The Issuer Says "Several". The Headlines Say Four. The Number Is Not in the Text — and We Can Show You Where It Came From.

Publicado: Actualizado: 2026/11/17 23:26 UTC

The Federal Reserve Bank of St Louis has published its president’s remarks of Tuesday 17 November, and the number every feed carried is not in them. Alberto Musalem told a conference at the Central Bank of Argentina that he expects the economy to require “several” interest rate increases before the spring of 2027. The word four does not appear in the document. It does not appear in the issuer’s own summary takeaways either. Yet headlines reaching us from a search on that speech split into two camps — one that renders it as several and one that renders it as four, or up to four — and the four is a figure a reader can only have produced by doing arithmetic on two of his other sentences and then printing the result inside quotation marks. We can show the arithmetic, because the issuer gives both inputs. What it produces is not four. It is a band from two to six, and four is the single value you get by taking the smallest of one bound and the largest of the other.

The issuer’s word is several, and the number was manufactured downstream

We read the prepared text in full, twice: the issuer’s PDF, which runs about twelve thousand characters and is marked “as prepared for delivery”, and the issuer’s HTML page for the same remarks, whose prepared-text section matches it. The document is titled Remarks on the U.S. Economy and Monetary Policy and was delivered at the 2026 Money and Banking Conference in Buenos Aires. Here is what it actually contains on the question everyone reported:

  • The count: “several interest rate increases”. No numeral, no range, no “as many as”. The issuer’s two summary takeaways use the same word.
  • The horizon: before the spring of 2027. He separately frames scenarios over the next eighteen months.
  • No meeting and no month is named for any increase. The only dated items are the September policy action, in the past, and the spring of 2027. Every mapping of a hike count onto consecutive meetings is the publisher’s, not his.
  • He does say markets expect the Committee to raise its policy rate within the next few months — which is his description of market pricing, not a timetable of his own.

What we can and cannot claim about the downstream renderings matters here, so we will be exact. We read the issuer in full. We could read one secondary publisher’s page only in part, and on that partial reading it carries four in its body as well as its headline, together with a phrase we will come to in a moment. One major outlet’s page returned a client error and we therefore cannot tell you whether its body matches its headline — only that the headline string that reached us renders the count as four. We are not asserting what any publisher’s body says on the strength of its headline. The claim we are making is narrower and harder to dodge: a specific numeral is circulating in the headline layer for a speech whose full text does not contain it, and the issuer is free, fast and unambiguous.

The reason this is a systems problem rather than a journalism complaint is that “several” and “four” are not two renderings of one field. One is a qualifier that no extractor can store as a number. The other is a number, and the moment it exists in text it becomes storable, comparable and chartable. A pipeline that pulls integers out of central-bank coverage will hold a four for this speech, with a timestamp, a speaker and a source, and nothing in the record will mark it as derived. The qualifier survives only in the camp that chose not to compute.

Where the four came from, and the one input he withholds

The issuer gives two statements about the real policy rate, and they are the inputs:

  • He expects the economy to require a real policy rate solidly between 2 per cent and 2.5 per cent before the spring of 2027.
  • He estimates the real policy rate is at the lower end of the 1 per cent to 1.5 per cent range at present.

So the gap he describes runs from 0.5 points, at the top of the range he puts it in now to the bottom of the target, to 1.5 points, from the bottom of that range to the top of the target. The secondary headline we could partly read puts the real rate “a full point” too low. One point is not a figure he gives; it is the minimum of the two bounds he gives, and it is printed as though he said it.

Now convert the gap into quarter-point moves, and add the sentence that completes his argument: it would be preferable for a higher real rate to arrive in a balanced way through both lower inflation and a higher nominal rate, however “the nominal policy rate may need to do most of the work”.

  • A 1.0-point gap, with the nominal rate doing all of the work, is four quarter-point increases.
  • A 1.0-point gap with the nominal rate doing most of it — more than half, less than all, which is what he said — is two to four.
  • Across the full gap he describes, the supported band is two to six.

The published four is the arithmetic of the narrowest version of his gap combined with the strongest version of his nominal-rate claim, and he made the gap wide and the claim weak. Take the minimum of one bound and the maximum of the other and a range collapses to a point estimate. That is the mechanism, and it is worth naming because it is not a typo and not a mistranslation: it is a defensible calculation, performed correctly, on inputs selected to produce a single number, and then presented without the word derived anywhere near it.

