The session prints its extreme early

Split the regular trading session (09:30-16:00 ET) into four equal quarters and ask which one prints the day's eventual high or low. The first quarter -- ending at 11:07:30 ET -- wins 45.6% of the time, against a volatility-clock null of 43.2%. That's a real, repeatedly-attacked excess of +2.4 percentage points, not an artifact of mornings simply being more volatile -- the null already knows that. A second, related-sounding claim looks at what happens next: once the first quarter's extreme is locked in, the session closes in the opposite half of its range 76.8% of the time. That number is true. It is also not information -- a null built with the same conditioning says 77.1%, and the two are statistically indistinguishable.

• ES and NQ futures, 2010-2026, RTH session (09:30-16:00 ET) split into four 97.5-minute quarters (Q1 ends 11:07:30 ET); compared against a volatility-clock null that preserves each time-of-day's own realized volatility, including the 9:30 open spike
• Q1 holds the session's defining extreme 45.6% of the time (held-out sample) vs. 43.2% for the null -- +2.4pp, p=0.005 (fitting sample: 47.9% vs. 43.2%, +4.6pp)
• The excess survives seven separate attacks: a volatility-matched null, ES and NQ tested apart, year by year (15 of 16 complete years positive), excluding 2020, excluding the opening 15 minutes, and excluding gap days -- alone and combined
• A finer, 22.5-minute version of the same claim does not survive: it collapses once gap/opening-print days are excluded (p=0.41) -- see "What died" below
• Conditional on Q1 holding the extreme, the session closes in the opposite half of its range 76.8% of the time -- but a same-conditioned null says 77.1%. The "rule" is true and carries no information over its own baseline

The null already knows mornings are busier. What's left over is the part that isn't just volatility.

More than a busy open explains

The regular session runs 09:30-16:00 ET, 390 minutes. Split it into four equal 97.5-minute quarters and track which quarter ends up holding the session's defining high or low once the closing bell rings. Q1 (09:30-11:07:30 ET) wins outright more than any other quarter -- but so does the null, because mornings are the most volatile part of the day and a bigger quarter's-worth of volatility means more chances to set an extreme. The question worth asking is not "does Q1 win the most", it's "does Q1 win more than its own volatility already predicts".

The null used here answers exactly that: it resamples days from the same market and era while preserving each time-of-day's own realized-volatility clock, the 9:30 open spike included. Whatever the null gets, "more volatility in the morning" is already baked in. What's left is the table below.

Q1 is the only quarter with a positive, significant excess in both the fitting and the held-out sample. Q4 shows the mirror image -- a significant deficit (30.6% vs. 31.5% null, HELD): afternoons under-extend about as reliably as mornings over-extend. Q2 and Q3 sit close to their null throughout.

It isn't only Q1

A related way to ask the same question: given that a specific quarter is currently holding the session's running extreme once that quarter ends, how often does it keep holding it all the way to the close? The table below reads as "checkpoint quarter" (rows) against "quarter the final extreme actually lands in" (columns collapsed into one column per row).

Only the first row's own diagonal (Q1 holds after Q1 -- the same 45.6% number as above) belongs to the corrected, pre-registered family this page reports on. The rest of the matrix is context, not an independently tested claim: once Q2 or Q3 is the one holding the running extreme, its own odds of keeping it (Q2: 32.8% vs. 31.6% null; Q3: 38.4% vs. 37.9% null) sit close to their null and aren't distinguishable from it at this sample size. And once Q1 has the running extreme, later quarters are correspondingly a little less likely than chance to take it away outright (Q1→Q3: 11.5% vs. 12.5% null; Q1→Q4: 30.6% vs. 31.5% null) -- the flip side of the same effect, not a new one.

What survived seven attacks

A +2.4 percentage point excess is small enough that it's fair to ask whether it's real or a residual of some confound the null didn't fully strip out. Seven attacks, built specifically to try to kill this number, run against the held-out sample:

None of the seven flips the sign or removes significance. The volatility-matched null -- the one attack expected to take a real bite, since it directly narrows the exact gap this page measures -- still leaves 91% of the baseline excess standing. Splitting by market, by year, and by whether the opening minutes or gap days are in the sample all land on the same small, positive number.

