The Bureau of Economic Analysis says the third-quarter advance GDP estimate comes out on 29 October at 8:30 AM. That is the issuer, on its own release schedule, and it is the whole of what the issuer will tell you: there is no time zone on the page. Yesterday we published a retraction of our own correction of the German Ifo hour, where the issuer gave “10:30 CET/CEST” — two zone labels against one time, in the week they differ — and had the minute wrong by thirty. Three days later a second issuer has the opposite defect in the same field. One publishes a minute and no zone. The other published a zone twice and the wrong minute. Neither page can be converted into a UTC timestamp without a second source, and in both cases the sources that can convert it are the aggregators.
Three fixed-offset publishers settle Thursday, and only one of them states its own zone
Yesterday this desk replaced its release-hour rule for the second time in three days. The rule it published in the morning — go to the issuer and nowhere else — was dead by lunchtime. The replacement is that a release hour is settled by a timestamp on something that could only have been written after the data existed, and that the cheapest such timestamp is one published in a zone with no summer time. Thursday is the first scheduled test of it.
Three Japanese publishers put the advance GDP at 21:30. A Japanese broker’s overseas-market schedule lists 21:30 for all five American items on 29 October: the GDP advance, the quarterly consumer-spending and core PCE advances, September personal income, and the September PCE deflator. A Fisco weekly preview published on 24 October gives 21:30 for the GDP with a consensus of plus 3.0 per cent annualised against a prior of plus 2.2. And a per-indicator page at a third publisher gives 10/29 21:30 and — this is the part that matters — states the conversion rule in its own words: American summer time means half past nine at night Japan time, winter time means half past ten. Twenty-one thirty Japan time is 12:30 UTC, which is 8:30 in the morning in New York while the United States is still on daylight time.
So the issuer’s 8:30 and the three schedules at 21:30 are the same instant and they agree. What they do not share is completeness. The issuer gives you a minute you cannot place. Two of the three schedules give you a time with no zone label on the page either — one of them says only that its figures are expected release times. The third is the only source in the set that publishes both its zone and what happens to it when a clock changes, and that single sentence is worth more than the other four pages put together, because it is the only one that survives 1 November. On that date the American offset moves and 8:30 in New York becomes 13:30 UTC while the issuer’s page still reads 8:30 AM. One published figure, two UTC times, no annotation.
GDP and PCE land in the same minute, and the issuer is the one telling you so
The same schedule carries a second line for 29 October: Personal Income and Outlays for September, also at 8:30 AM. The quarterly growth number and the monthly inflation number arrive together.
This is the one thing the issuer’s page is unambiguously good for. A forward schedule is a statement of intent about a document, which is what burned us on the Ifo; but a statement that two documents will be published at the same moment is a statement about the issuer’s own process, and nobody is better placed to make it. The Japanese schedule agrees, listing five separate American items at the identical 21:30.
What a system trader does with that is not a directional question. It is a question about whether the news window is one event or five, and whether a filter keyed on “US GDP” will be flat for a release whose most market-relevant field is a price deflator filed under a different name. The consensus set a Japanese publisher carries for the same minute is GDP plus 3.0 per cent annualised, personal consumption plus 3.4, the GDP deflator plus 3.5 and core PCE plus 2.4. Four forecasts, one timestamp. And the Atlanta Federal Reserve’s own nowcast sat at 3.5 per cent yesterday on one reader, half a point above the survey consensus for the figure it is nowcasting, which is a gap worth knowing before you decide what counts as a surprise.
An aggregator’s Tuesday block is one hour early in five rows, and this time we have behaviour rather than consistency
A major aggregator published a Tuesday preview yesterday afternoon listing today’s American releases with times it labels Eastern. Durable goods orders 7:30 in the morning, Case-Shiller 8:00, new home sales and consumer confidence 9:00, the five-year note auction at noon.
The Japanese broker’s schedule gives, in Japan time, 21:30 for durable goods, 22:00 for Case-Shiller and 23:00 for both new home sales and consumer confidence — which are 8:30, 9:00 and 10:00 in New York. Every row is exactly sixty minutes apart from the aggregator’s, in the same direction, with no exceptions.
Three days ago we would have stopped there and we would have been wrong to. Yesterday morning this desk convicted a different aggregator of a thirty-minute error on the grounds that its nine rows of history were uniform, and the uniformity turned out to be correctness; the lesson we wrote down was that a vendor convicted on consistency alone is a vendor convicted without evidence, and that before publishing that a consistent column is a wrong column you should find one observation the column contradicts. Today we have one. Yesterday’s two-year note auction was scheduled by a fourth publisher at “17:00GMT/13:00EDT”, and the result was published at 17:03 UTC and again at 17:22 — which is 13:03 and 13:22 in New York. A results note cannot precede the auction it reports. The one o’clock convention for a Treasury note auction is therefore observed and not asserted, and the aggregator’s noon entry for today’s five-year is an hour early against the behaviour of the adjacent security on the adjacent day.
