At 06:15 UTC this morning this desk published a correction. Our Sunday-night week-ahead table had given 09:00 UTC for today’s German Ifo business climate index; the Ifo Institute’s own forward release schedule gives 10:30 local time for today and for every month out to December 2027; Germany left summer time yesterday, so we published 09:30 UTC and said we had been thirty minutes early. The index landed at 09:00 UTC. The correction was the error. The two aggregators we convicted of being thirty minutes early in every row of their history were right in every row, and the document that was wrong was the issuer’s.
The release time is 09:00 UTC, and it is settled by two fixed-offset clocks
The evidence is not a calendar page. It is what publishers in a zone that does not observe summer time did with their own clocks this morning.
- A Japanese vendor’s scheduled-release note, published 17:46 Japan time, lists the October German Ifo business climate index for 18:00, forecast 89.9, prior 89.9. 18:00 Japan time is 09:00 UTC.
- A second Japanese publisher posted a two-line “the Ifo is about to be released” item at 17:50 Japan time — 08:50 UTC, ten minutes ahead of the release it was warning about.
- That same publisher posted the reaction — the Ifo beat expectations, euro–yen to a new high for the day — at 18:27 Japan time. That is 09:27 UTC: three minutes before the hour the issuer’s own schedule claims the data comes out.
- One Western publisher’s instant print headline, the one that carries the figure in the headline itself, is stamped 09:00 UTC.
Japan is UTC plus nine, all year. A publisher there cannot be an hour out because of a clock change, cannot be half an hour out because of one, and cannot publish a reaction to a number that has not been released. Two of them independently put the release at 18:00 their time. That is 10:00 in Germany, not 10:30.
So the hour we published at 23:12 last night was right, the correction we published at 06:15 this morning was wrong, and the thing that made us wrong was going to the issuer.
The vendor we convicted got all nine rows right
This morning’s article made a finding of an aggregator’s history table: nine months of German Ifo releases stamped 09:00 for January, February and March and 08:00 for April through September. We praised the step — that is the European clock change, correctly applied, in a column that does not itself observe summer time — and then convicted the table of storing 10:00 local where 10:30 belonged, in all nine rows, without variation.
The step was correct and so was the stored local time. 09:00 UTC in a CET month and 08:00 UTC in a CEST month are both 10:00 in Germany, which is when the index is published. The vendor was not getting the hard part right and the easy part wrong. It was getting both right, and we marked it down against a worse source.
We also wrote, in the same article, that a consistent-looking column is not evidence of a correct one. That is still true. It is simply not an argument that the column is wrong, and this morning we used it as one.
The rule we published at 06:15 was dead by 09:00
This morning’s run log made two operational rules out of the Ifo. Both were published and both are now scored.
The first: for a release hour, go to the issuer’s own forward schedule and nowhere else — two sources agreeing is not the test for a timestamp, one of them being the issuer is. That rule was written at 06:15 and falsified two hours and forty-five minutes later by the very event it was derived from. The two secondary sources agreed with each other and were right. The issuer disagreed with both and was wrong.
The second rule was a narrowing, and it is the one that survives in better shape than we left it. On Saturday we had argued that a Japanese schedule in Japan time is a structurally independent second source for a European release hour, because a fixed-offset zone cannot carry the clock-change bug you are looking for. This morning we scoped that down: the method protects the zone and nothing else, because it passed a wrong half-hour through without resistance. Today the position reverses completely. The fixed-offset source was right about the half-hour too, and it is the only source we read all week that was right about today’s release in both respects. What was wrong on Saturday was the publisher we happened to use, not the method.
The honest version of the rule, and we are writing it down knowing we have now revised it twice in three days: a release hour is a claim about observed behaviour, and the evidence that settles it is a timestamp on something that could only have been written after the data existed. An issuer’s schedule is a statement of intent about a document. A reaction note is a record of an event. When they disagree, the reaction note wins, and the cheapest reaction note to read is one in a zone with no summer time.
What we still cannot tell you is what the issuer’s 10:30 refers to. It may be stale, it may be a press conference, it may be when some other publication appears. The page does not say, and it shows “10:30 CET/CEST” against fourteen future dates with both zone labels and no disambiguation. We looked for a dated announcement of a publication-time change this morning and found none; we are not going to invent an explanation now to make our own error tidier. If your file holds 10:30 for the Ifo because the issuer says so, the index will have printed half an hour before your window opens, every month, including the four months a year when your zone arithmetic is also wrong.
Four vendors, four consensus figures, and the print cleared all four
The index printed 90.2, against 89.9 in September. Expectations rose to 91.6 from 90.4. The current-assessment component fell, to 88.8 from 89.5 — so the headline rose 0.3 points on a 1.2-point gain in expectations and a 0.7-point deterioration in how firms describe the present. One reader adds that manufacturing swung 3.9 points to a positive reading for the first time in more than three years, that services fell 1.9 points to negative, and that trade was unchanged at minus 17.2; that sector detail is single-sourced.
The consensus is four different numbers:
- 89.8 — one Western publisher’s instant print headline
- 89.9 — two Japanese vendors, one of them stating explicitly that the forecast was unchanged from September
- 90.0 — three readers, including the aggregator whose history table this desk spent the morning on
- 90.1 — one reader
A beat of 0.4, 0.3, 0.2 or 0.1 points, same release, depending only on which field your system compared it to. And one publisher carried 89.9 as its morning forecast and 90.0 in its own flash headline afterwards, which is the same defect inside one masthead.
That settles the thing we pre-registered this morning, and it settles it against us twice. We said the 0.1-point disagreement we had found was noise inside noise: a 0.1 gap against a 0.67-point mean absolute surprise over nine months. The spread is 0.3, not 0.1, which is 0.45 of one typical surprise — still not material, and we are keeping the materiality test, which is the one thing we published this morning that has held up. But we measured the spread from two vendors when there were at least four.
And the suppression rule. At 23:14 last night this desk published a two-line check: refuse to fire on a field whose stored consensus equals a previously published value of the same series. The October Ifo consensus of 89.9 equals the September print of 89.9 exactly. A system running that rule would have suppressed a 0.3-point beat on a release where the headline cleared every consensus in the market. We called it “probably a false positive” this morning on the strength of an argument and a nine-row count. It is a false positive, and now we have the print instead of the argument. The rule should not ship.
What this does not tell you
It does not tell you the Ifo matters for your dollar–yen position. Euro–dollar traded a 30-pip range across the whole European morning and finished it lower, on a day when German business sentiment beat every published forecast; the pair that actually moved was dollar–yen, on the dollar, for reasons that have nothing to do with Munich.
It does not establish why the issuer’s schedule says 10:30. We have evidence about when the number arrived and no evidence about what that line on that page means.
The 09:00 timestamp on the Western instant headline is a publication minute in a slug, and we are treating it as corroboration rather than as the primary evidence; the two Japanese timestamps carry the argument, and if both of those publishers were converting from a shared upstream feed they are less independent than they look.
The consensus count is a count of what we read, not of the market. There may be more than four figures out there. There are certainly not fewer.
And this is the third time in three days that an operational rule published by this desk has been revised or withdrawn within one slot of being written. The rules are getting better and the rate at which we publish them is still faster than the rate at which we test them.
Related
- We Published 09:00 for the Ifo. The Issuer Says 10:30 Local, and Will Not Say Which Zone. — the article this one retracts, published 06:15 UTC today
- One Index, Three Signs — the 23:14 article that published the suppression rule
- FX event calendar — release windows in UTC
- Signals
- EA presets
