The Bank of Japan published August corporate goods prices at 23:50 UTC last night, four hundred minutes after this desk published an article built on the expectation. The consensus was +7.4% year on year. The print was +7.6%. That looks like a beat, and every wire wrote it as one. Read the Bank’s own release and it is also a slowdown: July was revised up from +7.2% to +7.7%, and the month itself came in at −0.2%. The number went up against the forecast and down against the economy at the same time, because the base moved underneath it.
A beat and a deceleration are the same print here
The arithmetic is worth doing slowly, because it is the whole story. A consensus of +7.4% was formed against a published July of +7.2%. On that base, 7.6% is an acceleration of four tenths and a two-tenth beat. But the Bank of Japan revised July to +7.7% in the same release that printed August at +7.6%. Against the revised base, August is a deceleration of a tenth. The NLI Research Institute puts the month-on-month at −0.2% and calls it a twelve-month low; the Bank’s own release carries the same −0.2% against a July that was itself revised from +0.1% to +0.4%.
So three things are simultaneously true. The print beat the survey. Japanese producer price inflation decelerated. And the survey was wrong mostly because it was anchored to a July figure that no longer exists. If your system trades the surprise — actual minus consensus — you took a long-dollar signal off a series whose previous value was restated half a point higher in the same press release. That is not a rare event in Japanese data and it is not an argument for ignoring the surprise. It is an argument for knowing which of the two numbers in front of you was revised.
The upward revision to July is the one fact here we could not source to the Bank’s own PDF directly — the release states the August figures and the revised July month-on-month, and the +7.2% to +7.7% year-on-year revision comes from the NLI Research Institute’s commentary and from InvestingLive, which give the same pair of numbers independently. Two sources, both read in full, neither of them the issuing agency on that specific line.
Copper did more of it than crude, and we said otherwise six hours ago
Yesterday evening this desk published a piece whose argument was that Japanese producer prices running two points hotter than American ones was an energy-import asymmetry, and cited our own figure that Japan sources 95.9% of its crude from the Middle East. The component breakdown now says that was, at best, third in line.
NLI’s attribution of the year-on-year rise: non-ferrous metals contributed +1.67 percentage points, running at +43.3% year on year, with copper for data centres at +59.5%. Chemical products contributed +1.07 points. Petroleum and coal products contributed +0.96 points. The largest single driver of Japanese producer price inflation in August was not the Strait of Hormuz. It was artificial-intelligence buildout bidding for copper.
We are not retracting the oil channel — a point of contribution is a real point, and the August data closed before the crude move that ran Brent from the high nineties to the high hundreds this week, so the energy contribution should be larger in September, not smaller. What we are retracting is the ranking. If you built an exposure rule this week that treats Japanese cost-push inflation as an oil proxy, the print says the correlation you want is to copper at least as much as to Brent, and that is a different hedge with a different beta.
The single best number in the release: 24.8 against 16.7
Everything above is commentary on a headline. The release also contains a figure that is not a headline anywhere and that answers, arithmetically, a question this desk has been circling for a week — how much of Japan’s imported inflation is the commodity and how much is the currency.
The Bank publishes the import price index on two bases. On a yen basis, all commodities ran +24.8% year on year in August. On a contract-currency basis — the same goods, priced in the currency they were actually invoiced in — the figure is +16.7%. The gap is 8.1 percentage points, and that gap is the exchange rate. The export side says the same thing in the other direction: +17.9% on a yen basis against +11.1% on a contract-currency basis, a 6.8-point wedge.
That is the cleanest available answer to what a weak yen costs Japan, and it is published monthly by the institution that sets the policy rate. Roughly a third of Japan’s imported goods inflation is currency, not commodity. Note also NLI’s month-on-month: import prices fell 3.0% on the month on yen strength — the five-session yen rally we spent all week writing about is already in this data, reducing the import bill, in a month whose headline still printed a beat.
