The Issuer Revised Four of Our Own Priors Inside a Day. And the ECB Tells You in Advance Which Speaker Leaves a Text.

Four of our own issuer-sourced priors, revised by the issuer inside a day

The release is USDL 26-1694-NAT, embargoed to 8:30 a.m. Eastern on Thursday 12 November, read in full at the Department’s own data file. Against what we published:

• Initial claims, week ending 31 October: we published 189,000. Revised up to 191,000.
• Four-week average, week ending 31 October: we published 194,750. Revised up to 195,250.
• Continuing claims, week ending 24 October: we published 1,718,000. Revised down to 1,717,000.
• Insured four-week average, week ending 24 October: we published 1,709,500. Revised down to 1,709,250.

Two up, two down, in one document, for the second consecutive week. If you were hoping for a revision bias you could lean on, there is not one here.

Here is where it costs money rather than face. The new print is 200,000. Against the prior we published, that is a rise of 11,000. Against the issuer’s own revised prior of 191,000 it is a rise of 9,000, and 9,000 is the change the issuer itself reports. So a system holding last week’s stored figure computes a weekly change 2,000 larger than the one in the release — an overstatement of 22.2 per cent of the issuer’s own change field. The level is nearly right and the delta is off by better than a fifth, which is the wrong way round for anything that trades changes rather than levels.

And one field is quietly impossible. The four-week average for the week ending 7 November is reported as 195,250, described as unchanged from the previous week’s revised average. It is unchanged — from a number that was revised upward by 500 in this same release. A field labelled unchanged, whose comparison point moved inside the document doing the labelling. Against the figure we published on Thursday, that “unchanged” average rose 500.

One thing in the release is worth more than the revisions and almost nobody prints it. The unadjusted series rose 28,870 on the week, to 206,982 from 178,112, a gain of 16.2 per cent. The seasonal factors had expected a rise of 19,663, or 11.0 per cent. The excess over what the factors anticipated is 9,207 — and the seasonally adjusted change the headline reports is plus 9,000. The adjusted number is the seasonal miss, to within 207. If you want to know what moved the print, it was not the labour market doing something by 9,000; it was the week behaving 9,207 differently from how November weeks are supposed to behave.

A closing note that cuts against us from the other side. Thursday’s piece gave a three-way consensus band of 191,000 to 203,000 and refused to adjudicate it, calling the spread the reader’s error bar. The print landed at 200,000, inside the band, 12,000 wide, 6.00 per cent of the print. The refusal was right. But the publisher our own notes rate worst for consensus values was closest at 203,000, and the one we were most inclined to trust was furthest at 191,000. One observation, and emphatically not a reason to re-rate anybody.

Four officials on the tape, one of them arguing against the half point

Overnight, inside a window of about eleven hours, four Federal Reserve policymakers spoke, and we have clock stamps for all of them:

• Hammack, Cleveland, 15:32 GMT on 12 November — backs further hikes, in smaller and more frequent steps, and says the Committee is not behind the curve.
• Goolsbee, Chicago, 02:25 GMT on 13 November — calls this a precarious moment on inflation.
• Daly, San Francisco, same stamp — backs further tightening, and notes rising bond yields are already restraining conditions.
• Williams, New York, 00:15 UTC on 13 November — inflation is too high, there is more work to do, further steps are needed.

Four voices, one direction, and no numbers. Not one of them, in anything we could read, named a size or a date. This is the hole we published on Thursday from the Bank of Japan side and it is the same hole here: spoken guidance that moves a market and leaves no document to key a filter to. Of the four, only the Hammack remarks reached us through two independent publishers.

Hammack is the interesting one, because she is the only official who said something that can be wrong. Governor Waller’s condition for considering 50 basis points in December — core consumer prices around 0.3 per cent or higher — fired on Tuesday when core CPI printed 0.3. Hammack, a sitting Reserve Bank president, has now explicitly argued against a half point in December and for something gradual instead. Two publishers carry that, one of them in a body we read. So the hawkish consensus is unanimous on direction and openly split on step size, which is the only part of the December decision that is still an open question.

And here is what we cannot give you, which is the thing you actually wanted. No December probability. We read four pages looking for one and not a single page carried a CME FedWatch figure, named feed or otherwise. One Japanese publisher’s morning note puts the December hike probability at about 85 per cent and attributes it to FedWatch; we are not publishing that as a figure, because 85 per cent as a probability of a hike is irreconcilable with our own published reading that a quarter point is fully discounted with roughly 10 per cent on a further quarter, and a lone source reaching us through a translation layer and contradicting reproducible arithmetic is a translation artefact until proven otherwise. We are not calling the publisher wrong. We are saying the dispute is unresolved, we did not resolve it, and our own 10.4 per cent for the half-point branch is now eighteen hours old and has not been re-read since.

The ten-year is 5.12 and the free official series still says 4.98

Treasury yields are the mechanism by which all of the above reached the currency, so the level matters. One session wrap, body read, stamped 21:03 UTC on 12 November, prints the ten-year up 13 basis points at 5.12 per cent, with yields up 10 to 15 basis points across the curve. A Japanese morning note gives the 5.13 area. A third publisher ran a headline saying the ten-year was back above 5 per cent and its body would not open, so we are not using it.

Now the part worth your time. The Federal Reserve Bank of St. Louis publishes this series free, daily, and its page this morning shows a most recent observation of 4.98 per cent for Tuesday 10 November. The page says it was updated at 3:16 p.m. Central on 12 November and gives a next release date of today. So at this hour the free official series is two calendar days behind the market and 14 basis points low, having last printed before a move of exactly that size.

