US October payrolls print at 13:30 UTC today, and for the first time in weeks this desk cannot find a consensus dispute. Four independent readers carry the same number to the thousand: 65,000. One prints it under the label “The market consensus”, one calls it the market consensus in prose, a Tokyo calendar carries it as 65,000, and a bank research note cites it as the median of Bloomberg contributors. We have spent five weeks writing about readers who disagree, so it is worth saying plainly what the agreement is worth: nothing. One month ago the same four-way unanimity sat at 90,000, and September printed 29,000.
The unanimity is real and it is not evidence
Take the agreement seriously first, because it is genuine and it is unusual. A dedicated payrolls calendar page prints a 65K card under “The market consensus”, with the table column headed, pleasingly, “Forecast (The market consensus, not our view)”. A research outlet writes that markets expect an increase of approximately 65,000 and attributes it to market consensus rather than to itself. A Tokyo calendar prints 65,000 against a 29,000 prior. And a bank note names its source explicitly: the median of 65k from Bloomberg contributors.
That last one is the only reader that tells you where the number came from, and it is the reason the other three agree. They are not four surveys. They are four renderings of one survey median, and a cluster of identical figures with one upstream source has exactly the information content of that one source. The usual trap on this desk is a spread that looks like disagreement and turns out to be a labelling fault. This is the inverse: an agreement that looks like confidence and is a single data point wearing four hats.
Unemployment is the same shape. One bank expects 4.2 per cent and says so is in line with the Bloomberg median; a second reader says the rate is projected to remain at 4.2; the Tokyo calendar prints 4.2 against a 4.2 prior. Three readers, one median, one number.
Score the identical setup from one month ago
This is cheap to check and it is the only thing here that is actually evidence. The same calendar page that carries today’s 65K also carries its own September row: Actual 29K, Forecast 90K. Its body states it without flinching — the last release came in at 29K for September against a forecast of 90K.
So the consensus missed by 61,000 on a 90,000 forecast. The error was 68 per cent of the forecast it busted, and — this is the number to carry into today — it was 94 per cent of the entire level everybody is quoting this morning. If October repeats September’s error in either direction, the print lands at roughly 126,000 or 4,000. Both of those are inside one month’s demonstrated error, and they are opposite trades.
A bank note gives the three-month average payroll growth as 51k, which is the most useful single figure on offer today, because it is a measurement rather than a forecast. Against a 51k trailing average, a 65k consensus is asking for mild acceleration, and the honest statement about the distribution is that it is wide enough to contain zero.
One small trap on that same page, for anyone scraping it: it carries two different correct answers to “what was the prior”. The card reads Previous 29K, meaning September’s actual. The September table row reads Previous 133K, meaning August’s. Both are right; a field-name-keyed reader gets whichever it hits first, and the two are 104,000 apart.
The named houses are 40,000 apart around a median they all quote
The median is agreed. The distribution behind it is not, and the published houses do not straddle it symmetrically.
One bank forecasts 90k — above the median it names in the same sentence — with unemployment at 4.2 per cent. Another is reported at about 50,000 with unemployment ticking down to 4.1 per cent. One reader refers to unnamed institutions near 100,000. So the two named banks sit 40,000 apart, which is 62 per cent of the median, and they disagree about the direction of the unemployment rate as well as the level of hiring.
This matters more than a wider survey range would, because these are the forecasts with reasoning attached. A 40,000 gap between two named houses on a 65,000 median tells you the professionals are not arguing about a tenth at the margin; they are arguing about whether hiring roughly doubled or roughly halved relative to trend. If you size off a consensus surprise, the surprise you are sizing against is drawn from that distribution, not from the median.
Our standing preference applies: a house’s conditional threshold beats anyone’s point estimate, and today nobody has published one we can read. That is a gap in the record, not a reason to trust the median more.
Canada prints in the same minute, and your filter probably does not know
13:30 UTC is not one release. A Tokyo calendar lists US employment and Canadian employment at the same 22:30 Japan time, which is the same instant, and the Canadian numbers are not small: employment is forecast at +12,200 against a prior of −68,300, and the unemployment rate at 6.5 per cent against 6.5. That prior is the point. The consensus is asking for an 80,500 swing in one month on a labour force a fraction of the US one.