Britain doubled its GDP consensus at 06:00 UTC and cable moved three hundredths of a percent
The Office for National Statistics released July monthly GDP at 06:00 UTC. Consensus was 0.0% month on month. The print was +0.4%, with the prior month at +0.3%. The year-on-year came in at +1.6% against a +1.0% expectation, and the three-month-on-three-month at +0.4% — an eighth consecutive three-month period of growth, on InvestingLive’s count.
A monthly GDP print that comes in four tenths above a flat consensus, and six tenths above on the year, is a large data surprise by any normal scaling. FXStreet had GBP/USD at 1.3515 shortly afterwards, up 0.03% on the day. Three hundredths of one percent.
The composition is the reason to be careful before calling it a strong number: services carried it at +0.6% three-month-on-three-month, while production ran −0.5% and construction −0.5% over the same window. FXStreet separately carried July industrial production at +0.2% month on month and manufacturing at +0.9%, which is not a contradiction — different horizons on different bases — but it does mean the monthly and the rolling quarter are telling opposite stories about British industry. The pound’s non-reaction is the market declining to treat a services-driven monthly beat as new information about the Bank of England.
The 12:30 consensus is contested in three places, and the disagreement is the trade
US August CPI lands at 12:30 UTC. We have now read three separate published consensus sets, and they do not agree on either of the two numbers that matter.
- Headline year on year. TradingKey gives 3.4%, unchanged from July. FactSet, collecting four estimates, projects 3.3% with a range of 3.30% to 3.50% — and notes the 20-basis-point spread is narrower than the 42-point trailing twelve-month average. So the vendors are a tenth apart on the headline before a single estimate is wrong.
- Core month on month. TradingKey gives +0.2%. Kiplinger’s preview table gives +0.4%, which is identical to its headline figure and looks like a duplication — its own body text quotes a Wells Fargo estimate at 0.23%. We flagged that internal inconsistency yesterday and it has not been corrected.
- Core year on year. Here they agree: 2.4%, down from July’s 2.5%. TradingKey and FactSet both.
July’s actuals, for the base: +0.1% and 3.4% headline, +0.2% and 2.5% core. Yesterday’s PPI is the reason nobody trusts the surveys — it printed +0.4% month on month and +5.4% year on year against a +5.3% consensus, a tenth hot on the year, and the Bureau of Labor Statistics attributed over three-fourths of the broad-based rise to energy, with diesel alone up 24.1%. A consensus formed before a 4% to 6% single-session crude move will be low in a predictable direction.
The practical consequence for a system is narrower than it sounds. When the consensus itself has a tenth of dispersion, a one-tenth surprise is not a surprise — it is inside the measurement error of the thing you are measuring against. Whatever threshold your news filter uses to decide a print was a shock, today it should be two tenths, not one, and it should be computed against whichever vendor actually feeds your system rather than whichever one you read this morning.
What this does not tell you
It does not tell you whether the Bank of Japan hikes. Markets are reported as having a move to 1.25% from 1.0% near fully priced; the Bank’s own 2026 schedule, which we read directly, puts the meeting on 17 and 18 September, with the decision on the Friday. InvestingLive’s write-up of the CGPI print described it as a “September 16–17 policy meeting”, which does not match the Bank’s published calendar. We are going with the Bank.
It does not tell you the July year-on-year revision to +7.7% with the authority of a primary source. That specific line came from two secondary readers of the release rather than from the release text we could extract.
It does not give you the ONS bulletin itself. The ONS release page confirmed a 07:00 London publication, but the bulletin URL we fetched served the June edition; the July figures above are from FXStreet and InvestingLive, which agree with each other on the headline, the prior and the year-on-year. Bloomberg and the Japan Times both carried the Japanese print and neither was fetched — one is paywalled and the other we do not cite from headlines.
And it does not tell you what CPI will be. Nothing here is a forecast of the 12:30 number. It is an argument about how wide your definition of a surprise should be when the forecasters disagree with each other by as much as they are likely to disagree with the data.
