The twenty-year bond auction this desk has been chasing for two days cleared at 5.420% yesterday at 17:00 UTC — a record for the security in its modern form, 21.6 basis points above the August sale, with a two-basis-point tail and the weakest indirect bid in the auction’s six-year life. Three publishers had the result within hours. The calendar page this desk checked first still shows an empty box, more than eighteen hours later. Both of those facts matter today, because the Federal Reserve raises rates at 18:00.
The number, and the subtraction we owe you
Treasury sold $13 billion of twenty-year bonds at a high yield of 5.420%. The previous auction, on 19 August, cleared at 5.204%; July’s cleared at 5.163%. That is 21.6 basis points of repricing in a single month on the part of the curve that funds nothing and signals everything.
This desk has a standing rule against loose auction language, so here is the working rather than the label. When-issued traded at 5.40% into the sale. The auction cleared at 5.420%. 5.420 minus 5.40 is 0.02 percentage points — a two-basis-point tail. That is a genuine tail, not a rounding artefact, and it is small. Demand did not collapse at the margin; it was simply priced two basis points cheaper than the screen said it would be.
Bid-to-cover was 2.57 — $33.38 billion of bids against $13.00 billion accepted, which divides to 2.568 and rounds where the publisher put it. On the headline metrics this was an unremarkable auction at a remarkable yield.
The number underneath is the one that should bother you
Indirect bidders — the bucket that holds foreign central banks and overseas asset managers — took 52.5%, about $6.8 billion. Our source describes that as the lowest indirect share in the six-year existence of the modern twenty-year auction. A second publisher independently puts roughly $6.8 billion with foreign central banks and asset managers, which corroborates the notional if not the superlative.
Read those two facts together and the auction stops being unremarkable. A record clearing yield with an ordinary bid-to-cover and a record-low indirect share means the paper was taken down, but by a different set of hands than usual. Domestic buyers and dealers absorbed what foreign accounts did not. That is a composition change, and composition changes are the ones that show up in the currency rather than in the yield.
If you trade dollar pairs on a rates thesis, the thesis you have probably been carrying is that higher US yields pull the dollar up. That mechanism runs through foreign demand for the paper. An auction where the yield rises and the foreign bid falls to a series low is the configuration in which that mechanism is weakest, because the yield is not being paid to attract the flow you are assuming.
“Record” is a six-year word here
The 5.420% broke the previous record set in October 2023, when the twenty-year sold at 5.245%. We are printing that superlative because our source is specific and dated, and because this desk has been wrong about superlatives before and prefers them anchored.
But the anchor is short. The modern twenty-year bond was only reintroduced in 2020, which our source says in its own words when it calls this the lowest indirect share “in the six-year existence” of the auction. A record in a six-year series is a six-year high, not a generational one. That is not a reason to discount it — a six-year high in a long bond a day before a hike is still information — but it is a reason not to let the word “record” do work in your head that the data cannot support.
For scale on the rest of the curve, the same source has the ten-year at 5.0% and the thirty-year at 5.37%, and describes those as the highest since 2007. This desk independently established a 5.0390% ten-year print and the 2007 comparison from two readers yesterday, so that part is well founded.
Eighteen hours, and the box is still empty
At 11:37 UTC today — eighteen hours and thirty-seven minutes after the sale — the economic-calendar page this desk uses for auction results still shows the 15 September twenty-year row with no actual figure at all, and 5.204% sitting in the Previous column. This desk recorded the same field empty at 23:10 last night, six hours and ten minutes after the auction. It has now been empty for three times that long.
That is not a complaint about a vendor. It is a specification for your system. If you have anything that reads an auction result off a calendar feed — a filter, a risk switch, a note-to-self — it did not fire yesterday and it has not fired today. Meanwhile the result was readable in full from three independent publishers within hours of the close. The information existed; the feed did not carry it. Those are different failures and only one of them is yours to fix.
There is a second trap in the same place, and this desk walked into it yesterday. A search for this auction returns an article headlined “$18 billion of 20-year bonds at a high yield of 5.204%”, which looks exactly like the answer. It is dated 19 August. We now know why it looked so right: 5.204% genuinely is this auction’s previous yield, which is precisely what makes a month-old article indistinguishable from today’s at a glance. A search result that resolves your open question on the first try deserves more suspicion, not less.
What this means for the next six hours
The Federal Reserve announces at 18:00 UTC, with the Summary of Economic Projections and the dot plot alongside, and a press conference at 18:30. That is under six and a half hours from this writing. A 25 basis point move to 3.75–4.00% from 3.50–3.75% is expected; one reader put market pricing at 90% this morning and another had the CME FedWatch tool at 92% yesterday, so the decision itself is not the event. The projections are.
The trade here is a sizing decision and not a direction. You are being asked to hold rate-sensitive exposure into a dot plot on the day after the market demanded 21.6 extra basis points to take twenty-year paper and the foreign bid came in at a series low. That combination argues for smaller, not for short. A hawkish dot plot into a long end already clearing at record yields is the scenario where the move is largest in both directions, because there is no comfortable position to be squeezed out of.
One more date worth writing down: yesterday’s twenty-year settles on Friday 18 September, which is Bank of Japan day. The BOJ decision lands around 03:00 UTC, under forty hours from now, and the Bank of England sits between them at 11:00 UTC tomorrow. Settlement flow and a central bank meeting in the same session is not a forecast, but it is a reason to know which of your positions are open on Friday morning.
What this does not tell you
- We did not read the Treasury’s own auction results release. Every figure above — the 5.420%, the $13 billion, the 2.57, the 52.5% — is a secondary reading of a primary document we did not reach. The yield, size and bid-to-cover are each carried by three independent publishers, which is why we are printing them. The bidder breakdown and the when-issued level come from one.
- Two of our own numbers do not quite reconcile and we have not resolved it. Our source gives the indirect share as 52.5% and the notional as $6.78 billion. On a $13 billion auction, $6.78 billion is 52.15%, and 52.5% would be $6.825 billion. The gap is trivial in dollars and it is still a gap. We have published both figures rather than quietly picking the one that suits the sentence.
- “Lowest indirect share in the auction’s history” rests on one reader. A second publisher independently gives the notional but makes no claim about the ranking, and a third carries the direction in a headline we did not read in full and are therefore not citing. Treat the ranking as one desk’s view and the notional as well founded.
- We cannot tell you what the dot plot will say. One reader reports that the chair again declines to submit his own projection, which would be a notable feature of tonight’s release if true. That is a single source and we have not corroborated it. Do not size anything on it.
- The composition argument is an argument, not a measurement. We have one auction. A single low indirect print is an observation; a change in the foreign bid is a trend, and one data point cannot establish one.
