On the Treasury’s own par yield curve, the 20-year has closed above the 30-year on each of the last three business days — 5.38 against 5.34 on 18 September, and 5.33 against 5.29 on both 21 and 22 September. Four basis points, three sessions, no drift. The rest of the curve is conventionally upward-sloping: two-year 4.71, ten-year 4.96. Only the last ten years of it run backwards. And at 09:30 UTC this morning, inside the London session, Germany reopens a 2047 and a 2056 — the two points where the American curve disagrees with itself — in the same hour, for one billion euros apiece.
The inversion is at the long end, and it is not the one people mean
When a desk says “the curve inverted” it almost always means twos against tens, because that is the spread with a recession literature attached to it. That spread is not inverted. On 22 September the Treasury’s daily par yield curve put the two-year at 4.71 and the ten-year at 4.96: twenty-five basis points of positive slope, and it has been positive all week.
The inversion is further out. The 20-year point closed at 5.33 and the 30-year at 5.29. That is not a rounding artefact and it is not one day’s quote — the same four-basis-point gap sits in the 18 September row (5.38 and 5.34) and the 21 September row (5.33 and 5.29), which is every business day since the Bank of Japan moved. Three prints, one sign, one magnitude.
We are not going to tell you what it means, because we cannot establish that from a table of yields, and the honest version of this is duller than the macro version: a par curve is a fitted construction, the 20-year sector carries its own supply and its own buyers, and a four-basis-point kink between two points ten years apart is a small number in absolute terms. What we will say is what it does to you operationally.
“Long end” is not a point, and this week it is not even monotone
If you hold a position keyed to the long end of the US curve — a yield-differential filter on USD/JPY, a correlated-exposure rule that treats 20s and 30s as the same risk, a news window that widens when “long rates” move — you are implicitly picking a point on a line you assume is monotone. It is not, at the moment, and it has not been for three sessions.
The practical consequence is small and specific: a rule written against the 30-year and a rule written against the 20-year have been telling you different things about the level of long-dated yields all week, by four basis points, in opposite directions from the ten-year. That is not enough to change anybody’s direction. It is enough to change which side of a threshold you land on if your threshold happens to sit in the gap. Go and look at which maturity your filter actually reads before you assume the question is academic.
The rest of the move since the BOJ decision is easier to summarise. From 18 to 22 September the two-year fell five basis points, the ten-year five, the twenty-year five and the thirty-year five, while the belly moved less — the three-year two, the five-year three, the seven-year four. The wings came down together and the middle lagged. Whatever repriced last week, it did not reprice as a slope trade.
The auction is not the curve, and the gap between them is not a tail
Yesterday’s two-year note auction cleared at a high yield of 4.787%. The Treasury’s own par curve marks the two-year at 4.71 for the same date. A third reader, Trading Economics, has it at 4.75. Three numbers for “the two-year on 22 September”, spread across seven and a half basis points, and the temptation is to subtract the smallest from the largest and call the difference a concession.
Do not. They are three different objects. The auction high yield is the clearing price of one specific new security at one moment in the early afternoon in New York. The par curve is a fitted constant-maturity construction built from bid-side quotes near the close, which is a different instrument at a different time of day. The vendor quote is a third thing again. Subtracting across them produces a number with no owner.
A tail is the auction high yield minus the when-issued yield immediately before the bidding deadline, and we did not read a when-issued level yesterday, so we are not publishing a tail. This desk has said before that you should do the when-issued subtraction yourself and show it. We could not, so we are not asserting one — and the 7.7 basis points between 4.787 and 4.71 is emphatically not it.
Germany sells the same two points today, and your calendar shows one of them
The German Finance Agency’s own issuance calendar lists two reopenings for 23 September, both for one billion euros: DE000BU2T000, a 3.40% coupon maturing 15 May 2047, and DE000BU2D012, a 2.90% coupon maturing 15 August 2056. That is roughly the twenty-year point and roughly the thirty-year point, sold in the same session — the exact pair the US par curve has inverted.
Forex Factory’s calendar carries one event for this, a “German 30-y Bond Auction”, listed as tentative with no time, with a previous of 3.90 and 2.4. If you plan your morning off that line you will have half the supply on your board. The aggregator’s figures are correct as far as they go: the Finance Agency’s published result for 16 September shows the same 2056 ISIN reopened at an average yield of 3.90% with a bid-to-cover of 2.4.
What the aggregator does not carry is the more interesting half of that result. Germany offered 1,500 million euros of the 2056 last week and allotted 1,257.41 million. Roughly 242.6 million, a little over sixteen percent of the offering, was retained rather than sold. Retention is a normal mechanic in German auctions and not by itself a failure, but a cover ratio quoted without the allotment beside it is a ratio whose denominator you have not checked — and this desk has spent three weeks on the theme that an auction metric without its definition attached is a number you cannot use.
One more thing worth knowing if you are reconciling sources this morning: Investing.com’s German 30-year Bund auction history still shows 12 August 2026 at 3.650% as its most recent entry and does not carry the 16 September result at all, while FX.co reported the 3.900% the same morning it happened. The primary source settles it. Go there.
What today actually does to the picture
Germany’s curve is the conventional shape: two-year 3.18, ten-year 3.44, thirty-year 3.79 on 22 September, thirty-five basis points of positive slope from ten to thirty and no kink we can see. The last 2056 auction cleared at 3.90%, eleven basis points above where that maturity is quoted in the secondary market now. Today’s reopening is smaller — one billion euros against last week’s 1.5 billion offered.
So the London session gets two pieces of long-dated European supply into a curve that is behaving itself, while the American curve is doing something mildly strange at the same two maturities. We are not going to dress that up as a trade. The euro spent the Asian session making a fresh low since 29 July around 1.1425 on a firm dollar, with the dollar index near 100.70, and nothing in a one-billion-euro reopening is going to be the reason it stops or continues. The supply is worth knowing about because it puts a known flow into a known hour, not because it points anywhere.
What this does not tell you
It does not tell you why the 20-year sits above the 30-year. We have the fact from the issuing agency and no sourced explanation, and we have deliberately not manufactured one. It does not tell you whether the gap persists past this week; three sessions is three sessions.
It does not tell you where the German 20-year sits, because we did not find a published level for that maturity and we are not going to interpolate one and present it as a reading. It does not give you the time of today’s auctions from a primary source — the Finance Agency calendar gives the date and the volumes, the aggregator gives this series a 09:30 stamp on past results, and we have taken the 09:30 from the aggregator rather than the issuer.
And it does not tell you what the two-year auction’s concession was. That requires a when-issued level we did not read, and a desk that will not publish a tail it cannot compute is more use to you than one that estimates it.
Related
- FX events calendar — the supply and release schedule in UTC
- Signals
- EA presets
