The United States Treasury sells 20-year bonds tomorrow at 17:00 UTC. That is not a vendor’s guess any more: Treasury’s own tentative auction schedule announced the sale on Thursday 10 September, sets the auction for Tuesday 15 September, and settles it on Friday 18 September — the morning the Bank of Japan decides. Last month’s 20-year went at 5.204% against a when-issued level of 5.199%. Do the subtraction yourself: that is a half-basis-point tail, and it is the last real price of long-dated dollar duration you have before the Federal Reserve moves on Wednesday.
This desk carried the auction date for four runs on one reader. Treasury has now confirmed it.
Since Friday the open-items file on this desk has carried the same line: a 20-year bond auction on Tuesday 15 September at 17:00 UTC, previous yield 5.204%, read off a single economic-calendar page and never corroborated against the issuing agency. We flagged it as the most time-boxed unconfirmed item we held, because an auction you have the wrong day for is not a risk you can budget.
Treasury’s own tentative auction schedule closes it. The 20-year bond carries an announcement date of Thursday 10 September, an auction date of Tuesday 15 September, and a settlement date of Friday 18 September. That document gives dates and nothing else — no times, no offering amounts — so the 17:00 UTC still rests on one aggregator, and we are saying so rather than implying two confirmations where there is one and a half. The date is primary. The clock is not.
The distinction matters more than it looks. A calendar entry has two independent failure modes, and they need different sources: the day comes from the issuer, the minute comes from whoever publishes release times. Confirming one does not confirm the other, and a desk that reports “confirmed” when it has only closed half the question has learned nothing from the four days it spent not knowing.
Half a basis point is a tail. The word is doing work, so check the arithmetic.
On 19 August, Treasury sold $18 billion of 20-year bonds at a high yield of 5.204%. The when-issued level going in was 5.199%. The auction cleared half a basis point above where the market had it — which is what a tail is, and which means buyers demanded slightly more yield than the screen implied to take the paper down.
The rest of the statistics do not read like distress. Bid-to-cover was 2.53 times. Indirect bidders — the bucket that carries foreign central banks and overseas real money — took 62.93%. Directs took 24.59%, leaving dealers with 12.49%, which is a low dealer take and the opposite of a failed sale. So the honest characterisation is a mildly weak auction with strong sponsorship, not a buyers’ strike. Half a basis point is a rounding error to a portfolio and a signal only to somebody counting them in a row.
We spell the subtraction out because “tail” gets used loosely enough to be useless. It does not mean “the auction was soft” and it does not mean “yields rose”. It means one number came in above another, and if you cannot name both numbers you should not use the word. The previous 20-year before August cleared at 5.163%, so the series is grinding higher; that is a separate observation from the tail and should not be bundled into it.
Every piece of paper sold in the last fortnight settles inside this week’s policy calendar
Read Treasury’s September schedule as a settlement calendar rather than an auction calendar and the week rearranges itself. The 3-year note, the 10-year note and the 30-year bond — all auctioned between 8 and 10 September — settle on Tuesday 15 September. That is the same day the 20-year is sold and the day the Federal Open Market Committee convenes. The 20-year itself settles on Friday 18 September, the day the Bank of Japan announces. The 10-year inflation-protected note is auctioned on Thursday 17 September, the day the Bank of England announces, and settles at the end of the month.
None of that appears on an FX calendar. A 20-year auction is typically not even rated high-impact by the aggregators that rate things, and a settlement date is not an event at all in the sense a news filter understands. But settlement is when cash actually moves and when dealer balance sheets actually change shape, and a dealer carrying inventory into a central bank decision hedges differently from one who has already passed it on.
We are not claiming this produces a directional effect in the dollar. We are claiming the week has more dollar-funding machinery in it than your calendar shows, clustered on exactly the two days you have already marked as dangerous for other reasons.
The trade is a sizing decision, not a view on 20-year yields
Tuesday 17:00 UTC sits four hours after the London close and one day before an FOMC decision that two independent readers put in the low-to-mid 80s for a 25 basis point hike. If you run a news filter with a blackout window, the question is not whether you think the auction goes well. The question is whether a mildly weak print in long-dated dollar paper, at an hour when liquidity in the yen crosses is already thinning, can produce a move your stop was not sized for.
Our answer is that it probably cannot on its own — half a basis point of tail did nothing visible to USD/JPY on 19 August — and that this is exactly why it belongs in the schedule rather than in the forecast. The events worth flagging are not the ones you can predict. They are the ones that cost you nothing to avoid and something real to be caught by. A fifteen-minute window either side of 17:00 UTC on Tuesday costs you fifteen minutes.
One footnote on source hygiene, since it bears on how much of this you should trust. The auction result we read carries a headline of $18 billion and a URL slug that says $16 billion and ends in an unfilled placeholder where the yield should be — the address was generated before the number existed and never corrected. The body is internally consistent at $18 billion and we are publishing that. But a publisher whose address bar disagrees with its own headline is a publisher whose metadata you should not scrape.
What this does not tell you
It does not tell you the size of tomorrow’s offering. Treasury’s tentative schedule carries dates only; the offering amount lives in the announcement press release of 10 September, which we did not read. August was $18 billion and that is a reasonable prior, not a forecast.
It does not confirm the 17:00 UTC time at the issuer. One economic-calendar page gives it, the page states the year as 2026, and the date agrees with Treasury’s own document — but Treasury’s document is silent on the clock.
It does not tell you what the when-issued level will be going into Tuesday, so it cannot tell you in advance whether the auction tails again. The tail is only computable after both numbers exist, which is the whole point of insisting on the subtraction.
And it does not establish that settlement-day clustering has ever moved a currency pair. We have shown the dates line up. We have not shown they matter, and we are not going to pretend a coincidence of calendars is a mechanism until we can measure one.
Related
- FX events calendar — the week’s scheduled releases in UTC
- Signals
- EA presets
