Two statistical agencies published six and a half hours apart overnight — Japan’s September national CPI at 23:30 UTC on 22 October, Britain’s September retail sales at 06:00 UTC this morning — and between them they revised six of the seven prior figures they carried. Every revision ran upward. In Japan that makes the acceleration smaller than it reads: core inflation printed 2.1 per cent against a prior most readers still publish as 1.7, but the publisher that carries the revision puts the prior at 1.8, so the month-on-month step is 0.3 points rather than 0.4 — the published-prior version overstates it by 33.3 per cent. In Britain it makes the fall bigger: all four fields missed consensus, by an average of 0.45 points, and all four August bases were marked up, so the monthly swing in the headline is 1.1 points against the 0.9 you get from the unrevised number. This desk has a standing rule to check whether the prior was revised. It fired twice in one morning, in opposite directions, and in both releases the field with the largest surprise is a field no headline carries.
Japan: the field that beat is the one nobody headlines
Three fields, three different verdicts against consensus, and the ranking is the reverse of the attention each gets.
- All items (headline): 2.1 per cent year on year, consensus 2.1, prior 1.9. In line. This is the number in the wire headline.
- All items less fresh food (“core”): 2.1 per cent, consensus 2.1, prior 1.7 as published. In line. This is the number the Bank of Japan’s 2 per cent target is read against, and the first reading above it since December on one reader.
- All items less fresh food and energy (“core-core”): 2.4 per cent, consensus 2.3, prior 1.9 as published. Beat by 0.1. This is the only field in the release that surprised, and it is the one that appears in no headline we read.
That is the third time in four days this desk has sized for an event whose outcome was decided by which number the vendor put in the headline slot — Jakarta on the deposit facility, Ankara on three rates in parallel, Thursday’s US claims print where the initial figure beat and continuing claims deteriorated. We are not calling four cases a law. We are saying that a release with three fields has three surprises, and reading one of them is a choice you are making whether or not you know it.
The practical form of that for a system trader is narrow: if your news filter keys on a single value for “Japan CPI”, the release you just traded registered as a non-event, because the two fields a single-value feed is most likely to carry both came in exactly on consensus. The 0.1-point beat sat in the third field.
And the prior moved underneath it — twice, and only one reader says so
Here is the measurement problem, and it is a disagreement between readers rather than a mistake by any of them. Three of the four publishers in this sweep give the August core prior as 1.7 per cent and the core-core prior as 1.9. The fourth gives 1.8 and 2.0, and states both as revisions from 1.7 and 1.9.
Work the arithmetic both ways, because the conclusion changes:
- Core, published prior: 1.7 to 2.1 is a 0.4-point acceleration in one month.
- Core, revised prior: 1.8 to 2.1 is 0.3 — the first version is 33.3 per cent larger.
- Core-core, published prior: 1.9 to 2.4 is 0.5.
- Core-core, revised prior: 2.0 to 2.4 is 0.4 — 25.0 per cent larger.
Neither version changes the level, and the level is what the central bank talks about. Both versions have core above 2 per cent and core-core at 2.4. But the step is what gets written up as momentum, and a third of the momentum in the core number is a vintage artefact of which prior your source happens to carry. One reader has it; three do not; we could not reach the issuing agency’s own September release to settle it, only its release calendar, which confirms the 23 October publication date and the monthly cadence and nothing about the figures.
So the revision is single-sourced and labelled as such, and the honest statement is that the acceleration is either 0.4 or 0.3 points depending on a 0.1-point revision only one of four readers reports.
Britain, six and a half hours later: four fields, four misses, four upward revisions
The retail sales release at 06:00 UTC is the cleaner case, because it is unanimous in both directions at once. Every field missed and every base was marked up.
- Headline, month on month: −0.4 per cent against −0.1 expected — a 0.3-point miss. August revised from +0.5 to +0.7.
- Ex auto fuel, month on month: −0.7 against −0.1 — a 0.6-point miss. August revised from +0.6 to +0.9.
- Headline, year on year: 0.8 against 1.0 — a 0.2-point miss. August revised from 2.4 to 2.7.
- Ex auto fuel, year on year: 0.8 against 1.5 — a 0.7-point miss, the largest in the release. August revised from 2.7 to 2.9.
Four of four fields below consensus, mean absolute miss 0.45 points, largest 0.7. Four of four priors revised up, total 1.0 point, mean 0.25. And the same structural point as Japan, from the other side: the biggest surprise is in the ex-fuel year-on-year field, which no headline leads with, while the headline month-on-month field had the smallest miss of the four.
The revisions compound the direction rather than offsetting it. On the revised base the headline monthly swing is +0.7 to −0.4, which is 1.1 points; on the unrevised base it is 0.9. The ex-fuel swing is 1.6 revised against 1.3 published, 23.1 per cent larger. In Japan the revision made the news smaller; here it makes the news bigger. A revision has no sign preference and checking it is not a conservative habit — it cuts whichever way the agency’s back data went.
One caution on the source, stated because this desk has been caught by labelling before: the reader we used writes “weak UK Retail Sales data for August” in its market-reaction paragraph while its own table is September. The figures are September; the sentence is wrong. We used the table.
What our own carried figure pair turned out to be
This desk has carried an open item for seven slots: a Japan core CPI figure pair recorded as “+2.1 against +1.7” with no named basket, flagged repeatedly as needing the issuer to settle it. It is settled now, and it was simpler and more embarrassing than the versions we were entertaining.
The 2.1 was the consensus. The 1.7 was the August prior. They were never two competing forecasts, which is the shape this desk kept half-treating them as, and the basket — unnamed for seven slots — is the Statistics Bureau’s all-items-less-fresh-food series, the one conventionally called core in Japan and the one that excludes fresh food only, not energy. Consensus was met exactly. The 1.7 is the figure three readers still publish and one revises to 1.8.
The reason this is worth four paragraphs rather than a line in a log is the failure mode it demonstrates. A pair of numbers carried in a file without field labels will get read as whatever shape the reader is looking for, and this desk spends a great deal of effort on vendors mislabelling fields while carrying an unlabelled pair of its own for a week. Every figure we carry forward needs its field, its basket and its vintage attached at the moment it is written down, not at the moment somebody asks. The item is closed.
What this does not tell you
We did not reach either issuing agency’s own release. Japan’s Statistics Bureau gave us its release calendar and nothing else; Britain’s bulletin was not opened. Every figure in this note is a secondary reading, and the revisions in particular rest on one publisher in the Japanese case and one in the British case.
We have no consensus figure for Japanese core-core from a second reader, so the 0.1-point beat — the only surprise in that release — is single-sourced. We have no seasonal-adjustment detail on either release and are not offering adjusted-versus-unadjusted as an explanation of anything; that hypothesis has cost this desk a published retraction before.
Nothing here says what either release does to its currency. The yen was firmer after the Japanese print on three readers and sterling was marginally lower after the British one on one, and in both cases the session was already moving on other things — a 17-basis-point drop in the US ten-year, equity weakness tied to AI capital spending, and French fiscal politics. Attributing a currency move to a release that landed in the middle of all three is not something this note attempts.
And the policy inference is not ours. One reader says the data cements a December Bank of Japan hike to 1.50 per cent and that some analysts see a terminal rate above 2; another reports the 29–30 October meeting is widely expected to hold at 1.25 and that the bank may raise its inflation forecasts. We have no probability figures from any feed, so we publish none.
Related
- FX events calendar — the European flash PMI cluster from 07:15, and 13:45 US flash PMIs, in UTC.
- Signals — why a release with three fields has three surprises.
- EA presets — news filters that key on a single value for a multi-field release.
