Everything got quiet. The day session got quieter.

The claim, as it gets made

Some version of this has been going around for a month: the day session is dead, nothing moves between the bell and the close, and meanwhile Globex runs wherever it likes overnight. It is the kind of claim that is easy to feel and worth measuring, because the fix depends entirely on which half is true.

It splits cleanly. One half holds. The other is backwards.

Everything is quiet

Range normalised to percent of the session open — NQ has roughly doubled since 2021, so a points-based percentile misranks a multi-year sample. Recent is the last 15 sessions; the baseline is the prior five years, 1,289 sessions.

The day session really is compressed — 14th percentile of five years is not a mood. But the overnight range is at the 30th percentile. It is below normal too. Globex is not roaming more in absolute terms; it is simply giving up less than the day is.

Nothing expanded. The day session just contracted faster than the night.

That distinction matters because “the overnight is wild” and “the day has stopped participating” call for opposite responses. Only the second one is happening.

The number that actually moved

Take each session and ask a yes/no question: did the day range finish entirely inside the high and low already set overnight? Not near them — inside both, taking out neither side.

In a normal five-year stretch that happens 18.0% of the time. Over the last 15 sessions it happened 53.3% of the time — 2.96× the base rate. The day session is repeatedly failing to reach either edge of a range that was set before it opened.

We ran the same 21-session calendar window — late August into mid-September — for every year we have. It has no precedent.

The only year that comes close is 2015 at 46.7% — and 2015 was a high-volatility tape, the aftermath of an August shock. Containment in a violent market is a different object: the overnight gaps so far that the day cannot catch it. Containment this high in a quiet market has not happened in this sample.

Three explanations, and none of them survive

Every candidate below makes a prediction specific enough to fail. The base rates are built from 2011–2025 only, so the period being explained never votes on its own explanation.

1. Options positioning and expiry

This drop went out three days before a quarterly expiry, which makes the timing suggestive. It does not survive the base rate: quarterly expiry weeks are indistinguishable from any other week.

18.2% against 18.7%. There is no expiry effect here to explain a 53% reading.

2. The contract roll

NQ did roll to the December contract during this window, so the coincidence is real. The effect points the wrong way: roll weeks are historically wider and less contained than ordinary weeks.

If the roll did this, compression would show up as a narrower and more contained roll week. It is neither. An explanation that predicts the opposite of what happened is not a weak explanation; it is a rejected one.

3. September seasonality

The most quoted of the three, and the most clearly wrong. Across 2011–2025 this calendar window is unremarkable, and September carries the lowest containment rate of any month.

September is the least contained month of the twelve. Seasonality does not predict this regime — it argues against it.

The control that splits the problem in two

Running the identical measurement on the S&P contract is the most useful thing in this drop, because it separates two effects that look like one.

The range compression is market-wide. Both contracts sit at the 14th percentile of their own day-range history. That is the same number twice, and it means nothing about this is a Nasdaq story at the range level — it is a broad volatility floor that both are resting on.

The abandonment of the day session is not. The S&P is at 1.19× its own containment base rate — barely above normal — and its day share of volume hardly moved. The Nasdaq is at 2.96× with its volume share at the 14th percentile. Both markets got quiet. Only one of them emptied out between the bell and the close.

Around the clock

Split the 23-hour session three ways and the “Globex roams” half of the claim resolves into something much more specific. Median range in points, measured the same way on both sides of the comparison.

Asia is flat. Europe is the only window in the day that expanded, and its range now runs at 71% of New York’s against a 44% norm — the 82nd percentile of the trailing year. That is the correction to “Globex roams freely”: the overnight’s relative liveliness is almost entirely a European-hours phenomenon, which is where a book gets repositioned when the mover is macro rather than single-stock.

Inside the day, the open still works

This is the finding that explains why the tape feels worse than a 25% range loss should feel.

Split the day range into what the opening thirty minutes builds and what the remaining six hours add on top of it.

The opening range is down 19% — quieter, but roughly in line with the broad volatility floor everything is sitting on. The range added after the first half hour is down 44%. The open is still doing most of its normal job. The rest of the day has stopped adding to it.

The open still pays. The continuation is what died.

That is why the opening range now accounts for 62% of the whole day range against a 51% norm. Not because the open got bigger — it got smaller — but because everything after it stopped contributing.

What we keep

Three things, none of them a trade.

• The overnight range is functioning as the day’s frame. On more than half of recent sessions the day never reached either edge of it. Those levels are behaving like boundaries, not like targets.
• Range is built early and then not extended. A 44% collapse in post-10:00 expansion against a 19% drop in the opening range is a different market from one that is simply quiet all day.
• Position size keyed to points is now wrong by a quarter. A stop that was correctly scaled a month ago sits a much larger fraction of the day’s likely range away, because the range moved and the stop did not. This is the practical consequence and it does not require predicting anything.

And one caution worth more than the three: this is 14 sessions. Regimes that show up at 2.96× a base rate are, by construction, unusual, and unusual things revert. Nothing here says how long it lasts or what ends it.

What this does not test

This matters more than the result, so it gets its own section.

• Any expectancy claim. No entries, no exits, no barriers, no expectancy after costs. We measured the shape of sessions, not the profitability of anything done inside them.
• Whether it persists. A 14-session window describes the present. It carries no forecast, and we ran no test of one.
• Dealer positioning as a cause. The obvious story is that dealers are long gamma and pinning the tape. The book on the day of publication did not support it — spot sat between two different measures of the gamma flip, and the largest concentration near spot was put-dominated, which is a hedge stack rather than a pin. We hold dealer positioning as a modifier of realised volatility, not a magnet for price, and we did not test it as a cause here.
• Macro attribution. Europe expanded. We did not test why, and the candidates — rates, currency, a data calendar that has shifted its weight into European hours — are not separated by anything in this drop.
• Other markets. Two index futures. The split we found between market-wide compression and instrument-specific abandonment may not generalise past them.
• Formal significance. The three rejected explanations are base-rate comparisons on large samples, not walk-forward tests with a permutation null. They are strong enough to reject an explanation that predicts the opposite of what happened; they are not a pre-registered kill in the sense the other drops in this series use the word.