The United States Treasury changed its forward-guidance sentence yesterday, and it changed it by deleting two words. The August statement said Treasury “anticipates maintaining nominal coupon and FRN auction sizes for at least the next several quarters.” The November statement, on the reading of the only wire we could open, says the same thing without “at least.” We published two hours before the release that four banks had four forecasts for that sentence and that three of the four were forecasts about wording. All three were wrong, and they were wrong in an instructive way: every one of them predicted a substitution, and what happened was a subtraction.
A deletion is the one outcome a text-keyed reader cannot classify
Hold the two sentences next to each other. August: maintaining sizes “for at least the next several quarters.” November: maintaining sizes “for the next several quarters.” Two words gone, nothing added.
Now ask what your system reports. A reader keyed to a diff on the guidance paragraph flags a change — the string moved, the paragraph is not the paragraph. A reader keyed to the phrase “next several quarters” reports no change, because the phrase is still there, intact, in the same clause of the same sentence. Both readers are correctly implemented. They return opposite answers on the same release, and the answers are not a matter of degree: one says the anchor held and one says the anchor moved.
That is not a hypothetical about parsing. It is what the wire reports actually did. Reuters leads on the change — Treasury dropped the term “at least” — and a second reader headlines the same release as showing no expected changes to coupon auction sizes, which is also true, because the sizes genuinely did not change. If your news handler widens on “Treasury shifts language” and stands down on “no changes to auction sizes,” yesterday fed it both from the same thirty seconds.
The semantic content of the deletion is real and it points one way. “At least the next several quarters” is a floor: not before then. “The next several quarters” is a span with no floor under it. Removing a floor is the cheapest way a borrower can tell you that the date it previously would not go before is now a date it might. Reuters pairs the deletion with a new line that Treasury will evaluate future size changes based on trends in structural demand and the potential costs and risks of various issuance profiles, which is the same message in longer form.
Scoring the four banks: the conditional held, the three point forecasts on wording did not
We published the four house views and declined to pick one. Marking them:
- Citigroup expected reductions in 20-year and 30-year issuance announced this month. Wrong. The sizes did not change, and the guidance now says they will not for several quarters.
- UBS — Phoebe White — said a change in the guidance was more likely than an immediate adjustment to issuance itself. Right, and right as stated: the guidance changed and the issuance did not. This is the only one of the four that scored.
- JPMorgan expected the “next several quarters” language to survive through next year with sizes unchanged until August. Half right, and which half depends on the reader above. The phrase survived; the sentence did not. The sizes did go unchanged.
- Barclays expected the phrase to become “coming quarters,” with shifts in the timing and composition of issuance rather than in size. Wrong on the words — “coming quarters” appears in no source we read — and right on the second clause, for a reason Barclays does not appear to have named. See the next section.
This is the fourth consecutive time on this channel that a house’s conditional threshold has beaten everybody’s point estimate. UBS did not forecast a form of words. It forecast an ordering: wording moves before sizes move. An ordering is testable, it survives the exact phrasing being unguessable, and it is the shape a system can actually hold a position against. Three desks tried to guess a string and the string was not in any of their outcome spaces.
We should be honest about our own role here. We did not pick a winner, which was correct, but we also framed the event as having a textual outcome space with four candidate strings in it. The actual outcome space included “the same string, minus a qualifier,” and we did not put that on the list either.
The change nobody forecast is in the auction calendar, not the paragraph
Reuters reports that Treasury will convert the 7-year note into a quarterly auction with two reopenings. That is not a size change and it is not a wording change. It is a change to the shape of the issuance calendar, and for anyone running scheduled news windows it is the most consequential line in the release.
A monthly new-issue 7-year is twelve first-auction events a year. A quarterly auction with two reopenings is four new issues and eight reopenings, which is still twelve dates but no longer twelve events of the same kind: reopenings price differently, draw different demand, and are a weaker signal than a fresh tenor. Any schedule that keys a window to “7-year note auction” and sizes it uniformly is now sizing eight of twelve dates wrong, in the direction of too wide.
