"Edge-discovery batch 4 — new mechanisms (industry / event / diffusion)"

Date: 2026-09-25 | panel/split/costs as before. Candidates: H12/H12b industry-factor momentum (coarse-sector / GICS-industry EW 126d ret), H13 announcement premium (PIT-predicted earnings proximity, no surprise), H14 large-leads-small diffusion (trailing 5d return of a sector's top-5 $-volume peers -> non-large members). DSR N=15 cumulative. Placebo = within-date permutation (200 draws), gate at p<0.05.

#Scoreboard

#H12 industry-momentum -> RESEARCH

• OOS Sharpe 1.04 (IS 0.77), p vs 0 0.00, corrBASE nan, beta 0.96, placebo p 0.07 (null mean 0.98)
• v2: DSR 0.82 (N=15), PIT-200 0.95, lag2 1.02, tiered 1.04, pos-years 10/11
• cost ladder: optimistic=1.06, base=1.04, conservative=1.00, severe=0.93
• decay: H5=1.10; H10=1.08; H21=1.04; H42=1.09; H63=1.02
• yearly: 2016:1.76; 2017:2.66; 2018:0.38; 2019:1.83; 2020:0.81; 2021:1.38; 2022:-0.33; 2023:1.68; 2024:2.03; 2025:0.87; 2026:1.83
• score parts: {'oos_sharpe': 13.0, 'signif': 15.0, 'regimes': 15.0, 'costs': 8.9, 'perturb': 10.0, 'breadth': 5.0, 'capacity': 5.0, 'decorr': 0.0, 'plaus': 2.8}

#H12b industry-mom-gics-level2 -> RESEARCH

• OOS Sharpe 0.94 (IS 0.65), p vs 0 0.00, corrBASE nan, beta 1.04, placebo p 0.16 (null mean 0.88)
• v2: DSR 0.83 (N=15), PIT-200 0.77, lag2 0.90, tiered 0.91, pos-years 10/11
• cost ladder: optimistic=0.96, base=0.94, conservative=0.88, severe=0.77
• decay: H5=0.94; H10=0.97; H21=0.94; H42=0.91; H63=0.92
• yearly: 2016:1.91; 2017:1.77; 2018:0.09; 2019:1.73; 2020:0.20; 2021:1.16; 2022:-0.09; 2023:1.53; 2024:1.57; 2025:1.11; 2026:1.65
• score parts: {'oos_sharpe': 11.8, 'signif': 15.0, 'regimes': 15.0, 'costs': 8.2, 'perturb': 10.0, 'breadth': 5.0, 'capacity': 4.9, 'decorr': 0.0, 'plaus': 2.5}

#H13 announcement-premium -> RESEARCH

• OOS Sharpe 0.71 (IS -0.24), p vs 0 0.01, corrBASE nan, beta 0.20, placebo p 0.52 (null mean 0.71)
• v2: DSR 0.81 (N=15), PIT-200 0.70, lag2 0.51, tiered 0.49, pos-years 9/11
• cost ladder: optimistic=0.87, base=0.71, conservative=0.23, severe=-0.53
• decay: H5=0.71; H10=0.57; H21=0.53; H42=0.56; H63=0.57
• yearly: 2016:0.82; 2017:-1.32; 2018:-0.32; 2019:0.14; 2020:1.21; 2021:0.90; 2022:0.48; 2023:0.63; 2024:2.14; 2025:1.20; 2026:1.10
• score parts: {'oos_sharpe': 8.9, 'signif': 14.9, 'regimes': 15.0, 'costs': 0.0, 'perturb': 10.0, 'breadth': 5.0, 'capacity': 4.9, 'decorr': 0.0, 'plaus': 2.5}

#H14 large-leads-small -> RESEARCH

• OOS Sharpe 0.78 (IS 0.63), p vs 0 0.01, corrBASE nan, beta 0.90, placebo p 0.49 (null mean 0.78)
• v2: DSR 0.79 (N=15), PIT-200 0.77, lag2 0.79, tiered 0.75, pos-years 9/11
• cost ladder: optimistic=0.91, base=0.78, conservative=0.41, severe=-0.20
• decay: H5=0.78; H10=0.75; H21=1.01; H42=1.08; H63=1.01
• yearly: 2016:1.80; 2017:2.48; 2018:-0.12; 2019:2.07; 2020:0.56; 2021:1.53; 2022:-0.39; 2023:1.43; 2024:0.50; 2025:0.40; 2026:1.58
• score parts: {'oos_sharpe': 9.8, 'signif': 14.9, 'regimes': 15.0, 'costs': 0.0, 'perturb': 10.0, 'breadth': 5.0, 'capacity': 4.6, 'decorr': 0.0, 'plaus': 2.5}

#Reviewer note

• Sector/industry maps: sectors.json covers ~99 names (coarse 11 sectors); sectors_full.json (fetched this session) covers 497 of 503 discovery names at the GICS industry level (110 groups). H12 uses the coarse sector map (H12b re-tests it at industry level), H14 uses the sector map. Names absent from a map get NaN = no setup.
• H13 uses the naive PIT date predictor (hit rate ~0.90 within +/-15d); actual future announcement dates never enter the signal.

Runtime: 27.1 min.