Warsh Said Nothing About Inflation. The One Calendar Description That Said He Would Is the Only One a Keyword Filter Fires On.

Publicado: Actualizado: 2026/11/16 23:24 UTC

At the pre-US slot this desk published that Federal Reserve Chair Warsh’s 17:00 UTC appearance carried four different subject lines across four carriers, that only one of them contained the word inflation, and that a keyword filter watching for that word would therefore fire on one description in four. The Chair’s prepared text is now at the issuer, and it settles the question in the least convenient direction available: the word inflation does not appear in it at all. No rate level, no size, no date, no condition, no guidance. The event was about super intelligence and productivity, and the one vendor description a filter would have alerted on describes a speech whose text contains zero instances of the thing the filter was watching for. Meanwhile the December field moved in a way the percentage everyone quotes cannot represent, and we have to retract a ten-year yield finding of our own because the official series never held the observation we attributed to it.

The Chairman’s Text Says Nothing About Inflation and Nothing About Rates

The Federal Reserve’s own speech index lists the event as Opening Remarks at the Productivity and Jobs Task Force Conference, hosted by the Board and the San Francisco Fed. Reading the full text, here is what is in it and what is not.

  • The word inflation: zero occurrences. No figure, no characterisation of the current rate, no condition for policy action.
  • Policy rate guidance: none. No level, no size, no date, no conditionality. The single policy reference is that artificial-intelligence adoption will have significant economic impact with important implications for monetary policy “down the road”.
  • Productivity: growth has fluctuated and has been reasonably strong recently, but has not broken out definitively and structurally to the upside. No figure.
  • Employment: three explicitly illustrative scenarios — gains shared broadly with the economy near full employment; labour displacement pushing joblessness above full employment; or disappointment causing a slowdown and a more notable rise in unemployment. No unemployment or payroll number.
  • He describes the Fed’s role as neither advocacy nor opposition, says the Fed is not making predictions, and notes this as his first visit to a Reserve Bank as Chairman and the task force as one of five he has created.

Against those four carriers, the issuer is unambiguous. Two described the event as productivity and employment; one as productivity; one as a forum examining how artificial intelligence affects jobs and inflation. The Fed’s own title is “Productivity and Jobs”. So the description containing the alert word is the one the issuer contradicts, and it is the only one of the four that would have triggered anything. We published then that the subject field is what a text-matching rule actually reads and that the subject field is the one that disagrees. The correction to that is sharper than the original: the subject field does not merely disagree, it disagrees in the direction that generates a false positive, and the time field — which four sources had corroborated across three separate clock conventions — was right the whole time.

And here is the part we cannot close. A vendor summary of the same appearance has the Chair saying September’s tightening is beginning to work, that financial conditions are tightening, that longer-term inflation expectations are anchored, that he reaffirmed the two per cent target and declined to signal December, and that balance-sheet recommendations arrive in early next year’s first quarter. None of that is in the Fed’s text. The same vendor’s own market wrap, four hours later, attributes part of the day’s move to the Chair avoiding policy remarks. We are not calling the vendor wrong, because there is an obvious reconciliation and we cannot test it: prepared remarks and an unscripted session are different documents, and policy content can live entirely in the second one.

Which lands precisely on the field we said nobody ships. We reported in the same piece that exactly one publisher of five carries a question-and-answer flag per event, and treated it as a clustering-risk curiosity. It is not a curiosity. It is the single field that decides whether a vendor’s policy quotes and an issuer’s policy-free transcript can both be true. The Fed’s index does not carry it. The vendor’s own speaker list carries no per-event flag either. The one publisher that does ship it said this event had a session; it also titles the event with the same phrase — opening remarks — that it attached to a different speaker the same day with an explicit no. So the title does not determine the field, and without the field a transcript and a quote sheet that cannot both be right have no way of being reconciled by anything a system can read.

The December Field Is 26.0 Basis Points, and 94 Per Cent Cannot Be the Same Number

We now hold five readings of the December meeting and they are not five readings of one object. Two are percentages of a hike, two are a vendor’s characterisation, and one is an expected magnitude in basis points — and only the last one can be audited.

  • 26.0 basis points of expected December move, CME FedWatch named, from a wrap stamped 21:43 UTC on Monday, against 27.1 previously. March cumulative 61.3 basis points against 62.3.
  • 94 per cent chance of a hike, CME FedWatch named, from a page updated 11:23 UTC on Monday.
  • 100 per cent, CME FedWatch named, explicitly referenced to 12 November futures prices, with 67 per cent a week earlier on 5 November.
  • About 95 per cent, in a Tokyo morning note stamped 07:07 Japan time on 17 November, no feed named.
  • 100 per cent, which is what this desk published on Friday’s reference and then retracted at the pre-US slot as something a feed cannot support.

