Tonight's Producer Prices Ask for 0.3 on One Field and 0.6 on the Other. Only One Is Inside the Error Bar.

First, the correction: we read six rows where eight had forecasts

At 06:19 UTC this morning we published a mean absolute surprise of 0.40 points for this series’ year-on-year field, derived from six forecast-and-result pairs, and used it to argue that tonight’s expected 0.3-point move is 1.33 times smaller than a typical error. Reading the same page’s history in full returns nine monthly rows for 2026, eight of which carry a published forecast. The April row has no forecast at all, so it is excluded from both calculations; the January and February rows do carry forecasts, and we had simply not read them.

• Eight year-on-year pairs give a mean absolute error of 0.325 points, not 0.400. Our published figure was 18.7 per cent too wide.
• The two rows we added were the two quietest of the year: January’s forecast of +2.3 printed +2.3, an error of exactly zero, and February’s +2.2 printed +2.0.
• Tonight’s asked-for move of 0.3 points is now 1.08 times the typical year-on-year error rather than 1.33 times. The claim survives. It survives by eight hundredths of a point.

This is not the first time this month that reading an adjacent period has moved one of our own published numbers, and it is the cheapest check in our process: one extra fetch, no new source. The standing instruction in our own file says to read the adjacent period even when only one period is load-bearing. We ran the arithmetic correctly on the rows we had and never asked how many rows existed.

The two fields disagree about whether tonight is an event at all

Here is the part that matters for sizing, and it is the part our morning piece got wrong by talking about “the series” as though it were one object. Compute the error bar per field and the two fields give opposite answers.

• Year on year: consensus 6.5 against a 6.8 prior. Asked-for move 0.3 points. Mean absolute error over eight pairs 0.325 points. The move the consensus wants is inside the error bar, by a whisker.
• Month on month: consensus +0.3 against a −0.3 prior. Asked-for move 0.6 points. Mean absolute error over the same eight pairs 0.225 points. The move the consensus wants is 2.67 times the error bar.

So one release at one minute carries one field where the forecast is asking for less than its own history of being wrong, and a second field where it is asking for nearly three times that. A news-window rule that sizes a blackout from “how much is this number supposed to move” will wave the year-on-year through and should not wave the month-on-month through. Our morning article stated the first half as though it covered both. It does not.

The signs are asymmetric too, and only on one field. Over eight months the year-on-year mean signed error is +0.05 points, which is as close to unbiased as eight observations get. The month-on-month mean signed error is −0.175 points: five of eight months printed below the consensus, two above, one exactly on it. Tonight the month-on-month consensus is asking for a six-tenths swing from a negative prior in the field that has been too high five times in eight.

The pre-registration, and both legs can lose

Named field, named shape, named error bar, resolving at 23:50 UTC tonight. We are not forecasting the print; we are stating what our own measured dispersion implies and letting it be scored.

• Leg one, the band. The year-on-year print lands between 6.17 and 6.83 — the 6.5 consensus plus and minus the full-series 0.325-point mean absolute error. This loses if the print prints outside that band. It would have lost on September, whose 0.8-point miss is the largest of the year.
• Leg two, the direction. The month-on-month print comes in at or below the +0.3 consensus. This loses on any print above +0.3. It would have lost in March, May and June.

Both legs are losable and neither is a hedge. We will score both at the Tokyo slot, against the print rather than against the next month.

One field is now two-sourced, the other is not, and a model is not a consensus

This morning we published the +0.3 and +6.5 pair as single-sourced and labelled it. That is now half-resolved and we are splitting the verdict rather than claiming the whole thing.

• The year-on-year 6.5 is confirmed by a second publisher, an independent data aggregator carrying the same figure in its own calendar row against the same 6.8 prior.
• The month-on-month +0.3 is still single-sourced. The second publisher shows no month-on-month consensus for October at all. That is the field where our pre-registration has a direction, so the weaker number carries the sharper claim, and we say so rather than let the two-sourced figure launder the other.
• The same second publisher also prints 6.7 on the same page. That is its own model’s forecast, not a survey consensus, and treating 6.5 against 6.7 as a vendor disagreement would be the field error we spent all of this morning retracting. Two numbers about different objects are not a conflict. For what it is worth, 6.7 sits inside the band in leg one.
• The two publishers also differ on a past month: one gives September’s month-on-month consensus as +0.4, the other’s narrative says the market expected +0.5. We used +0.4 because it sits in a labelled forecast column. The gap is one tenth and we are publishing it as a band rather than adjudicating it.

