2Y 4.800 =5Y 4.890 ▼10Y 5.070 ▼30Y 5.410 ▼VIX 24.56 ▼MOVE 98.15 ▼DXY 103.79 ▲USD/JPY 159.41 ▼GOLD 3,756 ▼COPPER 5.71 ▼WTI 81.87 ▼BTC 63,115 ▼ETH 1,783 ▼S&P 7,324 ▼NDX 27,116 ▲MSTR 96.50 ▼
Rates · Liquidity · Scarcity · Credit · Power · 2026-11-18

The 20yr Year Bond Is the Hardest Thing the Treasury Sells, and It Sells the Least of It

The full market version

When the US Treasury sells a bond, it does not get the market price for free. It has to pay a small premium to whoever agrees to take the whole issue at a fixed moment, and that premium is visible afterwards as the gap between the yield the Treasury actually paid and the yield the market was showing minutes before the sale. The trade calls this the auction tail. It is a toll, and it is the cleanest read available on which maturity the market least wants to own.

This piece asks one narrow question. Take every government bond sale of 2026, sort them by maturity, and divide each toll by the number of dollars raised. Which maturity is hardest to place? The answer is not the longest one. It is the 20 year.

That matters because 2026 is a year in which the Treasury has funded almost everything with very short paper and left the supply of long bonds flat. If the toll were about the size of the government's borrowing, the 20 year would not be the most expensive place on the curve. It is, and that tells you the toll is being set by the balance sheets of the banks that have to hold the bond when nobody else bids.

Per dollar raised, the 20 year is the most expensive point on the curve

There are 66 completed sales of notes and bonds in the 2026 record used here, from 12 January to 8 October. Averaging their tolls is misleading, because a 70 billion dollar 2 year sale and a 13 billion dollar 20 year sale are not the same transaction. Divide by the dollars raised instead and the ranking becomes clear.

The 2 year costs 0.051 hundredths of a percentage point of toll for every billion dollars raised. The 3 year costs 0.044, the 5 year 0.027, and the two maturities that large institutions treat as near substitutes, the 7 year and the 10 year, cost almost nothing at 0.005 and 0.006. The 30 year costs 0.021. The 20 year costs 0.108.

That is 5.1 times the 30 year and 20 times the 10 year. Per dollar, the market charges 5 times as much to take a bond 10 years shorter.

The plain averages tell the same story with less force. The 20 year's average toll is 1.51 hundredths of a percentage point against 0.48 for the 30 year. That gap looks like noise until it is divided by size, and size is the whole point: the Treasury sells about 14 billion dollars of 20 year paper per sale and about 23 billion dollars of 30 year paper.

This is not a claim that the 20 year sales went badly. The average toll is under 2 hundredths of a percentage point, and only 1 of the 9 sales priced through the market rather than at a discount. The claim is about the price of the marginal buyer's balance sheet, and about what happens when a bond is sold in a size too small for the banks to distribute efficiently and too large for the end investors to absorb in a single sale.

The shortage is on the buying side, and the bank in the middle pays for it

Every auction result divides the accepted bids into 3 groups. Indirect bidders bid through a bank; that is where foreign official money and most long term institutional money sits. Direct bidders are domestic institutions bidding for themselves. Primary dealers are the banks obliged to bid at every sale, and what they end up holding is the portion nobody else wanted.

Across the 2026 sales, indirect bidders took 64.1 per cent of the 20 year's privately sold portion and 68.6 per cent of the 30 year's. Primary dealers took 11.6 per cent of the 20 year and 9.6 per cent of the 30 year. Two points of difference is not a scandal by itself. What makes it matter is the combination: the 20 year is the smallest sale on the calendar, it carries the most expensive toll per dollar, and the banks are taking a larger share of it than of a 30 year the Treasury sells at roughly twice the size.

The 15 September sale shows it most clearly. It was 13 billion dollars, it paid a toll of 2 hundredths of a percentage point, indirect bidders took 52.1 per cent and the banks took 16.7 per cent — the lowest indirect share and the highest bank share of any 20 year sale this year. Compare the 30 year on 8 October: a toll of 1.8 hundredths of a percentage point, 72.1 per cent indirect, 6.8 per cent to the banks, on a 22 billion dollar sale. Twice the size, less than half the bank burden, and a slightly smaller toll.

The mechanism is a mismatch inside the banking system. A 20 year bond carries the interest rate risk of a long bond and the trading liquidity of neither end of the curve. It is not a benchmark maturity, so it is expensive for a bank to finance while it holds it. It is not a 30 year either, so it does not match the long dated liabilities of pension funds and insurers, the investors who would otherwise want it. The bank underwrites the sale and then has to find a home for a bond that no client's mandate specifically asks for. The toll is the fee for that search.

The inversion has held for 45 sessions, which is why it is a price and not an accident

The 20 year has yielded more than the 30 year on 47 of the 51 trading sessions since 1 September, and on only 44 of the 167 sessions before that. The current run is unbroken: 45 consecutive sessions from 10 September. On the 13 November close the gap is 5 hundredths of a percentage point — the 217th widest of 218 sessions this year, against a median of zero and a range of minus 7 to plus 6. Only one session in the file has been wider.

An inversion that survives 45 sessions is not a market accident. Nor is it a dramatic one: the widest print of the year is 6 hundredths of a percentage point, and a 20 year that costs 5 more than a 30 year is well inside what a simple expectations model would tolerate. What makes it useful is its persistence. A gap that sits at the 99.5th percentile of its own year for 2 months, while the average extra compensation demanded for holding long bonds is falling, is telling you something about structure rather than about one failed sale.

