Yesterday morning this desk published, in print, that it carried an expectation of a Japanese flash manufacturing PMI at 00:30 UTC today, could not find it on the calendar, and was recording it as unresolved rather than publishing a time it could not stand behind. The release happened. It printed 54.1 against a 55.0 consensus. And it was on the calendar the entire time — one row up and one day back, because the calendar renders in New York and 00:30 UTC on a Thursday is half past eight on a Wednesday evening there. The calendar was not wrong. We read a display artefact as a fact about the world, and if your news filter keys on a vendor’s day column, so does it.
What we said, and what actually printed
The 24 September Tokyo article’s humility section stated that Forex Factory showed no Japanese releases at all on Thursday 24 September, against an expectation we had carried from the previous slot of a flash manufacturing PMI at 00:30 UTC with a 55.0 consensus and a 54.9 prior. That sentence was accurate about the page and wrong about the world.
The print, from three readers — the calendar’s own actual column, a market data aggregator and a wire desk:
- Manufacturing 54.1, against 55.0 expected and 54.9 previous — a miss of nine tenths and the softest manufacturing reading since February
- Services 51.6, from 52.5
- Composite 52.5, from 53.5 — the weakest since May
Nine tenths is not a rounding error. Every one of the four consensus splits this desk spent last week arguing about was one to three tenths wide. This was a single, undisputed consensus figure and the print missed it by three to nine times those gaps — the kind of surprise that is supposed to be worth a position, on a release we told you we could not confirm existed.
The row was never missing
The mechanism is boring and that is exactly why it survives. Forex Factory states its display timezone at the top of the page: America/New_York, currently four hours behind UTC. A release embargoed for 00:30 UTC on Thursday renders at 8:30pm on Wednesday. It is the same instant. It is on a different row of a different day.
The issuer settles the instant independently. S&P Global’s own flash Japan release carries an embargo line in the header of every edition — the July one reads, verbatim, “Embargoed until 0930 JST 24 July 2026 / 0030 UTC 24 July 2026”. Ninth-thirty Tokyo, half past midnight UTC, the twenty-fourth of the month. The publisher and the calendar agree perfectly. Only the day label differs, and the day label is a rendering choice.
So the honest version of yesterday’s finding is not “the calendar is missing a Japanese release”. It is “we asked the calendar a question in the wrong timezone and it answered correctly”. That is a worse result for us and a more useful one for you, because a missing row is a vendor bug you can report and a displaced row is a class of error that will recur on every single Tokyo-morning print for as long as you keep reading day columns.
It is not one row. It is the whole Asia session.
Look at what else sits under Wednesday 23 September on that page, after 8pm New York time:
- 8:30pm — Japan flash manufacturing PMI, 54.1 against 55.0, prior 54.9. That is 00:30 UTC Thursday.
- 9:30pm — Australian employment change, 39.5K against 21.5K expected, prior negative 15.9K. That is 01:30 UTC Thursday.
- 9:30pm — Australian unemployment rate, 4.6 percent against 4.5 expected and 4.5 prior.
- 9:30pm — NAB quarterly business confidence, negative 11 against a prior of negative 19.
Four Asia-Pacific releases, including an employment beat of eighteen thousand and an unemployment rate that rose anyway, all landing on Thursday in UTC and all filed under Wednesday. The calendar’s Thursday, as displayed, contains its first event at 3:30am New York — the Swiss National Bank. In UTC terms, Thursday’s first four hours had four prints in them and the page shows an empty morning.
The Australian pair is the sharper illustration, because it moved something. One Japanese desk attributes part of this morning’s AUD/JPY weakness directly to the employment report: jobs expanded, participation rose faster, the unemployment rate went up, and the Australian dollar was sold. AUD/JPY was 110.81 at 02:55 UTC. If your correlated-exposure check runs once a day against the events listed for that day, it saw none of this.
What your scheduler should key on
The fix is not to distrust the calendar. The calendar was right. The fix is to stop letting a vendor’s presentation layer define your event set.
Three things worth changing before tonight, in order of how cheaply they pay:
- Key news windows to UTC instants, never to a vendor’s day bucket. If your filter blocks trading thirty minutes either side of a high-impact JPY event and it built today’s list from the rows displayed under Thursday, it was open through 00:30 and 01:30 and closed for nothing. That is the failure mode: not a blocked window you did not want, but an open window you thought you had blocked.
- Pull a two-day span, always. Any Asia-hours release is on the previous calendar day in America/New_York, and it crosses back over on the US daylight-saving switch. A query that fetches day D and day D minus one, then filters by UTC timestamp, is immune to both. A query that fetches day D is wrong for the entire Tokyo session, every day, forever.
- Get the instant from the issuer when it is load-bearing. S&P Global prints the embargo in UTC in its own header. Statistical agencies almost always do. An aggregator’s time is a conversion; the issuer’s is the definition.
None of this is a market call and none of it will show up in a backtest, because a backtest built from the same displaced calendar reproduces the error faithfully in both directions. It shows up as one live session where a filter you believed in was not on.
And the previous column is doing it here too, in the other direction
Yesterday this desk published that US calendars carry the final PMI in the previous column while the release coverage compares to the flash, and asked out loud whether the same held elsewhere. One reader says it does here: it reports August Japanese manufacturing as revised lower to 54.9 from a preliminary 55.1.
If that is right, the calendar’s 54.9 prior is the final and September’s comparison against the flash is 55.1. September is a drop of eight tenths against the number in the previous column and a full point against the number the August release actually published. And the sign runs the other way from the US case — American manufacturing was revised up seven tenths for August, Japanese manufacturing down two tenths.
That is one reader, on one series, in one month. We are recording it as a single-sourced extension of yesterday’s finding, not as a confirmation of it, and it does not license a correction factor — if anything the opposite, since two regions revising in opposite directions is precisely what makes one impossible.
What this does not tell you
We do not know why the day-boundary problem bit us specifically yesterday rather than on any of the dozen previous Tokyo-morning releases this file has covered, and the honest answer is probably that it did bite before and we happened to have the release from another source. We have not audited our own archive for it. That audit is owed.
Our two reads of the same calendar, twenty-five minutes apart, differed on whether the Japanese actual column was populated — the first, a whole-page read, showed it empty; the second, a day-specific read, showed 54.1. We are treating the day-specific read as authoritative and making no claim about the first, including no claim that the vendor was slow. The two Australian reads also differed by one tenth on the previous field, negative 15.8K against negative 15.9K, which we mention only so that you do not take either from us as settled.
We could not reach S&P Global’s September flash Japan release itself. The embargo line quoted above is from the July edition of the same series; we are relying on the cadence being unchanged, which is an inference, and the September figures come from secondary readers rather than the issuer. A search for the September release returned a same-titled S&P Global page from 2023, which we discarded. Nothing here is sourced to a Japanese market summary we also found and cut, because it turned out to be the 24th of August.
And we have no view on whether 54.1 matters for the yen. It printed in a holiday-thinned handover, spot has since recovered rather than weakened, and attributing this morning’s move to a survey would be exactly the kind of story this article is about.
Related
- FX events calendar — times in UTC, which is the entire point
- Signals
- EA presets
