The Sunday Open "Free Money Glitch" Dies on 585 S&P Sundays. Our First Reading of the Drift Underneath It Was Wrong.

We tested the viral “Sunday open is a free money glitch” mechanic on the Nasdaq in July and killed it. The kill came with a consolation prize: doing nothing but holding long from 20:00 ET to Monday’s cash open made +$46,586 over that window.

We ran the identical mechanic on the S&P, on 585 Sunday sessions from our own ES tick archive going back to January 2014. It loses again, harder and with more significance, and the direction read it hangs on is a coin flip.

Then we got the control wrong. Our first pass measured the drift underneath the mechanic on the Sunday evening session alone, 18:00 to 24:00 ET, and reported +$15.84 a night at t = 0.49. We wrote that the consolation prize does not exist on the S&P. But the NQ figure we were holding that against was not an evening session — it ran to Monday’s cash open. On the same window, ES pays +$163.62 a Sunday at t = 3.28, which is a stronger drift than the Nasdaq study found. The glitch is still dead. The thing underneath it is not.

The rule, as pitched

• Sunday Globex session, 18:00 to 24:00 ET.
• Read the direction of the move from the 18:00 open to 20:00 ET. Higher means long, lower means short.
• After 20:00, wait for the first 5-minute fair-value gap in that direction: three bars where the third bar’s low sits above the first bar’s high, mirrored for shorts. Enter at the close of that third bar.
• 20-point stop, 40-point target. Otherwise out at the session end.
• One contract. Nothing re-tuned for the S&P.

Where a bar contains both the stop and the target, we resolve the stop first.

The read is a coin

Everything downstream of the pitch depends on one claim: the 18:00–20:00 direction tells you where the session ends. It does not.

The chart below puts the read next to a coin, with the 95% confidence interval around the measured 48.2%. That interval covers the coin’s height. That is the whole finding.

	Value
Sessions	585
Read agreed with the session outcome	48.2%
95% interval	44.1% – 52.3%
p	0.41

Fifty sits inside the interval, closer to the middle than to either edge. Twelve and a half years of Sundays cannot distinguish this read from a coin, and the point estimate leans the wrong way.

The full mechanic, and where it actually exits

Adding the entry pattern to an uninformative read does not fix it. It adds a round trip.

Full mechanic, 585 Sundays
Trades	543
Win rate	41.4%
Average per trade	−$50.13
Median trade	−$29.50
Total, one contract	−$27,218.50
Profit factor	0.63
Max drawdown	$27,513.00
Best / worst trade	$1,970.50 / −$1,029.50
t	−3.35 (p = 0.0009)

The exit breakdown is what makes this readable. Of 543 trades, 519 ended at the session timeout. Twenty-three hit the 20-point stop. One reached the 40-point target — one, in twelve and a half years.

So the bracket is decoration too. A 20-point stop and a 40-point target are wide relative to what an ES Sunday evening usually does in the hours after 20:00, and price almost never gets to either. What is left is a coin-flip position held to midnight and charged $29.50. The median trade is exactly −$29.50, which is the cost line and nothing else.

That is why the loss is significant while the read is not. An uninformative signal has no expectation of its own; the fee does. Run it 543 times and the fee is the entire result.

The controls: four ways to spend a Sunday night

Here is the comparison the original claim never offers. The chart carries four lines on the same 585 sessions. Long from the 20:00 read all the way to Monday’s cash open, long the evening session only, the read’s side held as a plain position, and the full mechanic.

585 Sunday sessions	Avg	Total	Win rate	t	p
Long, 20:00 to Monday’s cash open	+$163.62	+$95,717.50	55.9%	3.28	0.0011
Long the evening session	+$15.84	+$9,267.50	50.8%	0.49	0.63
Follow the 20:00 read	−$49.07	−$28,707.50	43.6%	−2.54	0.0115
Full mechanic	−$50.13	−$27,218.50	41.4%	−3.35	0.0009
Short the evening session	−$74.84	−$43,782.50	43.2%	−2.29	0.0221

Three things fall out of that table.

The entry pattern is not the problem. Following the read as a plain position loses $49.07 a session. Adding the fair-value-gap wait and the bracket lands at $50.13. The gap logic changes almost nothing, because the coin flip in front of it already decided the outcome.

The direction filter is what does the damage. On NQ the same filter turned a +$46,586 drift into a loss. That drift carried t = 1.24, and we said at the time it was not deployable on its own. On ES the filter turns the evening session from +$9,267.50 into −$28,707.50. Half of all Sundays it puts you short against a drift, which is the same mechanism on both indices.