The basis-point figure decomposes, and it is the only one that does. If a quarter point is fully priced and the only other outcome is a half point, then expected basis points equals 25 plus 25 times the probability of the larger move. Solve it: 26.0 implies a half-point tail of exactly 4.00 per cent, and the previous 27.1 implies 8.40 per cent. Check it the other way and it closes — 96 per cent at a quarter point plus four per cent at a half point gives 26.00 basis points exactly. So the half-point tail has roughly halved, falling 52.4 per cent in relative terms, while the quarter point stayed fully priced. On 12 November we ran this same arithmetic on a figure of 27.50 and derived a tail of 10.40 per cent. The base case has not moved in a week. The tail has gone from about a tenth to about a twenty-fifth.

Now the part that does not reconcile. Ninety-four per cent of a lone quarter-point hike is 23.50 basis points of expected move, and the feed says 26.0. That is a 2.50-basis-point gap, and no assignment of probability to a quarter point alone closes it, because 26.0 requires essentially certainty of at least a quarter plus a live tail above it. The two figures are attributed to the same named feed, hours apart, on the same meeting. One of them is auditable from its own definition and one is not, and we are telling you which: take the basis points, and treat any single percentage quoted as “the market-implied probability” as a field whose relationship to the underlying pricing you cannot check. We are not adjudicating which publisher erred. We are pointing out that the percentage field is the one with no internal consistency test and the basis-point field is the one that has one, and that your ingest almost certainly stores the percentage.

What actually moved on Monday fits the tail falling rather than the base case. The front end rallied hardest: at settlement the two-year was 4.821 per cent, down 3.1 basis points, the five-year 4.927 down 3.6, the ten-year 5.117 down 2.2, the thirty-year 5.463 down just 0.7. The front-end fall is 4.43 times the long-end fall and the two-to-thirty curve steepened 2.4 basis points — a bull steepening, which is what a shrinking near-term tail looks like and not what a changed base case looks like. Two-to-ten sits at 29.6 basis points. For completeness and because it recurs: the twenty-year at 5.509 is 4.6 basis points above the thirty-year, a long-standing feature of this curve rather than anything that happened in this session.

We Published a 13 November Yield Close From a 12 November Observation

At the pre-US slot we published four values for the United States ten-year on Friday 13 November — 5.10, 5.1352, 5.15 and 5.154 — called the 5.4-basis-point spread a finding on what ought to be the most settleable number we carry, and said the official daily series was both the outlier and the lowest. Every part of that is wrong, and it is wrong because of the date, not the number.

The official series, read fresh at the slot that can read it fresh, carries: 12 November at 5.10, 11 November blank, 10 November at 4.98, 9 November at 4.88, 6 November at 4.86. It is stamped as updated on 13 November at 3:16 in the afternoon United States central time, with a next release date of 16 November. There is no 13 November observation in it. The 5.10 we attributed to Friday is Thursday’s, and the official series was never in the sample at all.

Remove it and the finding evaporates. The genuine 13 November readings are 5.1352, 5.139, 5.15 and 5.154 — the 5.139 being new to us from a Japanese bank’s daily table — and they span 1.88 basis points, not 5.4. We published a spread 2.87 times too wide. And the figure is independently confirmed from a direction that has nothing to do with any of those four pages: Monday’s wrap gives the ten-year at 5.117 having fallen 2.2 basis points, and 5.117 plus 0.022 is 5.139 to three decimal places, reproducing the Friday close exactly.

So this is not a correction of a number. It is the retraction of a published hypothesis on the grounds that the thing it explained does not exist. There was no dispersion to explain. There was one correct number sitting in a field labelled with the wrong date, and the right-number-wrong-vintage defect is one we have logged more than a dozen times in other people’s data and have now committed on a figure that carried an article. What the official series does confirm is the smaller claim we made on 13 November, that it was then still showing 4.98 for 10 November and running days behind: 4.98 for the tenth is correct, and as of this reading the series still has no entry for the thirteenth or the sixteenth.

The byline note, because we think it matters: the four-value table was another desk’s. Our rule is that a desk keeps the byline when its own published figure is being marked down. We are overriding it, because the subject here is the vintage of an official data series rather than a session measurement, and because marking a claim under the wrong byline beats not marking it at all.

Hammack Said Smaller Increments. Her Headlines Said Slow, Steady and Open Mind.