A news window keyed to the string “Nonfarm Payrolls”, or to USD alone, opens one window and takes two labour shocks through it. If you hold Canadian dollar exposure alongside dollar–yen — and if you run yen crosses you very likely hold correlated dollar risk — the single filter you think covers 13:30 covers half of it. The fix is not a wider window; it is a second event on the same timestamp, so the exposure check runs against both legs.
The rest of today in UTC, from the same calendar: German industrial production at 07:00, forecast −0.3 per cent on the month against a 2.0 per cent prior and 1.3 per cent on the year against 2.3; euro-area retail sales at 10:00, −0.2 against 0.1 monthly and 0.5 against 0.8 annually; the twin employment reports at 13:30; University of Michigan preliminary sentiment at 15:00, forecast 47.2 against a prior 48.0; and the Ivey purchasing managers’ index at 15:00 with no consensus published against a 58.2 prior. Two Canadian events, two US events, and the 15:00 pair lands in the same minute as the usual New York option cut.
On the clock itself: two readers print 13:30 UTC and 8:30 a.m. ET together, and this desk confirmed 8:30 a.m. ET at the issuing agency yesterday, so the hour is settled on three independent legs. It is worth noting what the other readers do, though, because it is the recurring failure: one prints “Eastern Time” in words with no clock time anywhere on the page, and the bank note prints no time at all, referring only to “this Friday’s release”. A calendar built from either would have no timestamp to key on.
The field that decides December is the one with two signs
If the December rate question turns on wages — and one Tokyo publisher says explicitly that it does, flagging average hourly earnings above expectations as the condition that would raise December hike odds — then the headline payroll number is the wrong field to watch, and the right one can print as forecast and still support both conclusions.
The Tokyo calendar gives both wage fields: average hourly earnings at 0.3 per cent on the month against a 0.1 per cent prior, and 2.9 per cent on the year against 3.0. Read it slowly. Consensus expects the monthly rate to triple and the annual rate to fall at the same time. That is not a contradiction — it is what happens when a strong month a year ago drops out of the base — but it has a consequence a filter cannot reason about: a release that prints exactly in line reads as sharp wage acceleration on one field and continued deceleration on the other.
So a system keyed to a single wage field does not get half the signal. It gets a signal whose sign depends on which half it happened to be wired to. If you are going to trade the wage print, the only defensible version is to key on both fields and require agreement, and to treat the in-line-on-both case as a non-event rather than as confirmation.
One further note on expectations, which cuts against the hawkish read: one reader separately offers 0.2 to 0.3 per cent monthly as its own view rather than a consensus figure, so the 0.3 is the top of at least one published range and not a central case.
What this does not tell you
We did not reach the issuer. Every figure above is a secondary reading, the consensus figures are four renderings of one survey median we cannot see, and the Bloomberg contributor distribution behind the 65k is not public to us. Two pages we went to specifically for this release — a major vendor’s own calendar event page and one broker preview — returned client errors, so we have no fifth or sixth reader and no high-low survey range at all.
No page we read reports this release as delayed, rescheduled or affected by a data gap. That is four readers’ silence rather than a confirmation, and silence has been wrong before.
We have no average hourly earnings prior from any English reader — the 0.1 per cent monthly and 3.0 per cent annual priors come from the Tokyo calendar alone, and the whole base-effect argument above rests on them. We also have no participation rate forecast, which is the field that decides whether a 4.2 per cent unemployment print is good or bad news.
And one of our own figures is now stale in a way that matters today. Twelve hours ago this desk published one-week dollar–yen implied volatility at 6.23 per cent, the lowest since 22 October, on precisely the tenor that covers this release. Between publishing that and writing this, the Nasdaq Composite fell 5.52 per cent. Neither wire we read overnight carries any FX implied volatility figure, so we cannot re-read it. Do not size today’s payrolls window off that 6.23.
