Nine CPI releases, measured at the exact hour

US CPI is released at 8:30am Eastern. That is 13:00 UTC when New York is on standard time and 12:00 UTC when it is on daylight time, so the release hour changes with the clocks. We pinned each one rather than guessing, which matters: our first pass tried to find the release hour by looking for the busiest hour of the morning, and it kept picking 15:00 UTC, which is US equity trading, not the print.

ReleaseData forDayRelease hourHour moveHour volumevs average hour
Dec 18, 2025Nov 20250.84% lower13:00+1.82%$229M2.5x
Jan 13, 2026Dec 2025+4.51%13:000.04% lower$123M1.4x
Feb 13, 2026Jan 2026+3.90%13:00+0.28%$109M1.9x
Mar 11, 2026Feb 2026+0.35%12:00+0.27%$72M0.9x
Apr 10, 2026Mar 2026+1.64%12:00+0.18%$53M1.0x
May 12, 2026Apr 20261.52% lower12:00+0.12%$47M1.1x
Jun 10, 2026May 20260.35% lower12:00+0.87%$82M2.0x
Jul 14, 2026Jun 2026+4.35%12:00+1.77%$197M3.6x
Aug 12, 2026Jul 20260.19% lower12:000.19% lower$83M2.3x

Three things fall out of that table

The spike is smaller than anyone would guess. Mean 1.9x an average hour, and the release hour carries only 7.8% of the day's volume. On March 11 the CPI hour traded 0.9x, which means the most anticipated scheduled number of the month produced less activity than a random hour of the same day. Two others came in at 1.0x and 1.1x.

The hour does not tell you where the day ends. Release hour direction matched the day's close on five of nine. The days themselves closed higher on five of nine. Two coin flips stacked on each other.

The tape barely changes composition. Mean buy share in the release hour was 52.39% against 50.67% for the whole day. Under two points of difference in the hour that supposedly repriced the Federal Reserve.

What is real is size. Mean absolute move on a CPI day was 1.96% against a 1.62% baseline across the last 300 sessions. So these days run wider than normal. They are just not wider in a direction you can know in advance.

Then Thursday happened

The August PPI landed at 8:30am Eastern on Thursday September 10, the 12:00 UTC hour. Set it against the nine.

MeasureThursday's PPI hourCPI hours, meanCPI hours, most extreme
Volume vs an average hour4.8x1.9x3.6x
Share of the day's volume20.1%7.8%15.7%
Taker buy share29.83%52.39%44.81%
Hour move1.21% lower+0.57%+1.82%

Every column is an outlier. The hour traded $223 million, more than any CPI release hour in the sample, and carried a fifth of the day's total volume. Its buy share of 29.83% means aggressive sellers took just over 70% of it, and no CPI hour we measured came within fourteen points of that.

The session closed 1.50% lower at $77,129 after touching $76,676. Open interest fell 1.43%, so contracts were being closed rather than opened. That is positions coming off, not a new short being built.

The two hours before were stranger than the release

Hour, UTCMoveVolumeBuy share
09:000.15% lower$47M25.54%
10:000.17% lower$48M25.58%
11:000.03% lower$51M37.12%
12:00, release1.21% lower$223M29.83%

Two hours before the number existed, aggressive sellers were taking 74% of the tape. That is more one sided than the release hour itself.

We are going to report that and stop. Those hours ran on about a fifth of the release hour's volume, and a thin hour produces extreme ratios easily. We cannot see who was selling, whether it was one desk unwinding or a hundred small accounts, and anyone telling you it was informed positioning is guessing. It is a strange reading on a small sample, which is exactly the kind of thing that gets over explained.

A correction, and the streak is now seven

First the correction. In Wednesday's article we wrote that "US inflation data is due Thursday September 10". That was the PPI. The CPI, which is the number most readers had in mind, is Friday September 11. Our wording was technically true and practically misleading, and the two prints are a day apart for a reason worth knowing.

Second, the streak. Thursday made it seven consecutive sessions with a taker buy share below 50%, and at 42.20% it is the lowest of the run.

SessionCloseBuy share
Fri Sep 4$79,66148.36%
Sat Sep 5$79,83246.67%
Sun Sep 6$80,34247.58%
Mon Sep 7$79,11249.00%
Tue Sep 8$78,45644.42%
Wed Sep 9$78,30646.86%
Thu Sep 10$77,12942.20%

On Wednesday we tested whether a streak like this predicts a bounce, found that the only correctly computed version of the test gave a negative average five session return, and said there was nothing worth trading. Two sessions later the price is lower.

That is consistent with what we published and it is not evidence for it. Two sessions is not a sample. If price had risen 4% we would be writing the same sentence with the sign flipped, and we would rather say that now than after the fact.

What we are watching at 12:30 UTC

We are not forecasting the number or the reaction. But the nine releases above give a specific thing to check, and it takes one glance at the release hour.

If the CPI hour looks like the nine: roughly 1.9 times an average hour, a buy share near 52%, and about 8% of the day's volume. That would be the eleventh straight macro print that the market processed without a crowd, and it would strengthen a claim we have now made from two directions, the other being the nine CLARITY milestones that traded $864 million against an $863 million baseline.

If it looks like Thursday: four times or more a normal hour with a buy share in the twenties or thirties, then two consecutive prints have broken the pattern, and the pattern is the thing that needs rewriting, not the news. We will write that.

Either way the FOMC meets September 15 and 16, with the Senate's CLARITY cloture vote at 2:15pm Eastern on the 15th, so next week stacks two more dated events on top of this one.