Twice this morning this desk published that nobody — not a vendor calendar, not the issuing agency — had stated an hour for today’s euro area final inflation release. Both pieces turned that absence into an argument about calendar metadata. The release came out at 09:00 UTC. One of the calendars we did not read had printed that hour in advance, the issuer’s own page metadata carries exactly 09:00:00Z, and the first reaction piece we can find is stamped 09:00:49Z. We were wrong about the hour, and the way we were wrong is worth more than the hour is: we built a negative claim out of the four sources we happened to open. Meanwhile the release itself did something more interesting than confirming the headline. It confirmed the headline and revised two of the four components.
3.8 confirmed, 2.5 core confirmed, and the revision is entirely underneath
Eurostat’s final reads 3.8 per cent annual for September, up from 3.2 in August, with a monthly rate of 0.6. The flash on 2 October said 3.8 and 0.6. Core, which the flash release does not break out and a secondary reader supplies, is confirmed at 2.5 against 2.4 in August. On the headline, then, nothing moved and nothing was ever going to: a final is a revision, not a surprise, which is the framing both of this morning’s articles used.
The components tell a different story. Energy went from 18.8 per cent in the flash to 18.9 in the final. Food, alcohol and tobacco went from 1.4 to 1.3. Services held at 3.2 and non-energy industrial goods at 1.1. Two of four moved, each by a tenth, in opposite directions, and the headline is unchanged to one decimal because the two tenths cancel against very different weights.
The final also publishes the contribution breakdown the flash does not: energy 1.71 percentage points, services 1.52, non-energy industrial goods 0.29, food, alcohol and tobacco 0.25. Those sum to 3.77, which rounds to the published 3.8 — an internal consistency check the document hands you for nothing and that no secondary reader in our sweep reproduced.
So the component with the biggest contribution is the one that was revised up, and against August it is now 4.6 percentage points higher rather than the 4.5 the flash implied. If your model of euro area inflation is a model of energy, your September just got marginally worse, and you would not know it from any headline published today.
We said nobody published the hour. One calendar did, and we had not opened it.
This is the correction, and we are not going to bury it in a humility section.
At 06:18 and 06:19 UTC this desk published two articles, each of which stated that the calendar carrying today’s euro area data gave no time column and that Eurostat’s own flash release named a date and no hour. Both of those statements remain true about those two documents. The conclusion we hung on them — that the release could not be timed from published sources — does not survive contact with a third calendar, which listed the final euro area inflation for September at 09:00 GMT with the core forecast of 2.5 per cent attached.
Three independent confirmations now exist for that hour: the vendor calendar in advance, the issuer’s page metadata at 09:00:00Z, and a reader’s reaction piece at 09:00:49Z. The hour was knowable. We did not know it because we read four sources and treated the fourth absence as a property of the world.
The generalisation is the useful part, and it is uncomfortable because this desk has made the same move repeatedly: a negative claim about publication is a claim about your own sweep until you have checked a source that would have carried it. “No calendar states the time” needs a list of the calendars checked, in print, or it is not a finding. Every negative claim in today’s other article carries that list for exactly this reason.
Three documents, three different answers to the same metadata question
Having been wrong about one of them, the comparison is still worth making, because the three documents in front of us this morning fail and succeed in three distinct ways.
The Bureau of Labor Statistics release calendar states the date, states the time, and states the timezone rule for the whole page in a single line: all times on the calendar are Eastern Time. Today’s import and export price indexes are listed there at 8:30 a.m. Eastern — 12:30 UTC — with the next release of the same series on 17 November at the same hour. That is the complete shape. Nothing is left for a reader to infer.
Eurostat’s release states the date and the next release date, 18 November, and separately gives 4 November as the next flash covering October. It states no hour anywhere in its text. The hour exists only in the page’s machine-readable timestamp, which is to say the issuer published the time to machines and not to readers.
And the week-ahead calendar we read yesterday declares a timezone in its header and then carries no time column at all — the only one of the three that supplies the metadata and withholds the data it describes.
The ranking is obvious and the lesson is narrower than “calendars are bad”: if the issuing agency publishes a schedule page, read the schedule page. The agency that does this properly does it for every release on the page, and it costs one fetch to have the whole month rather than one event.
We looked for a margin-of-error column. Three calendars, none has one.
Yesterday this desk published that a retail sales beat of four tenths sat inside the issuer’s own stated interval of plus or minus four tenths, and asked a falsifiable question: does any vendor calendar carry the margin of error at all?
Today was the natural test, because a final estimate is the one release where the revision is reliably smaller than any plausible interval. We checked three calendars and a release-schedule page. None of the four carries an uncertainty column, an interval, or any field describing the precision of a figure. Every one of them publishes previous, consensus and actual as three bare numbers.
Eurostat does not help here either: its HICP release carries no confidence interval, which is defensible for a price index built from administered collection rather than a sample survey, and which is a different situation from the sampled American releases where the issuer does publish one. So today’s negative is clean on the vendors and inapplicable to the issuer.
The honest statement of the result, with the sweep named rather than implied: of the three vendor calendars and one agency schedule this desk read on 16 October, none published a margin of error. Not “no calendar does”. We made that mistake once already today.
What is left on today’s board, in UTC
Import and export price indexes at 12:30, confirmed at the issuer rather than at a calendar. Industrial production and capacity utilisation at 13:15. The New York option cut at 14:00. The Commitments of Traders report as of the 13 October close at 19:30, and Treasury international capital data at 20:00.
Both of the morning releases are second tier, and the honest framing for a release window is the one both of this morning’s pieces reached independently: when the information content is low and the clustering is tight, treat the window as a liquidity event rather than an information event. Nothing in a final inflation print or an import price index changes a rate path. What they change is the depth of the book for a few minutes, and a filter that sizes on expected surprise will size these wrong in the direction of too large.
The one thing on the list that will actually settle something is the 19:30 positioning report, and it is measured at the 13 October close — before this week’s inflation print. The first snapshot containing the reaction is still a week out.
What this does not tell you
The 09:00 release hour is established by three mutually supporting pieces of evidence, none of which is a sentence in the issuer’s own release text saying so. If Eurostat states its publication hour somewhere we did not look, that is one more source we did not open, and the correction above would then be understated rather than overstated.
The core figure of 2.5 per cent comes from one secondary reader; the issuer’s release as we read it gives the headline and the four component rates and does not break out a core aggregate. We have labelled it accordingly rather than presenting it alongside the component figures as though it had the same standing.
The component revisions are each one tenth, which is the smallest move a one-decimal series can make. We are reporting them because the direction is informative and because nobody else published them today, not because a tenth on food prices is a trade.
And we have not established why the two components moved. A final estimate incorporates late national data; which countries revised, and in what direction, is in the detail tables and we have not read them.
Related
- Event calendar — today’s remaining releases in UTC, with the issuer-confirmed times.
- Signals — how we treat a release whose revision is smaller than its own precision.
- EA presets — news-window settings for a low-information release block.
- Two Level Sets, One Cross, Sixty-Seven Minutes, Nothing in Common — one of the two pieces this article corrects.
- We Argued 1.1 Versus 1.2. The Issuer Says 1.3. — where the margin-of-error question was opened.
