The pre-holiday effect on SPY (spec ph_spec.md). Run 2026-11-04T10:34:56.862Z; 500 control runs per hold. Trade results in % of the price (total return, close to close), cost 0.04% a round trip.
data: 8489 days 1993-01-29 .. 2026-10-20; 136 dividends; cash = 3-month T-bill (FRED DTB3)

=== PH: buy at the close of the day before the pre-holiday day, sell at the close of the pre-holiday day (the last trading day before a scheduled holiday) ===
  DEV      n= 109 net -0.82%  avg -0.008%  PF 0.98 t -0.08 win 50.5%
  VAL      n=  89 net 5.80%  avg 0.065%  PF 1.20 t 0.59 win 55.1%
  HOLDOUT  n= 100 net 13.97%  avg 0.140%  PF 1.77 t 1.97 win 62.0%
  ALL      n=298 net 18.95% (sum of trade %) avg 0.064% (gross 0.104%); PF 1.22 t 1.17; at 0.10% cost: sum 1.07%
  winners 166/298 = 55.7% (95% Wilson 50.0%-61.2%); without the 10 best -8.38%, without the 20 best -23.30%
  10 best: 2009-04-09 (Good Friday) 3.93%, 2008-11-26 (Thanksgiving) 3.82%, 2000-12-22 (Christmas) 2.96%, 1999-09-03 (Labor Day) 2.88%, 2002-11-27 (Thanksgiving) 2.77%, 1999-01-15 (Martin Luther King Jr. Day) 2.57%, 2022-05-27 (Memorial Day) 2.41%, 2003-02-14 (Washington's Birthday) 2.15%, 2016-02-12 (Washington's Birthday) 2.02%, 2008-03-20 (Good Friday) 1.80%
  10 worst: 2009-07-02 (Independence Day) -2.77%, 1997-03-27 (Good Friday) -2.68%, 2018-12-24 (Christmas) -2.68%, 2011-09-02 (Labor Day) -2.59%, 2000-11-22 (Thanksgiving) -2.43%, 2011-11-23 (Thanksgiving) -2.25%, 2001-02-16 (Washington's Birthday) -2.24%, 2000-02-18 (Washington's Birthday) -2.19%, 2016-01-15 (Martin Luther King Jr. Day) -2.19%, 2007-11-21 (Thanksgiving) -2.09%
  by year (n, sum %): 1993 7 -1.80 | 1994 8 -1.30 | 1995 8 -0.34 | 1996 8 -2.96 | 1997 8 -0.57 | 1998 9 -0.17 | 1999 8 9.15 | 2000 9 -0.36 | 2001 9 -3.73 | 2002 9 -0.80 | 2003 9 2.03 | 2004 8 -0.76 | 2005 9 0.79 | 2006 9 -0.17 | 2007 9 0.34 | 2008 9 3.99 | 2009 9 1.64 | 2010 8 -0.01 | 2011 9 -1.41 | 2012 9 1.73 | 2013 9 0.63 | 2014 9 0.36 | 2015 9 -1.30 | 2016 9 0.44 | 2017 9 -0.82 | 2018 9 -0.41 | 2019 9 3.93 | 2020 9 2.30 | 2021 8 2.33 | 2022 10 1.36 | 2023 10 1.68 | 2024 10 1.91 | 2025 10 0.54 | 2026 7 0.71
  by holiday (n, avg %, PF, win): Washington's Birthday 34 -0.193 0.53 35.3% | Good Friday 34 0.366 3.19 67.6% | Memorial Day 34 0.014 1.05 55.9% | Independence Day 34 0.198 2.10 70.6% | Labor Day 34 -0.025 0.93 50.0% | Thanksgiving 33 0.166 1.62 66.7% | Christmas 33 0.156 1.82 57.6% | New Year 28 -0.341 0.35 32.1% | Martin Luther King Jr. Day 29 0.163 1.54 62.1% | Juneteenth 5 0.190 3.19 60.0%
  in the turn-of-month window n=106 avg 0.024% | outside n=192 avg 0.086% PF 1.30 t 1.21
  every other day of the file (same cost) n=8190 avg 0.006% (gross 0.046%) win 51.8%
  control (one-day holds at random days of the same months, 500 runs, 0 s): median avg 0.004%, 5-95% -0.099 .. 0.117%; real avg 0.064% at percentile 82.60
  [PASS] 1 PF >= 1.2 in validation: VAL PF 1.20
  [PASS] 2 >= 50 trades in each period: 109 / 89 / 100
  [FAIL] 3 t >= 2.0 on development: DEV t -0.08
  [FAIL] 4 above 95th pct of the control: percentile 82.60
  [FAIL] 5 positive without top-10 trades: -8.376%
  [PASS] 6 holdout net positive (flagged used): HOLDOUT net 13.970% PF 1.77
  VERDICT Buried (3/6 gates)
  whole strategy with cash interest, 1993-02-11 .. 2026-10-20: x2.73, 3.0% a year, worst drawdown -5.3% (2015-05-20 -> 2016-01-15), invested 3.5% of days
  buy-and-hold SPY (total return), same dates: x31.28, 10.8% a year, worst drawdown -55.2%
rule 3: not applicable (the only gaps in the SPY file are real market closures; unscheduled closures are not holidays)

=== Neighbouring holds (descriptive, after the gates) ===
  PH 2 days          n=298 (109/89/100) avg 0.093% PF 1.21 DEV t -0.67 VAL PF 1.41 HOLD 20.864% pct 75.00 w/o top10 -9.18% -> 3/6
  Across the holiday n=298 (109/89/100) avg 0.012% PF 1.03 DEV t 0.71 VAL PF 1.10 HOLD -9.896% pct 54.20 w/o top10 -30.81% -> 1/6
  Before and across  n=298 (109/89/100) avg 0.116% PF 1.24 DEV t 0.74 VAL PF 1.29 HOLD 7.925% pct 79.40 w/o top10 -7.83% -> 3/6