What died: a finer version that was just the opening print

The same question, asked first on a finer 22.5-minute quarter grid (09:30-11:00 ET, so Q1 ends at 09:52:30 ET instead of 11:07:30), looked even stronger at first glance: +1.76pp (held-out sample, p=0.005). It didn't survive the same scrutiny the 97.5-minute version did.

On the fine grid, 80-83% of the quarter's eventual extremes sit in the first 15 minutes -- the session's opening print itself. Exclude the days where the extreme literally is the opening print, and the excess collapses to +0.33pp (p=0.41, not distinguishable from noise). The coarser 97.5-minute version above doesn't have this problem to nearly the same degree: only 43-49% of its Q1 extremes are the opening print, and its excess survives the identical exclusion (+1.58pp, p=0.005). The finer claim measured the opening print wearing a bigger name. The coarser one measures something the opening print alone doesn't fully explain -- which is why this page reports the 97.5-minute version, not the finer one.

A rule that is true and worthless at the same time

Take only the sessions where Q1's running extreme held all the way through -- the same 45.6% of days from above. Ask a natural follow-on question: does the session then close in the opposite half of its range? If Q1 set the day's high, does price finish below the range's midpoint; if Q1 set the low, does it finish above?

Yes, 76.8% of the time (held-out sample). That reads like a strong rule -- more than three sessions out of four apparently "fade" back across the middle after an early extreme.

A null built with exactly the same conditioning says 77.1%. The rule is true. It is also almost exactly what a session with no memory of its own open would produce.

The held-out sample can't tell the real data from its null (p=0.64) -- and in the fitting sample, the real rate was significantly below the null (75.1% vs. 77.1%, p=0.005), the wrong direction for an edge. Neither reading supports "fade the early extreme" as a rule that adds anything.

The reason is geometry, not behavior. Fix one end of a range early -- the session's high prints by 11:07 ET and holds for the rest of the day -- and the close has only one side left to plausibly land on: somewhere relative to the range's midpoint, on the side away from the already-fixed extreme. A session that spends its remaining hours wandering with no memory of where it opened will still land on "the other side of the fixed extreme" most of the time, simply because the alternative (closing back at a new extreme past the one already fixed) is the minority outcome by construction. Conditioning on "the extreme already happened and held" does almost all of the work; there is very little left over for a genuine reversal tendency to add. That is why the real rate and the null rate land within noise of each other -- both are measuring the same geometry, not a market behavior.

This is a specific, useful shape of false pattern to recognize: state a fact that is true (76.8%!), skip the step of asking what a null built with the same conditioning would say, and an honest base rate gets sold as an edge. It happens to be measurable and it happens to be worthless, in the same finding.

Honest limits

• Two markets only. The headline number above is measured on ES and NQ at 15-second, close-path resolution -- the only two markets with data at that granularity. A coarser 15-minute version tested on four other markets (gold, crude oil, 10-year note, EUR/USD futures) found larger, "significant" excesses in three of the four -- but that same coarse method fails to even recover a significant result on ES (p=0.28) despite ES being solidly significant on the primary test. A method that misses a known-real effect is not good evidence for or against an effect on markets it hasn't been checked on directly. Read the other four markets as unresolved, not confirmed.
• The excess is smaller in the newer years. +2.4pp in the held-out sample (2018-2026) versus +4.6pp in the period the hypothesis was built on (2010-2017) -- roughly half. That is a typical pattern for a real but modest effect and not a red flag by itself, but it means the smaller, held-out number is the one to trust going forward, not the larger fitting-sample one.
• No direction. This measures when the session's extreme prints, not which way price moves. It says nothing about whether the extreme is a high or a low, or which way price is likely to go around it.
• No trading signal. By the time Q1's extreme is confirmed to be holding, more than an hour and a half of the session has already passed, and nothing here specifies an entry. The obvious next question -- "so should you fade it?" -- is exactly the rule the previous section shows carries no edge over its own null.