Two further things about that real rate, one in his favour and one against. In his favour: his own internal check reproduces. He says the real rate in the 1990s was around 3 per cent, or at least twice what it is now. Three divided by 1.5 is exactly two and three divided by 1.0 is three, so “at least twice” holds at both ends of the range he gave — it is exact at one end and conservative at the other, which is what a carefully stated claim looks like. Against him: he does not tell you the deflator. A real rate is a nominal rate minus an inflation measure, and with the policy range at 3.75 to 4.00 per cent, a real rate of 1.0 implies an inflation gauge of about 2.875 per cent. In the same remarks he says headline and core inflation continue to run higher than 3 per cent, and his own estimates for October put headline and core personal consumption expenditure inflation at 3.8 and 3.2 per cent — figures he labels as implied by the consumer and producer price releases rather than as published data. Subtract either from the policy midpoint and you get 0.075 or 0.675, nowhere near the 1 to 1.5 he states. So the real policy rate he is steering by is not computed against any inflation number in his own speech, and a derived level whose inputs are not stated cannot be reproduced, only believed.

Two of the four scheduled speakers said nothing, and the field we have hunted for weeks was sitting at the issuer

Four Federal Reserve officials were on the schedule. Ranked by how closely a rate-path system would watch them, the order is roughly the reverse of what they produced.

  • Waller, a voter, 14:10 UTC, remarks at a payments conference. Nothing on policy in anything we opened.
  • Hammack, a 2026 voter and one of the two October dissenters, 14:30 UTC, remarks on the Treasury market. Nothing we could find.
  • Musalem, a 2028 voter, 12:30 UTC, a conference in Buenos Aires. Produced the entire rate-path story above, at the issuer, in full text.
  • Williams, a voter, 19:25 UTC, a fireside chat. Produced the day’s market-moving line.

The Waller silence is the expensive one, because his own stated condition had been met. One feed carries his trigger as a half-point increase possibly being warranted if October core consumer prices came in around 0.3 per cent or higher. October core printed 0.3 against 0.2 expected. So a system watching the most-watched voter for a named, quantified, pre-announced condition got the condition satisfied and then got a payments conference. The one piece of content that predicted this was not a structured field at all: the preview we read flagged in prose that he might stay clear of monetary policy given the venue. The event type was the field that mattered and no calendar row carries it.

Which brings us to something genuinely useful. This desk has spent several slots noting that exactly one publisher ships a question-and-answer flag, that the flag changes how much unscripted risk a window carries, and that a prepared text and an unscripted session are two different documents whose reconciliation we could not perform. The issuer solves it for nothing. Musalem’s prepared text is marked as prepared for delivery and mentions no question-and-answer session. The issuer’s own landing page for the same remarks states that he took part in one, moderated by two of the host central bank’s governors. The scripted document cannot know about the questions that followed it, and the page sitting beside the document does. Stop buying this field from vendors and read the central bank’s remarks index, which carries the text, the title, the venue and the fact of a Q and A on one page.

On Williams, the headline layer failed the same way it failed Musalem, and we can prove it this time because we read the body. The verbatim line is that he does not think the situation calls for “dramatic action”. The article’s own headline says he “rules out” a jumbo half-point increase for December; the body says he poured cold water on it. Rules out is not in the body. That is two publishers in one day whose headline phrasing is absent from their own text, on two different officials, in the same direction — toward more specificity than the source supports. He also said the data gives a pretty clear signal, that the Committee has work to do on inflation, and that if artificial-intelligence investment keeps pushing semiconductor prices up it is a factor arguing that rates may need to rise further.

The percentage went up and the expected move went down, and the day’s yield change was larger than either figure we published

At the pre-US slot we said the December pricing had two fields, that the basis-point field carries an internal consistency test while the percentage field does not, and that a 2.5-basis-point gap between them could not be closed by any probability assignment. An independent publisher has now closed it, and the basis-point field was the right one.

  • Monday, named feed: 94 per cent, alongside a published expected move of 26.0 basis points. As a lone quarter point, 94 per cent is 23.5 basis points — the 2.5-point gap we complained about.
  • Tuesday, before Williams spoke: 96 per cent for a quarter and 4 per cent for a half, from a publisher that does not name its feed. That pair reproduces 26.00 basis points exactly. Our decomposition of the published figure was right and the 94 per cent was simply the wrong field to compare it against.
  • Tuesday, after Williams spoke: 98 per cent, again with no feed named. On a 98-and-2 split that is 25.5 basis points; as a lone quarter point, 24.5.