The levels say the same thing from the other side. The 20 year closed at 5.52 per cent on 13 November, the 203rd highest of 218 sessions, and the 30 year closed at 5.47 per cent, the 204th. Both sit near the top of the year's range. The 20 year sits nearer. The peak of the curve has moved inside 30 years and stayed there.

Short paper carried the funding, so the long end was never asked to

The supply story cannot explain a toll specific to one maturity, and the issuance record says why. Over the 90 days to 16 November the Treasury settled 5,742 billion dollars of short bills and 815 billion dollars of notes and bonds, so bills were 87.6 per cent of 6,557 billion dollars of gross issuance. In the previous 90 days, notes and bonds were 1,234 billion dollars on an 84.7 per cent bill share. Settlement of notes and bonds is down 419 billion dollars, or 34 per cent, while the bill share rose 2.9 points.

This is the political channel, and it runs the opposite way to the obvious one. A government that cannot pass new spending does not need to sell new long bonds, so the quarterly refunding keeps the 3 year, 10 year and 30 year sizes flat for a fourth consecutive quarter and the funding need is met in short paper. Long bond supply does not rise, so the extra compensation demanded for holding long bonds does not rise either. The New York Fed's model estimate of that compensation — the term premium, meaning the extra yield a buyer demands for locking money up rather than rolling short paper — fell 21.1 hundredths of a percentage point to 0.8738 per cent between 6 October and 6 November, the latest print available.

So the average compensation for holding long bonds is falling while the 20 year's specific premium over its neighbours is the widest of the year. Those two facts only reconcile one way. What is being priced at the 20 year is not the scale of government borrowing. It is the cost of placing a bond the government insists on selling, in a size the banking system cannot distribute.

One measurement that cuts against this desk's own argument

The obvious test is whether the gap tracks the auction result. Plot the 20 year's toll against its gap over the 30 year on each of the 9 sales, and the statistical correlation is minus 0.22, which is to say nothing. The 19 August sale paid a toll of 3.4 hundredths of a percentage point while the gap was minus 2; the 15 September sale paid 2.0 while the gap was plus 4. If a bad sale were driving the inversion, the two would move together. On 9 observations they do not — far too few to settle anything, but enough to refuse the easy version of the argument.

What that leaves is a gap set by the standing conditions of the maturity rather than by any single sale: the size, the way the bond is financed while a bank holds it, the client base, and the bank's own cost of capital. Last week the gap widened 2 hundredths of a percentage point, from 3 on 6 November to 5 on 13 November, but a second measure moved the other way. That measure — the 20 year's yield against the average of the 10 year and the 30 year, which prices the same bond against both of its neighbours instead of one — fell from 46 to 42. Neither moved the way the bad-sale story requires. The 20 year is not being progressively punished. It is being punished at a roughly constant rate, which is a different and much more durable claim.

One October sale exists in the record and has no price in it

A data caveat belongs on the record before the conclusion. The Treasury's public auction record carries a 20 year sale dated 21 October with every result field empty: amount offered, amount accepted, bid to cover, high yield, bidder splits. The row exists and the price does not. So the 20 year statistics here rest on 9 sales through 15 September, not 10, and the October sale is absent from every average quoted above. If it was the worst sale of the year, the per dollar figure for the 20 year is understated; if it was the best, the ratio is flattered. The 217th of 218 ranking and the 5 hundredths of a percentage point gap are unaffected, because those come from the daily yield curve rather than the auction record.

The 20 year sales of 2026 fell on 21 January, 18 February, 17 March, 22 April, 20 May, 16 June, 22 July, 19 August and 15 September — a third week of the month rhythm on a Tuesday or Wednesday. The next one in that cadence lands in the third week of November. That is the nearest test of everything above, and it is 2 days away.

The mechanism, the instrument, and the level at which this is wrong

The mechanism, in plain terms. Divided government means no new long bond supply: settlement of notes and bonds fell 419 billion dollars in 3 months while bills carried 87.6 per cent of gross funding. No new long bond supply means no new pressure on long term compensation, which is why the term premium fell 21.1 hundredths of a percentage point in the month to 6 November. But the Treasury still sells about 14 billion dollars of 20 year paper 9 or 10 times a year, into a client base that does not specifically want it, through banks that must hold it. The price of that sale does not rise with the deficit. It rises with the banks' cost of carrying the bond and falls as the sale gets bigger. That is why the toll per dollar is highest exactly where the sale is smallest.

The instrument is the gap between the 20 year and the 30 year, at 5 hundredths of a percentage point on the 13 November close, or the toll per billion dollars raised, at 0.108 against 0.021 at the 30 year.

This is wrong if that gap closes below zero for 3 consecutive sessions. A 20 year that again yields less than a 30 year would mean the market has stopped charging this maturity a separate price, and the whole construction collapses into a rounding error. There is a weaker second test: a 20 year sale that pays a toll of 0.5 hundredths of a percentage point or less while the gap holds at 4 or more would demote the claim from structural to a matter of trading liquidity.

DISCLOSURE — NOT INVESTMENT ADVICE
NoticeMacroTape publishes macroeconomic analysis and market commentary for information and education only. Nothing published here is financial, investment, legal or tax advice, or a personal recommendation, and no author is a registered investment adviser. Calls, Levels and invalidation points are analytical views about how a market mechanism works, not instructions to trade. Readers act on their own responsibility and carry their own losses.