Methodology

Built from the quarterly-cascade addendum (rangeprob_research, run 2026-08-26). Every RTH session (09:30-16:00 ET) is split into four 97.5-minute quarters; the "extreme" is read from the cumulative close-to-close return path, not intrabar wicks, computed identically for real days and for the null. The null is a slot-permutation resample drawn from the fitting-period donor pool, matched to preserve each time-of-day's own realized-volatility clock (including the sharp 9:30 open) -- so an excess here already has "mornings are more volatile" priced out of it, rather than just re-discovering it.

Nine related hypotheses -- spanning several window definitions and quarter lengths across the New York morning and the full regular session -- were tested together under Šidák correction for multiple comparisons (n=9, alpha=0.00568, 200 resampled seeds per market). This page reports the two members of that family that survived on the 97.5-minute regular-session window: Q1's own share of extremes, and the reversal-arc question above -- plus the seven follow-up attacks run specifically against the surviving excess, and the comparison against a 22.5-minute morning-only version that did not survive the same scrutiny.

ES and NQ continuous-contract futures, 2010-2026 (2026 is partial, read as a still-accumulating year, not a verdict input -- see the years-positive count above). The fitting sample is 2010-2017, the period the hypotheses were built on; the held-out confirmation sample is 2018-2026, disjoint from it. Every headline number on this page is the held-out one unless labelled otherwise.

This measures the timing of a session's own extreme against its own volatility-adjusted baseline. It is not a claim about profitability, and neither finding on this page implies a tradable entry, exit, or direction -- see "Honest limits" above.

See also

This page reports a narrow, heavily-attacked slice of a popular time-quartering framework that assigns fixed roles -- accumulation, manipulation, distribution -- to recursive quarters of a session, day or week. That broader framework was tested at three scales (90-minute morning quarters, six-hour quarters of the day, weekday quarters of the week) and found dead: none of its role-based, directional or hand-off claims survived correction. What is left, and what this page reports in full, is two narrow, specific facts about timing -- a small excess that seven attacks did not kill, and a "reversal rule" that looks strong until its own null gets checked.

Related measurements

• The -4.045 opening-range level — measured touch odds for a community level ladder.
• The sweep funnel is a distance curve — the same touch-curve approach applied to a different published claim.
• Hit rate is only half the story — touch odds paired with the path cost of getting there.
• Daily wrap-up — where each session's realized high, low and close get tracked as they happen.
• Graveyard — short verdicts on every claim we tested and rejected, including the rest of this framework.
• Methodology — how every number on this site is measured and reported.

Research and education, not financial advice. Independent, not affiliated with any third party.

FAQ

Does the session really print its extreme early?

More often than chance, yes, by a small margin: 45.6% of regular sessions have their defining high or low set within the first 97.5 minutes (by 11:07:30 ET), against a volatility-adjusted baseline of 43.2%. The +2.4 percentage point gap survived seven separate attacks built specifically to try to kill it.

Is this the same as the time-quartering framework in the graveyard?

No. That broader framework claims fixed roles for each quarter of a session, day or week -- accumulation, manipulation, distribution -- plus directional and reversal rules at several recursive scales. All of that was tested and found dead. What survives here is much narrower: a measured excess in how often the first quarter specifically holds the session's extreme. Nothing about roles, and nothing about direction.

Can this be traded?

No. It measures when an extreme prints, not which way price moves next, and by the time it is confirmed roughly half the session's hours are still open and undetermined. The natural next question -- "so fade the early extreme toward the close" -- is the second finding on this page, and it carries no edge over its own null.

Why doesn't "76.8% close in the opposite half" count as an edge?

Because a null built with the identical conditioning -- an extreme already fixed early, on one side -- lands in the opposite half 77.1% of the time too. Once a range's extreme is nailed down at one edge, the close has nowhere else likely to be, whether or not the market has any actual tendency to reverse. The 76.8% figure is true, and it is also close to what geometry alone would produce with no reversal tendency at all.

Why do only ES and NQ count here?

They are the only two markets with the 15-second, close-path resolution the primary test needs. A coarser 15-minute version was tried on four other markets and found larger effects in three of them, but that same coarse method also failed to recover a significant result on ES -- a market already known to be significant at full resolution. That failure means the coarse method cannot be trusted as confirmation on markets that haven't been checked directly, so those four stay unresolved rather than confirmed.