The top row is the one we got wrong. Long the evening session is +$15.84 a night at t = 0.49, which is the interval around zero. Long from the same 20:00 read to Monday’s cash open is +$163.62 a night at t = 3.28. The next section is about how the first number came to be published as the answer.

Shorting the evening session is worse than either losing rule. Both sides of the evening null lose once costs are in.

The window we measured, and the window we should have measured

Our first pass compared the ES evening session, 18:00 to 24:00 ET, against the NQ number from July. The NQ number was not an evening session. It was long from the 20:00 read to Monday’s 09:30 cash open, and it carried the whole night. We wrote “the consolation prize does not exist on ES” off a six-hour hold measured against a hold more than twice its length.

Run the NQ window on ES and the prize is there.

Long, same 585 Sundays	Evening session only	To Monday’s cash open
Win rate	50.8%	55.9%
Average	+$15.84	+$163.62
Total, one contract	+$9,267.50	+$95,717.50
Sharpe	0.14	0.98
Max drawdown	$25,073.00	$12,366.00
t	0.49	3.28
p	0.63	0.0011

Same Sundays, same side, same contract. The only thing that changes is where the position is closed. The P&L goes up tenfold, the worst drawdown halves, and the t-statistic moves from inside the noise to well outside it.

That also makes the S&P the stronger of the two indices on this measurement, which is not what we expected. The NQ version carried t = 1.24 over 291 Sundays, and we called it real but not tradeable alone. The ES version carries t = 3.28 over 585.

The lesson is ours to keep: a control is only a control if it runs over the same window as the thing it is controlling.

The years

Year	Evening session	To Monday’s cash open
2014	−$475.00	+$1,688.50
2015	+$737.50	+$4,356.50
2016	+$1,625.00	−$1,049.00
2017	+$825.00	+$855.50
2018	+$6,225.00	−$7.00
2019	+$287.50	−$3,190.00
2020	+$14,325.00	+$18,576.00
2021	−$2,475.00	−$3,058.00
2022	−$9,462.50	−$1,890.00
2023	−$712.50	+$6,848.50
2024	+$4,725.00	+$7,539.50
2025	+$4,937.50	+$27,475.00
2026	+$5,962.50	+$37,572.00

Our NQ archive starts in 2019 and our ES archive starts in 2014, and the five extra years are not where this drift lives. They contribute $5,844.50 of the $95,717.50. Eight of the thirteen years are green. 2025 and 2026 together carry $65,047.00, so this is a drift with a heavy recent tilt rather than a metronome, and 2019 lost $3,190.00.

The evening-only column shows why the six-hour read went nowhere. The blue line in the equity chart is the same series: above water at the end, and it spent 2021 through 2024 giving back more than half of what 2020 gave it.

What we changed in our own book

Nothing, and that is the honest answer. We never traded the Sunday mechanic, and we said in the NQ article that we do not trade the overnight drift as a standalone product either.

Three things did change in how we talk about it.

1. We published the wrong verdict on the S&P drift and we are correcting it here. The first pass held for six hours and the comparison held overnight. On the NQ window, Sunday-night drift on ES measures +$163.62 a session at t = 3.28.
2. Every control now inherits the window of the thing it controls. That is a harness change, not a wording change. Hold length is part of a comparison, and ours was not carried across from the NQ study.
3. The exit histogram goes into our standard output. 519 timeouts against 23 stops and 1 target told us more about this strategy than the P&L did. When a bracket almost never fills, the tested thing is not the strategy anyone described.

The pattern from the NQ write-up survives intact on the S&P: a viral claim is a background effect plus expensive decoration. Ask what doing nothing pays before asking what the rules pay. Then hold the nothing for as long as you hold the something. On 585 S&P Sundays, doing nothing from 20:00 to Monday’s open pays $163.62 a night, and the rules pay −$50.13.

If you want to re-run this yourself, the inputs are ours and they are for sale: ES ticks back to 2014 with the aggressor side on every print, the same archive this test used. The historical data packages carry the same files. The original Nasdaq verdict is here: the Sunday open glitch on NQ.

Methodology: ES Sunday Globex sessions, 18:00–24:00 New York time, 585 sessions from 5 January 2014 to 6 September 2026, built from our own tick archive with the real aggressor side on every print. Direction read: sign of the 18:00→20:00 move; outcome: sign of the 20:00→24:00 move. Entry at the close of the bar that completes the first 5-minute fair-value gap after 20:00 in the read’s direction, 20-point stop, 40-point target, otherwise out at 24:00. One contract, $4.50 commission plus two ticks slippage per round trip ($29.50 on ES). Where a bar spans both levels, the stop resolves first. Controls on the identical 585 sessions: long the evening session, short it, the read held as a plain position, and long from the 20:00 read to the next regular-session open, which is the window the NQ study used.