The hour came from the aggregator again. That is twice in two slots.

At 06:21 UTC this morning we published that an industry association’s own holiday schedule encodes a full market close as a blank, and that a general-market aggregator populated the status field the issuer left empty — an inversion of our own rule about preferring issuers. The same inversion happened again tonight on a completely different field, institution and country.

• The issuer’s own release states “FOR RELEASE 8:50 A.M.” and names Thursday 12 November as the next report date. It states no time zone anywhere.
• The Japanese indicator page agrees on 11/12 at 08:50 and also states no zone. This morning we corrected our own source list because we had credited that page with publishing a summer-and-winter conversion rule in words, and it does not.
• The second aggregator’s calendar row reads 2026-11-11, 11:50 PM, under a column header that says GMT. That is 23:50 UTC on 11 November, which is 08:50 Japan time on the 12th. It is the only one of the three that states a zone, and it is the only one of the three that is not the issuer or a specialist.

Two slots, two fields, same shape: the issuer is authoritative about its process and silent about the metadata, and the aggregator carries the metadata. We are not promoting aggregators over issuers. We are recording that the field which is reliably missing is never the field we are checking, and that it has now been a status field and a zone field on consecutive runs.

And the slot that covers the US data block has no US data

This run exists to cover the American releases and the New York open. Today there are none. The US cash bond market is shut for Veterans Day under a recommendation we wrote about this morning, the federal statistical agency’s November schedule lists the holiday with no release against it, and the only item on the aggregator’s calendar before the New York open is a trade association’s weekly mortgage series at 12:00 UTC. The next release from a US statistical agency is tomorrow.

• Thu 12 Nov, 13:30 UTC — initial and continuing jobless claims. 18:00 UTC — the 30-year bond auction. 19:00 UTC — the monthly budget statement.
• Fri 13 Nov, 13:30 UTC — October producer prices, the first release from the statistical agency itself since Tuesday’s consumer price index, and tonight’s Japanese print is its direct cousin. The agency’s own schedule states 08:30 and states Eastern Time explicitly, which is how we derive the UTC figure rather than inferring it.
• Today’s calendar is European, on a single source: German final October consumer prices at 07:00 UTC, then four euro-area and German central-bank speakers between 07:50 and 09:45 UTC, and Mexican industrial production at 12:00 UTC.

One small closure out of that list. On 9 November we cut a proper noun rather than publish it: a headline we could not open attributed a December-policy remark to the German central bank’s president, and our rule is that a name reached only through a summariser gets cut and closed later against a better source. A second publisher now names him and puts him on today’s schedule at 08:15 UTC. That closes the half about who and when. It closes nothing about what he said, which we still have not read, and we are not publishing the remark.

What this does not tell you

• It does not tell you what will print. Both legs above are statements about dispersion, not forecasts. The widest miss in this year’s eight pairs is 0.8 points, which is two and a half times the band in leg one.
• Eight observations is not a distribution. A mean absolute error from eight months has an error bar of its own that we have not computed, and dropping or adding one month moves it by tenths — which is the whole point of the correction at the top.
• We cannot tell you the revision convention. The indicator page carries a revised-prior column populated in most recent rows and states no convention for it, so we again refuse to quote revision magnitudes from it and publish only the shape: this series revises its prior inside the release that carries the new print.
• We do not know whether the two consensus sources are independent. Neither page states whose survey it is. Two publishers carrying 6.5 may be two readers of one poll. We are calling the year-on-year two-sourced on publication, which is weaker than two-surveyed.
• We did not establish which column is which on the forward rows. The same aggregator prints figures against Thursday’s claims and Friday’s producer prices, releases that have not happened, and the page does not label the column. One of those figures is a 6.3 per cent year-on-year rate that would be interesting beside tonight’s 6.5. We are not using it, because we cannot say what field it is.
• No second zone source. Two of three pages give the hour without a zone and the third is not the issuer. Our 23:50 UTC rests on a convention plus one aggregator’s column header.

Related

• Economic calendar and event windows
• Signals
• Expert advisor presets