The turn-of-the-month effect on SPY (spec tom_spec.md). Run 2026-11-04T09:48:20.746Z; 500 control runs per window. 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; 405 month-ends; cash = 3-month T-bill (FRED DTB3) === TOM: buy at the close of day -2 (second-to-last trading day), sell at the close of day +3 (third trading day of the next month) === DEV n= 155 net 65.15% avg 0.420% PF 1.62 t 2.33 win 58.7% VAL n= 120 net 10.28% avg 0.086% PF 1.11 t 0.44 win 50.8% HOLDOUT n= 129 net 34.68% avg 0.269% PF 1.47 t 1.60 win 58.9% ALL n=404 net 110.11% (sum of trade %) avg 0.273% (gross 0.313%); PF 1.41 t 2.60; at 0.10% cost: sum 85.87% winners 228/404 = 56.4% (95% Wilson 51.6%-61.2%); without the 10 best 56.70%, without the 20 best 14.48% 10 best: 2009-03-30 6.90%, 2002-12-30 5.46%, 2015-09-29 5.46%, 2000-01-28 5.34%, 2010-08-30 5.26%, 2020-02-27 5.12%, 2011-11-29 5.10%, 2008-12-30 5.02%, 2001-09-27 4.93%, 1997-10-30 4.82% 10 worst: 2002-07-30 -7.92%, 2018-01-30 -6.37%, 1998-09-29 -6.00%, 2019-07-30 -5.66%, 2020-03-30 -5.18%, 2009-02-26 -5.18%, 1998-07-30 -5.07%, 2008-08-28 -4.77%, 1998-08-28 -4.70%, 2022-10-28 -4.67% by year (n, sum %): 1993 11 -1.55 | 1994 12 -1.50 | 1995 12 12.49 | 1996 12 7.66 | 1997 12 16.32 | 1998 12 -7.19 | 1999 12 7.37 | 2000 12 15.37 | 2001 12 6.14 | 2002 12 -11.16 | 2003 12 10.82 | 2004 12 3.18 | 2005 12 7.21 | 2006 12 -0.87 | 2007 12 -3.05 | 2008 12 -2.90 | 2009 12 0.82 | 2010 12 13.94 | 2011 12 -1.17 | 2012 12 8.81 | 2013 12 -0.20 | 2014 12 -3.16 | 2015 12 -1.95 | 2016 12 2.32 | 2017 12 4.61 | 2018 12 -2.55 | 2019 12 2.48 | 2020 12 14.51 | 2021 12 0.21 | 2022 12 -2.10 | 2023 12 5.31 | 2024 12 -1.83 | 2025 12 0.81 | 2026 9 10.90 quarter-end turns n=135 avg 0.388% PF 1.59 | other turns n=269 avg 0.215% PF 1.32 rest of the month (from day +3 close to the next day -2 close, same cost) n=403 avg 0.608% (gross 0.648%) PF 1.50 t 3.10 average day around the turn (total return %, share up): -5 0.028 (55.2%) | -4 0.145 (54.0%) | -3 0.063 (54.0%) | -2 0.101 (54.5%) | -1 -0.041 (46.0%) | +1 0.219 (62.5%) | +2 0.076 (56.5%) | +3 0.066 (54.1%) | +4 0.031 (53.1%) | +5 0.046 (53.1%) | days +6..-6: 0.024 (53.8%, n 4434) control (4-day holds at random places in the same months, 500 runs, 0 s): median avg 0.142%, 5-95% -0.016 .. 0.311%; real avg 0.273% at percentile 90.40 [FAIL] 1 PF >= 1.2 in validation: VAL PF 1.11 [PASS] 2 >= 50 trades in each period: 155 / 120 / 129 [PASS] 3 t >= 2.0 on development: DEV t 2.33 [FAIL] 4 above 95th pct of the control: percentile 90.40 [PASS] 5 positive without top-10 trades: 56.697% [PASS] 6 holdout net positive (flagged used): HOLDOUT net 34.681% PF 1.47 VERDICT Buried (4/6 gates) whole strategy with cash interest, 1993-02-25 .. 2026-10-20: x5.43, 5.2% a year, worst drawdown -18.0% (2002-05-01 -> 2002-10-31), volatility 7.9%, invested 19.1% of days buy-and-hold SPY (total return), same dates: x31.71, 10.8% a year, worst drawdown -55.2% (2007-10-09 -> 2009-03-09), volatility 18.5% crashes, strategy / buy-and-hold: 2000-02 dot-com 1.5% / -47.5% | 2008-09 financial crisis -12.6% / -55.2% | 2020 Covid 5.2% / -33.7% | 2022 rate hikes 1.4% / -24.5% years better than buy-and-hold: 11 of 34 rule 3: not applicable (the only gaps in the SPY file are real market closures: Sept 2001, 2 Jan 2007, Oct 2012) === Neighbouring windows (descriptive, after the gates) === TOM -2..+3 n=404 (155/120/129) avg 0.372% PF 1.52 DEV t 2.55 VAL PF 1.13 HOLD 59.109% pct 95.60 w/o top10 87.80% -> 5/6 TOM -3..+3 n=404 (155/120/129) avg 0.434% PF 1.58 DEV t 2.41 VAL PF 1.39 HOLD 56.484% pct 96.60 w/o top10 109.81% -> 6/6 TOM -1..+2 n=404 (155/120/129) avg 0.209% PF 1.37 DEV t 2.27 VAL PF 1.08 HOLD 26.363% pct 89.20 w/o top10 40.73% -> 4/6 TOM -1..+4 n=404 (155/120/129) avg 0.299% PF 1.40 DEV t 2.78 VAL PF 0.96 HOLD 44.789% pct 85.20 w/o top10 67.40% -> 4/6 TOM -1..+1 n=404 (155/120/129) avg 0.137% PF 1.27 DEV t 2.46 VAL PF 0.95 HOLD 15.131% pct 86.40 w/o top10 14.57% -> 4/6