The rest of the mechanics, all secondary: buybacks of up to $70 billion across the curve for liquidity purposes, with Treasury continuing to evaluate enhancements such as changes to maximum purchase amounts, plus up to $25 billion in the one-month to two-year bucket for cash management. Short-term bill auction sizes are expected to be cut in December and raised across the bill curve in January, on seasonal fiscal outflows — a stated, dated, two-step change that is more actionable than the guidance sentence everybody watched. One reader assumes an $850 billion cash balance at end-December and a Treasury General Account peaking at $1.025 trillion plus or minus $50 billion in late November and late January.
The knowable part was knowable, and we got it right
We published, pre-announcement and labelled as secondary, that next week’s refunding would be $58 billion of three-year notes on 9 November, $42 billion of ten-year notes on the 10th and $25 billion of 30-year bonds on the 12th, for $125 billion, with no Wednesday auction because 11 November is a market holiday. All of that is confirmed: the sizes, the total and the three dates. One reader adds that the $125 billion refunds roughly $97 billion of privately held notes maturing 15 November and raises roughly $28 billion in new cash.
That is worth stating plainly because it defines where the uncertainty actually was. The dollar figures were derivable from a document published in August and a borrowing estimate published on Monday, and a desk that spent yesterday morning trying to forecast them was spending effort on a solved problem. The prose was the only unsolved part, and the prose was unsolvable — not hard, unsolvable, because four well-resourced houses with the same information produced four strings and the correct string was a fifth.
The operational conclusion has not changed from what we published before the release. On an event whose outcome space is textual, there is no such thing as being correctly positioned. There is only being smaller through the window, and the width of the window is set by how long it takes the wires to agree on whether anything changed — which, on the evidence of two publishers leading in opposite directions, is longer than the release itself.
What this does not tell you
We did not read the statement. Treasury’s own quarterly refunding index names the November policy statement and links it, and that link returned a client error on every attempt. Every word of the new sentence in this article reaches you through a wire. We have the August text from the issuer and the November text from a reporter, which is not the same standard, and the one-word difference between two secondary renderings — one writes “FRN,” the other “floating-rate note” — is exactly the kind of gap that a direct read would close and a summary will not.
It does not tell you who signed it. One reader attributes the statement to the Assistant Secretary for Financial Markets; the August statement we did read was issued by a Deputy Assistant Secretary for Federal Finance. We are not going to assert a name off one machine-readable page.
It does not give you a yield reaction, and we are refusing one again. Three readers with no common clock: one has the ten-year down 7 basis points to 4.90 per cent with the two-year down 5 to 4.59 and the 30-year down 3 to 5.30, explicitly intraday; a Tokyo desk has the ten-year rising back to 4.96 and then falling to 4.9; a third has it moving from the 4.96 area to the 4.89 area. Those are compatible with each other and with almost any story. None of them is a close, none of them isolates the 13:30 statement from the ADP print fifteen minutes earlier or from an 8 per cent collapse in crude, and an issuance article priced off a borrowed handle is worse than one priced off none.
It does not establish how long “several” is. Reuters reports that many analysts read it as two quarters, which would put changes in the first half of 2027; the same piece has dealers expecting the cuts, when they come, at 20 years and longer, funded by more issuance in the 2-to-7-year sector and possibly in ten-year notes. Those are analyst views about an undated intention, and the deletion of “at least” makes them more likely to be early rather than more likely to be right.
And it does not tell you that any of this moved price. The day belonged to an election result and to crude. A reader who widened for the refunding and narrowed for everything else had the window in the wrong place.
Related
- FX event calendar — the 9, 10 and 12 November auctions, and the February refunding already dated 1 February 2027.
- Signals — why an ordering you can test beats a string you have to guess.
- EA presets — news windows keyed to a 7-year auction that is about to stop being a monthly new issue.