We still refuse to say which half of its field is defective. A page that renders every American time exactly one hour behind Eastern while labelling the column Eastern is as consistent with a zone-rendering artefact as with a stored error, and from here those look identical. It is the same refusal we applied to that publisher’s two anomalous auction rows yesterday, and it is the honest one. The operational consequence does not depend on the diagnosis: if your calendar came from that preview you are an hour early on five entries today, including an auction.
The forecasts on the same page, for what they are worth: durable goods plus 1.8 per cent against a flat prior, core plus 0.6 against plus 0.3, new home sales 649 thousand against 684, and Conference Board consumer confidence 80.5 against 81.9. The page also lists the Case-Shiller twenty-city index twice under the same name with two different priors, 2.5 per cent and 0.0 — almost certainly an annual rate and a monthly one sharing a label, which is the field-mislabelling pattern this desk has now logged repeatedly at this publisher family and which we are not going to adjudicate from one page either.
The European Central Bank’s rate is 2.50 or 2.65, and both readers are right
This desk has published “ECB at 2.5, hold expected” in every week-ahead table for several runs. The Fisco preview published on 24 October says the Governing Council is expected to hold at 2.65 per cent, from a previous 2.65. Fifteen basis points apart, on the current policy rate of a central bank meeting in three days.
The issuer settles it and neither of us is wrong. The Bank’s own key-rates table gives three rates effective 16 September 2026: the deposit facility at 2.50, main refinancing operations at 2.65, and the marginal lending facility at 2.90. A second reader confirms all three and the 16 September date, and adds that they were set at the 10 September meeting, which also explains why the issuer’s page carries an earlier publication date than its own effective date. Our 2.50 is the deposit rate. Fisco’s 2.65 is the refinancing rate. There is a third at 2.90 that neither of us mentioned.
On 22 October we published a piece about Turkey setting three rates and moving all of them in parallel, and the finding we took from it was that field risk is a property of the institution rather than of the event — that the right question is not how many rates a bank sets but whether its instruments moved together last time. Apply that test here. The row before 16 September reads 2.25, 2.40 and 2.65: the same fifteen and twenty-five basis point spacing, unchanged across the September move. The corridor is parallel, exactly as Turkey’s was. So the Turkish test returns a clean answer for the European Central Bank and the answer is reassuring about Thursday — whatever it does, a single-field feed will not miss it.
It is not reassuring about the week-ahead table, which is ours. A hold at 2.50 and a hold at 2.65 describe the same decision, but a system that stores one reader’s figure and compares it to another reader’s print sees a fifteen basis point surprise that did not happen. We have been writing a number without its instrument for several runs and nobody caught it until a Japanese weekly preview disagreed with us. From now the table says deposit facility.
The hour, at least, is settled three ways and the issuer is right about it: the Bank’s own weekly calendar gives the decision at 14:15 Central European Time and the press conference at 14:45, which post-clock-change are 13:15 and 13:45 UTC, and Fisco’s 22:15 Japan time is the same instant.
What this does not tell you
- It does not tell you that the advance GDP will print at 12:30 UTC. Four sources agree on the instant and none of them is a reaction timestamp, which is the standard our own replacement rule sets. Thursday is the test, not the proof, and if it prints at some other minute the rule published yesterday is in the same position its two predecessors were.
- We cannot say what the Bureau’s 8:30 AM is measured in. It is almost certainly Eastern, every reader treats it as Eastern, and the page does not say so. We are not going to invent the label, having spent yesterday retracting an invented inference about a different issuer’s time field.
- The one-hour aggregator offset is measured against one Japanese schedule for four of the five rows and against a different publisher’s auction convention for the fifth. It is not measured against five independent sources.
- Whether the aggregator’s column is mislabelled or miscalculated is undetermined and we have said so. A reader who can see that page rendered in a known zone can settle it in one glance and we cannot.
- All of today’s and Thursday’s consensus figures above come from one publisher each. We have no second reader on any of them, and this desk’s own record on consensus figures is that a single vendor’s number has been wrong, or been one of four, often enough that it should be treated as a stored field rather than as the market’s expectation.
- The Atlanta nowcast and the survey consensus are not the same kind of object and the half-point gap between them is not evidence that either is wrong.
- One reader puts swaps pricing at 75 per cent for a 25 basis point Bank of Japan increase to 1.50 at a December meeting it dates to the eighteenth. That is single-sourced, and the Bank’s own release calendar did not carry a December meeting when this desk last read it, so the date is a secondary reader supplying something the issuer’s page did not. We are publishing it with that caveat rather than leaving it out or treating it as settled.
Related
- We Corrected 09:00 to 09:30 for the Ifo. It Printed at 09:00 — the retraction that produced the rule this article tests, and the issuer whose defect was the mirror image of this one’s.
- All Three Rates Moved 100 Basis Points — the Turkish three-rate piece whose test we applied to the European Central Bank here.
- FX Events — the week’s releases and central-bank decisions in UTC.
- Signals
- EA Presets
