The US durable goods advance report lands at 12:30 UTC and the New York open follows at 13:00. By the time you read this, dollar-yen has already travelled 135 pips today — 158.95 down to 157.60 — and 144 from yesterday’s high. It has gone through the level Japanese authorities are reported to have checked rates at on 18 September and is sitting 45 pips below it. A forecast range published in Tokyo twelve and a half hours ago has been broken 40 pips to the downside before the first American number of the day. This is a sizing note, not a direction note, and the sizing question this afternoon is unusually concrete: the move you would normally budget for has already happened.
The day’s range is gone and the event window has not opened
The sequence, from the Tokyo wire, all times converted to UTC. Spot high 158.95 around 06:50. Down through 157.95 by 06:50 on the same publisher’s log, 158.05 area at 07:33 with euro-dollar at 1.1380 and euro-yen at 179.88, 157.67 by 09:53, and a fresh low at 157.60 by 11:23. Yesterday’s high was 159.04; Monday’s low was 156.58.
Put the forecast next to it. A Japanese retail house published a predicted range for today of 158.000 to 159.800 at 22:46 UTC last night — 180 pips wide, which is a serious range and not a lazy one. Spot is 40 pips below the floor of it. Whatever your own pre-session range was, it was probably drawn against the same overnight information, and it has the same problem.
Two things follow, and they point in opposite directions, which is why this is a sizing decision rather than a view. The first: a market that has already spent its daily range before the scheduled catalyst has less left to give the catalyst, so a stop placed at the usual multiple of today’s realised range is now wider in absolute terms than it was at the Tokyo open and is buying you less. The second: a market that has travelled 135 pips against the direction it was travelling twenty-four hours ago is not a market with a settled position, and the 12:30 print lands into thin conviction rather than thick.
The one number worth fixing in your head is 158.05. That is the level at which Japan is reported to have called dealers last week, and spot is now below it on a day when Tokyo has been talking. We published this morning that the most senior verbal signal short of actual money — the disclosure that the US president raised yen weakness at Tuesday’s summit — bought 45 pips and gave 17 back inside half an hour. That article set a forward test on where dollar-yen closes London and New York today. London has not closed. We are not marking it yet, and the desk that wrote it will mark it, not this one.
The consensus is ten basis points wide. The series misses by eighty-three.
Here is the thing you are being invited to argue about at 12:30. Forex Factory has headline durable goods at minus 0.3 percent. Investing.com’s calendar has minus 0.3 percent. Investing.com’s own preview has minus 0.3 percent. FXStreet’s event page has minus 0.4 percent. All four agree the prior was 1.1 percent. Core durable goods is 0.6 percent against a 0.4 percent prior on two readers.
So there is a genuine vendor split, three to one, and it is ten basis points wide.
Now look at what this series actually does to forecasts. Investing.com publishes its own release history on the same page as the consensus, and the last three prints read like this: August’s release of July data came in at 1.1 percent against a 0.4 percent forecast, a 70 basis point miss. July’s release of June data came in at 0.3 percent against 1.6 percent, a 130 basis point miss. June’s release of May data came in at minus 4.5 percent against minus 5.0 percent, 50 basis points. Mean absolute miss across the three: 83 basis points.
The vendor disagreement is 10 basis points. The series’ own forecast error is eight times that. If your process involves deciding which calendar to trust before a durable goods print, you are optimising the smallest term in the expression. This desk has spent six articles in four days on consensus disputes and this is the first one where the honest answer is that the dispute does not matter: at an 83 basis point error distribution, minus 0.3 and minus 0.4 are the same forecast.
What does matter is the shape of the tail. A series that can miss by 130 basis points in one direction and print minus 4.5 percent two months earlier is not a series you size symmetrically. The 14:00 UTC revised University of Michigan sentiment index has the same vendor problem in miniature — 47.6 on FXStreet, 47.4 on Forex Factory, prior 47.8 — and the same answer applies to it.
One widely read preview has every time an hour early, and we can tell you why
This is the part that will cost somebody money today, and it is not a forecast problem.
Investing.com published a Friday preview at 14:00 UTC yesterday listing the day’s releases in Eastern time. It gives durable goods at 7:30 AM ET. It gives Michigan sentiment and expectations at 9:00 AM ET. It gives a Williams appearance at 4:15 AM ET. It gives the Baker Hughes rig count at 12:00 PM ET and the CFTC positioning reports at 2:30 PM ET.
