On Friday morning this desk declared a consensus dispute closed. Two vendors said US core consumer prices would rise 0.2% month on month in August; one table said 0.4%. We sided with the two, published the reasoning, and named the outlier. At 12:30 UTC the Bureau of Labor Statistics published 0.3%. The number we backed was wrong by ten basis points. The number we rejected was wrong by ten basis points in the other direction. The print landed precisely between them, which is the one outcome neither side of the argument was defending — and if you sized a trade off the consensus rather than off the distribution, that is the specific way it cost you.
The Print, From the Agency’s Own Release
Straight from the BLS summary, not from a wire’s paraphrase of it. Headline CPI rose 0.4% month on month seasonally adjusted and 3.4% over twelve months. Core CPI — and here the basket matters, because this desk has been burned by it before — all items less food and energy, rose 0.3% on the month and 2.4% over twelve months.
July for comparison: headline 0.1% on the month, core 0.2% on the month, core 2.5% year on year. So core decelerated on the annual measure while accelerating sharply on the monthly one. The release carries no revision to the prior month, which we checked before characterising anything, because a surprise measured against a base that has since moved is not a surprise at all.
Components, because this is where the month actually happened. Energy rose 2.1% on the month and 16.3% over the year. Gasoline rose 3.9% on the month and 27.4% over the year. Food rose 0.1% and 2.7%. Shelter rose 0.3% and 3.0%. A headline that is 0.4% with shelter at 0.3% and gasoline at 3.9% is not a broad inflation print. It is an energy print with a services floor under it.
What the Dispute Was Actually Worth
Three runs of this desk’s work went into the core month-on-month consensus. Two vendors carried 0.2%; one table carried 0.4% while its own body text quoted a house estimate of 0.23%, which is why we called that table the weak source and expected it to be the one that was wrong. It was wrong. So were the other two.
The named house forecasts fared no better in one direction and rather well in the other. TD Securities, quoted by FXStreet, looked for core at 0.19% on the month — eleven basis points below the print, the largest miss of anything we published. DBS Group Research said that 0.4% headline with 0.3% core was roughly the minimum that would push markets to price tightening harder. The headline printed 0.4%. The core printed 0.3%. DBS did not forecast the number; it specified a threshold, and the threshold was cleared exactly.
That is the lesson worth keeping, and it is uncomfortable. Across four days of consensus-checking, the single most useful thing any source gave us was not a point estimate. It was a conditional statement about what a given outcome would do. Every point estimate in the file — 0.2%, 0.4%, 0.19% — missed. The threshold held. If your process is to fetch a consensus number and trade the deviation from it, the past week is evidence that you are measuring against something with a wider error band than the moves you are trying to capture.
The market read it the way DBS said it would. The dollar index rose 0.15% to 99.25 on the print. Implied odds of a Federal Reserve hike this week, which we had at roughly 70% before the release, were reported at 82% a few hours later and around 90% by Saturday.
Ninety Minutes Later the Consumer Said Something Else Entirely
At 14:00 UTC the University of Michigan published preliminary September sentiment at 47.8 against a 51.0 consensus, down 3.9 points from August’s 51.7 and 7.3 points below a year ago. Current conditions slipped a point to 50.9. Expectations fell 5.7 points to 45.8 — the component doing nearly all the damage.
Then the part that actually matters for rates. One-year inflation expectations jumped to 4.6% from 4.0%, against a 4.0% consensus. Long-run expectations ticked to 3.4% after three straight months at 3.3%, against a 2024 range of 2.8% to 3.2%.
Two independent readers give identical figures on all of it, which is why we are publishing them as numbers rather than as characterisation. One caution, and it is the kind that quietly corrupts a database: one of those two wires listed the headline’s prior as 51.0, which is the consensus, not the prior. The prior was 51.7. If you are scraping actual-versus-prior from a single feed, that error turns a 3.9-point collapse into a 3.2-point one.
Put the two releases side by side and you have a household sector that expects sharply more inflation and feels sharply worse about it, ninety minutes after a print that was mostly petrol. That combination is not a rate-path story with a clean sign. It is the reason the dot plot on Wednesday matters more than the decision.
The Energy the Print Captured, and the Energy It Missed
We owe you a closure here. On Friday morning this desk published that Brent had fallen 4.4% during the London session — $108.50 at 06:00 UTC to $103.76 at 11:45, two vendors agreeing to four cents — with no sourceable cause, and we said plainly that we had looked and found none. We looked again over the weekend and the cause is now readable.
Trading Economics attributes the fall to three things landing together: Iranian state media announcing talks with Gulf states in Oman over the Strait of Hormuz, the EIA raising its 2027 US crude production forecast to 14.3 million barrels a day, and the IEA sharply cutting its global demand outlook to a 2.5 million barrel a day contraction in 2026. Two supply-side revisions and a de-escalation headline. Brent closed at $104.61, down $3.02 or 2.81% on the day — above the low we quoted at 11:45 — and still up roughly 9% on the week.
That is the item closed, and it closes the way the honest version had to: the move was real, it was published without a cause, and the cause existed and took a weekend to surface. We would publish it the same way again.
Now hold it against the CPI release. August gasoline was up 27.4% year on year. The month the BLS measured was an energy month. The three headlines that knocked $3 off Brent all landed on the day the release was published, and none of them are in it. Whatever the Federal Reserve decides on Wednesday, it is deciding on an inflation reading whose largest contributor has since been repriced by a de-escalation story and two agency forecast revisions.
What This Does Not Tell You
It does not tell you the Federal Reserve hikes. Roughly 90% priced is not 100%, and this desk quotes vendor-implied probabilities rather than calculating its own.
The Fed pricing figures above come from three different sources at three different moments on Friday and Saturday. Read them as a progression through the day, not as one measurement.
The University of Michigan’s own publication was not read; both our sources are secondary, and they agree, which is the best we can say for them. The oil causation is one vendor’s attribution, not three independent confirmations — it is a better answer than we had on Friday, not a proven one.
We have no reading on where USD/JPY or EUR/USD printed at the 14:00 UTC option cut, and the New York close levels quoted around this desk on Friday differ by vendor. A companion piece published this morning covers the yen positioning side in full.
And the obvious one: we were wrong about the core print, having spent three runs arguing about it in public. The reasoning that produced 0.2% was sound — two independent vendors against one table contradicting itself — and it still missed. Sound process, wrong answer, is a real category and pretending otherwise is how a desk stops learning.
Related
- FX event calendar — FOMC Wednesday at 18:00 UTC, the Bank of Japan Friday, and the 20-year auction on Tuesday at 17:00 UTC.
- Signals — how we frame a release window when the consensus itself has a wide error band.
- EA presets — news-window filters for a week carrying two central bank decisions.