So across Tuesday the probability rose from 96 to 98 while the expected move fell from 26.0 to about 25.5, and both readings are correct about their own field. A half-point tail coming out of the curve raises the chance that something happens and lowers how much is expected to happen. A store holding the percentage reads that afternoon as more hawkish; a store holding basis points reads it as less. Williams removing a jumbo move is precisely a tail-compression event, which is the mechanism, and the 25.5 is our arithmetic on an assumed split rather than a published number. Neither Tuesday percentage names a feed, so we give you the band of 96 to 98 and decline to call any of it the market-implied probability.

The curve agrees with a tail coming out rather than a base case moving. On readings stamped 18:03 UTC: the two-year fell 4.7 basis points to 4.783, the ten-year 4.1 to 5.091, the thirty-year 2.6 to 5.448, with the ten-year breakeven up 0.8 to 2.220. The front end moved 1.81 times the long end, against 4.43 times on Monday — still a bull steepening, less lopsided. And because the breakeven rose while the nominal fell, the ten-year real yield fell 4.9 basis points on the day Musalem argued the real policy rate must rise by one to one and a half points. That is not a contradiction — one is a policy rate and the other a ten-year market rate — but it is the gap he has to close, moving the wrong way at the long end on the day he described it.

Now the part that marks our own work down. At the pre-US slot we gave the ten-year at 5.105 and offered the day’s move as either minus 1.20 or minus 2.61 basis points, depending on which of two publishers’ Monday levels you used. Both are too small. The 18:03 reading carries its own differenced field of minus 4.1 basis points, and that pair is internally consistent: 5.091 plus 0.041 implies a Monday level of 5.1320, which sits 0.09 of a basis point from one publisher’s Monday figure of 5.1311 and 1.50 basis points from the other’s 5.117. Two of three publishers now agree on Monday’s close to within a tenth of a basis point. We previously declined to convict either, on the rule that a vendor is not convicted on consistency alone; we now have the observation that was missing, and the evidence favours the higher Monday level. Our two candidate day-moves were short by factors of 3.42 and 1.57 for one reason: we computed a daily change from a level read at hour ten of the session. A differenced field built on a mid-session level is a partial-day move wearing a daily label.

The same error, in a different field, on the same day, was already marked down once — and the closing data has now confirmed the correction. At the pre-London slot the desk called the dollar-yen session 0.726 of a standard day; at the pre-US slot it corrected that to 1.186 on a 74-pip range. The four-values table for 17 November gives the day at 159.47 open, 159.61 high, 158.87 low, 159.38 close — a range of exactly 74.0 pips, which is 1.186 standard days on the one-week implied volatility we used. The correction was right to three decimal places and both extremes survived the remaining twelve hours untouched. The pre-London figure was 1.45 times too small. Note what that does to the mark this desk proposed at the pre-US slot about re-reading whichever extreme sits nearest to price: at the pre-US slot the low was the near extreme and the high the far one, and neither moved again. The hazard argument is about probability and a single negative draw does not refute it, but the first out-of-sample test of that prescription would have spent its fetches on a number that was already final, and we are recording that rather than quietly dropping it.

Wednesday 18 November, with every time in UTC:

  • 19:00 — the minutes of the 27 to 28 October meeting, at which the Committee held its target range at 3.75 to 4.00 per cent. Double-sourced and both non-issuer, with the hour now carried by a live feed as well as a calendar. This is the document that can settle, at the issuer, whether the two October dissents were for an increase; one feed now names both dissenters, where a week ago we had the claim from a single publisher.
  • 15:00 — the New York option cut, carrying 160.50 and 159.75 on the structured order board, neither of them with a size band, on a third consecutive structured read.
  • 23:50 on Tuesday 17 November — Japan’s October trade balance, 08:50 Japan time Wednesday.
  • Euro-area October inflation, United Kingdom October inflation, and a 20-year auction we carried from an earlier calendar and could not confirm: the one live feed we read lists only bill auctions for the 18th and 19th, which is a negative from a single source and therefore weak.
  • Context from the same feed: six-week bills cleared at 4.055 per cent with a bid-to-cover of 2.82, and the Atlanta Fed’s fourth-quarter growth nowcast is unchanged at 3.4 per cent.

One standing absence closes here, and it closes against us. For more than ten consecutive sessions this desk has printed that it cannot give you the notional on the 160.00 dollar-yen strike, because the only board carrying size bands leaves that cell empty. A second publisher’s expiry list for the 17 November cut gives 160.00 at 1.06 billion dollars and 157.50 at 1.81 billion. The board bands 157.50 as large and leaves 160.00 blank — and the board’s own footnote defines large as one billion units or more. So the blank cell was not a small order and not an absent option. It was a 1.06-billion-dollar option that the publisher’s own vocabulary would have labelled large. Where the band exists it is correct; where it is missing it is missing data. That is a mark-down of a claim this desk published under another byline, and the desk that made it keeps the correction, so the full scoring belongs to the next flow piece rather than to this one. We are closing the absence and naming the debt.