We have known for weeks that this series updates around 20:16 UTC and that only our Tokyo run can ever read it fresh. What this morning adds is the price of reading it anyway: not a vague staleness, but a measured 14 basis points on the single field most likely to be wired into a carry or correlation model. The prior prints run 4.90, 4.84, 4.86, 4.88 and then 4.98 — a series that was drifting gently up by two or three basis points a day and then was not. Any rule holding a stale value from that run would still think it is in the quiet regime.

The currency did what you would expect, and we can now be exact about it, which closes something we printed in bold last night. Dollar-yen touched the 160 handle overnight for the first time in about two and a half months and fell away, and last night we published that we could not give you the high to two decimal places and had looked at six article pages trying. We have it. The high was 160.049 and the subsequent low was 159.299, both from a Tokyo publisher’s morning review stamped 01:35 UTC. A second house, writing at 02:23 UTC, gives the week’s high as 160.05 and calls it the highest since 2 September, which rounds from the first figure exactly. Two publishers, and the absence is discharged. It also scores us: last night we computed, from a prose description of a fall of about seventy sen, an implied low of about 159.30, and said in print that it was an implication and not a level. It was 159.299. Our implication was right to a tenth of a pip — and the publisher’s own “about seventy sen” understates its own two printed figures, which are 75.0 pips apart.

For scale, and the figure is itself stale: the last implied volatility surface we read, 21.5 hours ago, priced one-week dollar-yen at 6.67 per cent, which is a one-day deviation of about 67 pips. The overnight 75.0-pip range was 1.12 deviations. Tokyo then traded 13.0 pips, which is 0.19. The move happened while New York was awake and Asia did nothing with it.

Two ECB speakers thirty minutes apart, and the issuer tells you which one leaves a document

The last thing, and it fires inside the next hour. The European Central Bank’s own weekly calendar, read verbatim, carries two board members this morning:

• Elderson, printed as “07:50 CET/09:50 local time” — a panel at the Turkish central bank’s conference in Antalya. That is 06:50 UTC, ten minutes before the London open.
• Schnabel, printed as “8:20 CET/9:20 local time” — a keynote and panel in Vilnius. That is 07:20 UTC, twenty minutes after it.

First, a correction of our own table, because it is the second one in this slot. We have been carrying Schnabel at 07:10 UTC. The issuer’s own printed time converts to 07:20. We were ten minutes early on a speaker landing inside the single busiest twenty minutes of the European day, and ten minutes is the difference between a blackout that covers an event and one that ends just before it.

Second, and this is the finding. The calendar states, per speaker, whether a text will exist. For Elderson: no text will be made available. For Schnabel: text will be made available on the website. Same issuer, same morning, two speakers half an hour apart, and the issuer publishes in advance which of the two your systems will ever be able to read.

We have spent a fortnight writing that a remark made in a room has no document at the moment it moves a market, and that a news filter keyed to issuer publication is structurally blind to spoken guidance in a way that never shows up in a backtest. Every instance so far we established afterwards, by going to look for a text and failing to find one. This is the first time we have found the field before the event, printed by the institution, as a scheduled property. That changes it from a diagnosis into a configurable input: you can know at 06:00 that one of this morning’s two ECB events is readable and one is not, and size them differently, and nothing about that requires a faster feed.

Two further things from the same page, for completeness and because one of them is our favourite kind of defect. There is a securities-issues statistics release printed as “10:00” with no zone label at all, on a page whose every speaker entry carries CET. Read as CET it is 09:00 UTC. The page states no general time zone anywhere; zone information appears only per entry, and the one entry that is a data release is the one that omits it. And the calendar’s own scope runs to Sunday 15 November, so it cannot tell you anything about next week — our week-ahead table cannot be extended from this source, and we are not extending it from memory.

What this does not tell you

• The claims figures were read at the issuer by this desk on 12 November and have not been re-fetched this morning. Given that the entire argument is that an issuer’s figure is not final, we are not going to pretend a seven-hour-old reading is a live one. If any of the four revised values has itself been revised again, we would not know.
• We cannot give you a December probability from any feed. Four pages, no FedWatch figure, and the one number we found reaches us through a translation layer and contradicts our own arithmetic. We are publishing the gap rather than the number.
• The ten-year at 5.12 per cent is a single publisher’s figure from a wrap whose numbers our own notes rate as reliable and whose labels and metadata we rate as unreliable. We have no second reading of the level from a page we could open.
• We have no two-year level. One headline says the two-year is nearing 5 per cent and we could not read its body, so we are not quoting it.
• Of the four Fed officials, three reach you through one publisher’s round-up. Only Hammack is two-sourced, and only Williams’s remarks were read in an article body rather than a round-up.
• We cannot tell you what Elderson or Schnabel will say, obviously, and we also cannot tell you whether Elderson’s panel will be reported at all. The issuer says there will be no text; it does not say there will be no coverage.
• The ECB calendar entries were read as verbatim text only after a second fetch at an offset. The first fetch returned a model summary whose spacing of the time strings differed from the page, and we are not quoting from it.
• Tokyo’s session levels and the finance minister’s remarks this morning are not scored here. Japan’s Finance Minister Katayama said at 00:47 UTC that moves at current levels look one-sided and sharp, and that Tokyo would coordinate closely with the United States Treasury. We have that from one article body and two headlines at a second publisher, we are reporting it as said, and we are not attributing the overnight fall from 160 to it or to anyone — one publisher does make that attribution and it is a single translated causal claim, which is the class of claim we keep the level from and cut the “because”.
• The published fixing this morning was 159.91 and the Tokyo range was 159.79 to 159.92. We are leaving the session analysis to the slot that can do it properly rather than appending it here.

Related

• FX events calendar — central bank speakers and release times in UTC.
• Signals — how we turn a reaction function into a sizing question.
• Presets — filter and correlated-exposure settings for policy events.