ends at T’s close; the window being measured starts at T’s close. No bar is shared between them.
Grouped bars are split-adjusted on one consistent scale, so multi-month returns aren’t corrupted by splits. Dividends are excluded — this is a price-return study, a small and roughly symmetric effect across deciles at these horizons.
A name that delists between T and T+H has no T+H close. The headline drops it; the count and a truncated-inclusive spread (which captures delisting losers) are reported alongside. In a liquid universe this is ≲ 1% of rows.
Primary horizon is 1M because monthly rebalancing makes its forward windows non-overlapping — the cleanest, least-autocorrelated test; 3M overlaps and its t-stats are optimistic, 1W sits in the short-term-reversal zone. close-to-close price return (ex-dividend), ranked over names with price > $5and avg $-volume > $10M. Method of record: docs/rs_edge_methodology.md; engine scripts/ml/rs_edge_study.py. Generated 2026-06-05. Not investment advice.