What this does not tell you

  • We cannot tell you what any secondary publisher’s body says about the hike count, except one, and that one only in part. One major outlet’s page returned a client error. Our claim is about a numeral circulating in the headline layer against an issuer text that does not contain it — not about what each newsroom wrote.
  • The two-to-six band is our arithmetic, not his forecast. He said several and gave the inputs; we did the conversion to show where a four can come from, and publishing our own derivation as the answer would repeat the error we are describing.
  • We never found Tuesday remarks from Waller or Hammack. Absence of coverage in the four or five places we looked is not proof they said nothing about policy. If either did, the third section is weaker than it reads.
  • Waller’s half-point trigger and the October core consumer price print come from two different secondary sources, neither an issuer. The condition-was-met claim rests on both being right.
  • Neither Tuesday December-probability reading names its feed, so the 96-to-98 band is not comparable to Monday’s 94 per cent in the way a single series would be. The 25.5 basis points is ours, on an assumed split.
  • Every yield in the fourth section is stamped 18:03 UTC, which is not a close. One other publisher carries the ten-year near 5.10 with a change of about 2 basis points on an undated snapshot; we are not adjudicating that against the 18:03 pair, because two readings at two different times are not two readings of one object, and one of those times is unknown.
  • The real-rate deflator is the most interesting thing we could not establish. If someone can reproduce a 1 to 1.5 per cent real policy rate from a published inflation series and a 3.875 per cent midpoint, that would tell you more about his reaction function than the hike count does.
  • Nothing here is a view on direction. A speech that gives a range and a market that prices a point estimate are a sizing problem, and the question it should change is how wide your window is around 19:00 on Wednesday, not which way you lean into it.

Related

Economic calendar and event windows · Current signals · EA presets

Sources, read 17–18 November 2026.

  1. Federal Reserve Bank of St Louis, Remarks on the U.S. Economy and Monetary Policy, Alberto Musalem, Central Bank of Argentina, 17 November 2026 — the issuer’s prepared text, read in full. stlouisfed.org
  2. Federal Reserve Bank of St Louis, the issuer’s landing page for the same remarks, carrying the summary takeaways and the question-and-answer fact. stlouisfed.org
  3. Yahoo Finance, report of Williams at a fireside chat, 17 November 2026 — the source of the verbatim dramatic-action line and of the headline-versus-body gap. finance.yahoo.com
  4. The Private Banker, reports on Williams and on Musalem, 17 November 2026 — read only in part; the hike count and the full-point phrase are from that partial reading and are labelled as such above. leprivatebanker.com
  5. Newsquawk, the Fed-speak preview for 17 November 2026, which supplied the schedule, the venues and the prose caveat about the payments conference. newsquawk.com
  6. Newsquawk headline feed, 17 November 2026 — the minutes time, the dissent naming, the bill auction, the growth nowcast. newsquawk.com
  7. Newsquawk, notable option expiries for the 17 November New York cut, published 08:52 UTC — the source of the 1.06 billion and 1.81 billion notionals. newsquawk.com
  8. DZH Financial Research order board via Myforex, published 21:05 UTC 17 November against a 159.22 reference, read in full including its size-band footnote. myforex.com
  9. Minkabu FX, the four-values table for 17 November and the 18:03 UTC market readings, published 18 November 2026 Japan time. fx.minkabu.jp

Pages we tried and could not open, so that no figure above is implied to rest on them: one major outlet’s report of the Musalem speech returned a client error, which is why this piece makes no claim about that publisher’s body text. The dated option-expiry page for the 18 November New York cut failed for a third consecutive attempt, so the notionals above are for the 17 November cut and not for Wednesday’s. A per-indicator page that states its own time-zone conversion rule also returned a client error. Four article pages failed on their bare addresses and were recovered with a query-string variant; their figures come from the recovered pages, read in full.

All times UTC. Figures are as published by the sources named; commentary and interpretation are our own. Where a claim rests on a single publisher, or on a page we could read only in part, we have said so in the text rather than in a footnote.

Nothing here is investment advice. This is market commentary for system traders and it is not a recommendation to buy or sell any instrument. Trade your own research and your own risk limits.


Macro Desk
Macro Desk