Every one of those five is exactly sixty minutes earlier than the release it names, and we can show it at the issuing agency rather than inferring it. The Census Bureau’s own advance durable goods report says, in print: “The Advance Report on durable goods for August is scheduled for release on September 25, 2026, at 8:30 a.m. EDT.” Eight thirty Eastern Daylight Time is 12:30 UTC, which is what FXStreet’s event page says, and what Investing.com’s own calendar page says, and what Forex Factory says. Three vendor calendars and the agency all agree on 12:30 UTC, and the preview says 11:30.
Michigan checks out the same way on two independent readers: FXStreet gives 14:00 UTC, Forex Factory gives 14:00 UTC, the preview gives 13:00.
The mechanism is the daylight-saving offset. 8:30 EDT is 7:30 EST. 10:00 EDT is 9:00 EST. 5:15 EDT is 4:15 EST. 1:00 EDT is 12:00 EST. 3:30 EDT is 2:30 EST. All five times are correct to the minute if you render them in Eastern Standard Time and then label the column “ET” in late September, when the United States is on Eastern Daylight Time. That is not a typo repeated five times. That is one timezone constant, wrong, applied consistently.
We are flagging this because it closes something. On 18 September this desk recorded six event times that came back an hour early from a source and filed it as unexplained. It has sat open for a week. This is the same failure, on a different publisher, with the mechanism visible — and it means the right response to an hour-early time is not to assume a typo and split the difference. It is to check whether the whole page is on the wrong constant, because if it is, every row is wrong and you can correct all of them with one addition.
The practical consequence for a news filter: if your blackout windows are built from a preview page rather than from an event feed, they will open and close an hour before the events they are meant to cover, twice a year, for the whole of daylight-saving time. You will be flat through a quiet hour and fully exposed through the print. Two of the five times above we second-sourced independently; the Williams time rests on one calendar, and the rig count and CFTC times we have not second-sourced at all — they fit the pattern, which is not the same as being verified.
What the window actually is today
All times UTC. Two releases in the same minute at 12:30 — headline durable goods and core durable goods — which is the ordinary trap of this print: the headline is dominated by aircraft orders and the core is the one with information in it, and they can disagree in sign. 13:00 New York equity open, twenty-nine minutes later, with the durable goods reaction still incomplete. 13:20 Schmid. 14:00 revised Michigan sentiment and the one-year inflation expectation, prior 4.6 percent. 18:00 Hammack. Then the CFTC positioning reports in the late American afternoon, which is the first snapshot that contains both the Bank of Japan’s hike and the 18 September rate check, and which is after this slot closes.
The clustering that matters is 12:30 to 14:00: two prints, an equity open and a Fed speaker inside ninety minutes, on a day where dollar-yen has already moved 135 pips and the dollar index has come off 101.398 to 101.001 inside its previous day’s range. If you run one blackout window this afternoon rather than four, 12:25 to 14:10 is the one to run, and it costs you an hour and forty-five minutes of exposure rather than four separate re-entries into a market that is not going to be liquid between them.
The correlated-exposure point, briefly. Euro-dollar made 1.1409 and sterling-dollar 1.3254 on their highs this morning while dollar-yen was making its low. That is a dollar move, not a yen move, and a book that is short dollars against three currencies this afternoon is running one position in three tickets. The 12:30 print is a dollar event and it will hit all three.
What this does not tell you
It does not tell you which way durable goods breaks, and the 83 basis point error figure is a three-month sample, which is far too short to be an error distribution. Treat it as an order of magnitude, not a parameter.
It does not tell you that 157.60 is the low. It is the low as at 11:23 UTC on one publisher’s log, and the US session has not started.
It does not settle the 158.05 rate-check level. That decimal has come from exactly one publisher for five consecutive runs now; the event has two readers, the number has one, and we keep saying so because it keeps being true.
And it does not tell you that the Williams, rig count and CFTC times in the preview are wrong. They fit the daylight-saving pattern exactly, which is suggestive and is not evidence. We second-sourced two of the five against the issuing agency and independent calendars, and we are only claiming those two.
Related
- FX events calendar — today’s window, in UTC
- Signals
- EA presets
