The Curve Peaks at 20 Years, and the Dealer Is the Buyer There

The 20yr closed at 5.37% on 23 October against 5.33% at the 30yr, a 4.0bp inversion that has held for 31 consecutive sessions and has made the 20yr the peak of the curve on 109 of 204 sessions this year. Primary dealers took 12.6% of the last 3 20yr auctions against 6.0% of the last 3 30yr. Bills carried 86.4% of gross offering for 9 months with coupon supply pinned at $343bn-$352bn, so the long end has not yet been asked to fund the deficit the midterm decides.

There is a place on the US Treasury curve where the government pays more to borrow for 20 years than for 30. On 23 October the 20yr printed 5.37% and the 30yr 5.33%, so the curve peaks at 20 years and slopes down after it. That is not a forecast about inflation or growth. It is a statement about who shows up to buy. A pension fund or an insurer buys 30yr paper because its liabilities are 30 years out; almost nobody has a 20-year liability, so the 20yr has to be absorbed by the firms whose job is to warehouse what nobody else wants. The midterm election reaches this through one channel: how much the Treasury has to borrow, and at which tenor. This piece reads the auction record rather than the polls to see which of those two facts is priced.

The peak of the curve is not the longest bond

The 20yr closed at 5.37% on 23 October against 5.33% at the 30yr. That is a 4.0bp inversion, and it means the most expensive money the US government can borrow is money it borrows for 20 years, not 30. On 2 January the ordering was the normal one: the 30yr paid 5.0bp more than the 20yr. The long end has repriced by 9bp against itself while both yields rose — the 20yr is up 56bp on the year to the 30yr's 47bp.

This is not a one-day artifact. The 20yr has been the highest-yielding point on the nominal curve on 109 of 204 sessions this year, against 95 for the 30yr. For most of a year the market has said the same thing: the 20-year is the worst place on the curve to own duration.

The explanation is the buyer base, not the duration. A 20yr bond is too long for the cash and bank-treasury bid, which sits inside 2 years, and too short for the duration bid that a pension or insurer matches against liabilities 30 years out. The 20yr falls in the gap between the two natural buyers, in the one sector of the curve where neither has a reason to be.

The dealer is the buyer of the 20yr

How long that has been true is measurable. The 30yr minus 20yr spread has been negative for 31 consecutive sessions, from 10 September through 23 October, and has spent 77 of 204 sessions this year below zero. The last time it closed at or above zero was 9 September.

The auction record says who absorbed it. Primary dealers are the firms obligated to bid at a Treasury auction — the balance sheet of last resort when end investors do not show up. On the last 3 20yr auctions they took 12.6% of what was accepted. On the last 3 30yr auctions they took 6.0%. Across 2026 the means are 10.7% for the 20yr and 8.4% for the 30yr.

A dealer share is the cleanest read available on whether an auction cleared through end investors or through the Street's inventory. The 20yr is not failing — it prices, every month, within a basis point or two of the curve. It is clearing through the dealer, which is a different and more durable fact than a tail.

The spread of outcomes is wide enough to be informative. The 20yr dealer share ran from 5.8% in January to 16.7% on 15 September. The 30yr ran from 2.2% on 10 September, an auction that stopped through 6.2bp with indirect bidders taking 79.3%, to 14.7% in June. When the 30yr has a real-money bid the dealer disappears from it. When the 20yr has one, the dealer is still there.

The 21 October auction the fiscal feed still has not published

The most recent 20yr auction was a reopening on 21 October. As of this morning the Treasury Fiscal Data record for it still carries null in the total accepted, high yield and bid-to-cover fields. The result was published on the day by TreasuryDirect, and it is this: a 5.469% high yield on $13.0bn accepted, a 2.73 bid-to-cover.

Against a par print of 5.47% on the same day, that is a 0.1bp stop-through. The auction cleared through the curve rather than behind it. It still went through the dealer at 10.2%, with indirect bidders at 57.7% and direct at 31.3%.

The tails are the reason not to overread the dealer number in either direction. Across 10 20yr auctions this year the mean tail is +1.4bp and 2 of the 10 stopped through. The 30yr mean is +0.5bp with 4 of 10 stopped through. On pricing, the 20yr is the weaker of the two by under a basis point on average. On who ends up holding it, the gap is more than 4 points of dealer share. The auction is not expensive; the bond is hard to place.

Nine months of funding at the front end

None of this has been tested by supply, because the supply has gone somewhere else. Through September, bills were 86.4% of gross bill-and-coupon offering, in a range of 85.5% to 87.7% across 9 months. The coupon calendar offered $343bn to $352bn a month and held exactly 8 coupon auctions in every one of those months.

That is the stability worth noticing. Whatever the political noise of the past 9 months, the coupon calendar has not been asked to fund anything. The entire variance in monthly coupon supply across the year is $9bn. The bill share is the relief valve, and it has been held open.

October is incomplete — 4 of 8 coupon auctions had settled by the 23rd — so its 92.3% bill share is not comparable to a full month and should not be read as a jump in front-end funding. What it does show is the same behaviour: the coupons are back-loaded into the last week, and the month so far is running at the front end.

How the midterm reaches the 20yr

One: the fiscal path. The mechanism is gross issuance, not sentiment. A unified government writes a budget resolution with new spending and has to fund it. Divided government does not. That is the only channel through which a vote reaches a bond price, and it runs through the deficit rather than through expectations about the deficit.

Two: the funding mix. The Treasury chooses bills or coupons. For 9 months it has chosen bills at 86.4%, because bills add no duration and meet no term premium. A larger funding need eventually pushes that share down, not because anyone decides to term out but because the bill share cannot rise without limit.

Three: the tenor. Every dollar that leaves bills and enters coupons has to land somewhere on the curve. The 30yr has a natural buyer: indirect bidders took 70.4% of the 8 October auction. The 20yr does not: 57.7% indirect with the dealer at 10.2% on 21 October. Marginal duration therefore lands in the 20yr first, and the price of placing it there is a wider 30s20s inversion. That is the instrument to watch, and it sits at -4.0bp.

Where this is wrong

First, the spread. A 30s20s close above +2.0bp would put the 20yr back under the 30yr and say the inversion was dealer warehousing rather than an absent buyer base. The spread has not closed above +2.0bp since 20 August, when it printed +3.0bp. That is the level, and it invalidates the mechanism rather than merely hurting it.

Second, the mix. A bill share below 85% for 2 consecutive months would mean the Treasury is already terming out. The Jan-Sep range was 85.5% to 87.7%, so a print under 85% is outside anything this year has produced. If that happens the 20yr concession is being paid, not threatened, and the trade is late rather than early.

Third, the direct measurement. A 20yr dealer share below 8% on two consecutive auctions would say real money is taking the paper down. The 2026 mean is 10.7% and the low was 5.8%, so it is achievable — it simply has not happened since January.