RSI+Accel Neutral

Market-neutral long/short book. Each day every liquid US stock is ranked within-day on RSI and on 20-day momentum acceleration; the two percentile ranks are averaged, the top 10 are bought and the bottom 10 shorted at the NEXT open, held 45 trading days with overlapping cohorts marked daily.

• Signalaverage of within-day percentile ranks of RSI and mom_accel_20d
• Constructionlong top-10 / short bottom-10, equal weight, entered next open, 45-day hold, 10bps one-way cost
• Universestock_close ≥ $5, ADV20 ≥ $20M, ETFs excluded
• Why it was promotedresearch/173 reported hedged Sharpe 1.50 at beta −0.04 — the best true alpha in the corpus once market beta was stripped, with K, hold, ADV floor and cost all swept rather than cherry-picked
• CORRECTION 2026-07-30that 1.50 was a WINDOW ARTIFACT. research/173 filtered the panel to its own fold range and only ever measured 2024-01 to 2026-05. Same cell, four windows: 173 window +1.47, FULL panel +0.69, 2022-23 only +0.05, since-freeze OOS +0.42. The strategy produced zero alpha across 2022-23 and the published figure excluded those years.
• Not tradeable as writtendaily rebalance × 45-day hold = ~45 overlapping cohorts = ~880 legs = $11 per position on a $10k book. A turnover sweep shows cutting rebalance frequency is nearly free (rebal 21d = 36 legs at $275, hedged +0.59), but the K=10 surface is rough (neighbour corr +0.27) so every cell in that family is the same ~0.6 ± noise.
• Honest caveathonest expectation is ~0.5–0.7 hedged, not 1.50. Forward record starts 2026-07-30 at 0% (an earlier 42-session backfill was removed as replay). Paper/shadow only, no orders.