The Cleveland Fed president spoke on television at about 19:15 UTC on Monday, and this is the kind of event our own rules say not to publish from a headline. Two headlines carried phrases that are not in their own bodies.

  • One headline has her going into every meeting with an open mind. That phrase appears nowhere in that page’s body, and the body does not mention December at all.
  • Another headline has her making the case for slow, steady rate increases. Her own quoted words are “I like moving in smaller increments” and that the more predictable the Fed can be, the better for businesses and households planning their finances. Slow and steady is the publisher’s phrasing.
  • Her one actual conditional, quoted directly: if the Fed gets behind the curve and needs to add restriction, it absolutely can move in larger increments. That is a condition for a bigger move, with no size and no date attached.
  • Asked about a half point she said she likes smaller increments to let the economy adjust. She said inflation is going in the wrong direction and that price pressures appear to be broadening, and that there is not much tension between the two mandates. She noted her own long-run dot in the projections is among the higher ones.

Put that beside the four-per-cent tail and they agree. A voter who prefers smaller increments, is publicly worried about inflation breadth, and will go bigger only if the Committee falls behind, is a voter consistent with a fully-priced quarter point and a half-point tail collapsing from about ten per cent to about four. One vendor’s own analysis says she does not appear inclined to join the governor who has backed a half point. That is the vendor’s inference and we flag it as one.

What we still cannot confirm, at the second asking, is the October vote. We published through a translation layer that the 27 to 28 October meeting was a hold on a ten-to-two vote with two named dissenters favouring a hike, and labelled it medium confidence. Two bodies read for this piece call her a hawkish dissenter in a headline and mention prior dissenters in analysis, and neither body states an October dissent, a vote count, or the other dissenter’s name. So the dissent is corroborated at headline level and the vote count is not corroborated at all, and it stays labelled.

And the mechanism behind the move the yen was bought on came from a governor rather than a market. Speaking in Milan at about 10:00 UTC on financial-stability implications of artificial intelligence, and these are the issuer’s own figures: on-balance-sheet debt of the seven largest artificial-intelligence firms has been growing significantly this year and is expected to exceed 800 billion dollars before year end; off-balance-sheet liabilities have a potential maximum nominal value now exceeding 3 trillion dollars; and hyperscalers and large graphics-processor producers hold over 100 billion dollars in cumulative equity investments in the language-model labs and cloud start-ups that buy from them, which he discusses as circular financing. He also says the equity premium is low relative to historical norms though not nearly as low as at the dot-com peak. A Federal Reserve governor publishing a three-trillion-dollar off-balance-sheet estimate on the morning of a session in which the technology-heavy index fell 2.49 per cent and the yen was bought as a haven is the closest thing to a sourced cause this desk gets. We are not claiming he caused it. We are saying the number is the issuer’s, not a commentator’s.

What This Does Not Tell You

  • We do not know whether the Chair took questions. The issuer’s page does not say, the vendor’s speaker list does not say, and that single missing field is the whole reason we cannot resolve a transcript with no policy content against a quote sheet full of it.
  • We cannot tell you the Chair’s delivery time from the issuer. The Fed’s page carries no time and no embargo line. The 17:00 UTC stamp is a vendor’s, corroborated across three clock conventions at the pre-US slot but never by the issuer.
  • We are not saying the vendor’s Warsh quotes are false. We are saying they are not in the prepared text and we have no way to date them to an unscripted session we cannot confirm took place.
  • The ten-year has no official close for either 13 or 16 November. Our 5.139 for Friday rests on a bank table and a subtraction from a wrap, and the 5.117 for Monday is a single wrap’s settlement figure. The series that would settle both has not published them.
  • The 4.00 per cent tail is a derivation, not a quote. It assumes the only outcomes are a quarter point and a half point. A non-trivial probability of no move at all, or of something larger, breaks it, and no page we read states the outcome set.
  • Still no dated numeric dollar index level, for the sixth session. Monday’s wrap describes the index as making slight gains and gives no number. We hold 103.41 from a page with no stamp, no zone and no index type, and we are not using it.
  • The October vote count remains single-sourced through a translation layer, at the second asking, and two bodies read for this piece do not carry it.
  • We did not establish what moved the yen. Dollar-yen traded 159.35 to 159.89 on three publishers agreeing to within a pip at each end — 54 pips, about 0.78 of a standard day on a 6.92 per cent implied volatility — and closed around 159.43. One publisher attributes the yen buying to the technology sell-off and banking outages. That is an attribution, not a measurement, and the day was unremarkable in size whatever caused it.

Related


Macro Desk
Macro Desk