February's Auction Showed 50% Foreign Demand. The Fed Wrote 14 Points of It.

The 10yr and 30yr auctioning on 10 and 12 November are the same sizes the Treasury sold in February, 42bn and 25bn, but without the 28.3% SOMA add-on that lifted February's total accepted to 53.90bn and 32.08bn. Measured against total accepted, indirect bidders took 50.1% of the February 10yr; measured against the 42bn the private sector had to buy, 64.3%. The gap is an identity, not a correlation: it holds to 0.02pp on all 75 nominal coupon auctions this year. The 10s30s sits at 40.0 basis points, the 18th percentile of 214 sessions.

Every Treasury auction produces a table of who bought it: primary dealers, direct bidders, and indirect bidders. Indirect is the bucket that matters. Foreign official accounts and domestic funds that bid through a dealer land there, so a rising indirect share is read as evidence that real money wants the paper. The reading only works if the denominator is the paper the public had to buy. It often is not. When the Federal Reserve's Open Market Desk reinvests maturing holdings it bids non-competitively at the auction, and its purchase is added to total accepted. The offering does not change. The denominator does. In February that add-on was 28.3% of the offering at both the 10yr and the 30yr, which turned a 64.3% indirect take into a printed 50.1%. This week the same 42bn and 25bn are sold against a Desk that has been shrinking rather than reinvesting. The piece tests what that does to the demand signal everyone is about to read, and what it means for a 10s30s already at the 18th percentile of the year.

The same auction, two denominators

The 3yr auctions today, the 10yr tomorrow and the 30yr on Thursday are the same sizes the Treasury sold in February: 58bn, 42bn and 25bn, settling 16 November. What differs is the denominator. In February the Federal Reserve's Open Market Desk added 28.3% to the 10yr and the same to the 30yr by reinvesting maturing holdings at the auction: 11.90bn and 7.08bn. Total accepted became 53.90bn and 32.08bn against offerings of 42bn and 25bn.¹

Indirect bidders are the bucket that holds foreign official accounts, domestic funds bidding through a dealer, and pensions. In the February 10yr they took 50.1% of total accepted and 64.3% of the private take. The 30yr reads 54.4% one way and 69.8% the other. Same bidders, same session, 14.2 and 15.4 percentage points apart.²

The gap is not a correlation that happens to hold. It is arithmetic. Indirect share of total accepted equals indirect share of private take divided by one plus the add-on as a fraction of the offering, and that identity holds to within 0.02 percentage points on all 75 nominal coupon auctions in this sample. Change the add-on and the printed share changes by a known amount. Nothing about who showed up has to change at all.³

The regression says the same thing from the other side. Across 75 auctions the add-on carries a slope of -0.58 on the printed indirect share at a correlation of -0.60, and a slope of -0.06 at a correlation of -0.08 on private take. The first number is a mechanical consequence of the identity. The second is what demand actually did, and it is flat.

What the add-on is, and how large it became

A SOMA add-on is reinvestment, not stimulus. The Desk holds Treasury paper that matures, and when it rolls that principal back into an auction it bids non-competitively as one more participant. The purchase is added to total accepted rather than subtracted from the offering. It is not a buyer the market had to find. It was always going to be there, standing inside the statistic everybody quotes.

The size is not marginal. Across the three refunding blocks this year — February, May and August — the 3yr, 10yr and 30yr carried 96.72bn of add-on on 375bn of offering and 471.7bn accepted, which is 20.5% of everything those nine auctions absorbed. In the matching mid-quarter months the Desk has mostly stayed out: the largest primary 3yr/10yr/30yr add-on in March, June or September was 0.03bn.⁴

So the bidder statistics carry a step, not a slope. A 2026 30yr reopening shows a median 63.3% indirect take; a refunding-month 30yr new issue shows 53.7%. Set the two side by side and foreign demand looks like it fell apart. Compare indirect share of private take and they print 66.7% and 66.6%: the same crowd, before and after.⁵

The correction runs against this desk's own 2 November numbers

None of this is a fault in the Treasury data, which publishes every component separately and correctly. It is a statement about what this desk's own auction coverage has been measuring. On 2 November this desk published a concession ladder putting indirect bidders at 65.5% of the 10yr and 61.0% of the 30yr, with primary dealers at 7.6% and 8.4%, all computed against total accepted. Every one of those figures is arithmetically right and structurally misleading.

Measured against private take the same 2026 auctions put indirect bidders at 73.1% of the 10yr and dealers at 8.7%. The 30yr median on private take is 66.6% indirect against 10.7% dealer. The correction is worth 7 to 8 points at the 10yr and 5 to 6 at the 30yr, in the direction of more real money and less Street inventory than the headline suggested.

It matters most where the add-on is largest, and that is not evenly spread. The 2yr has taken a median 9.7% add-on this year because the Desk has used it for reinvestment more than any other tenor, so its printed indirect share is the most distorted of the seven. The 20yr, which takes a 9.4% median, is the least. A single cross-tenor comparison built on total accepted is therefore comparing seven different denominators.

The long end has been paying for this all year

The long end has spent the year pricing the same transfer. The 10s30s, which is the 30yr par yield less the 10yr and so the price of duration beyond the note sector, closed at 40.0 basis points on 11/06/2026 against a 2026 median of 53.0. That is the 18th percentile of 214 sessions.

The 20yr tells it from below. The 20s30s closed at -3.0 basis points on 11/06/2026, negative means the 20yr yields more than the 30yr, and it has been negative for 41 consecutive sessions. The 20yr remains the most expensive money the Treasury borrows.

Two facts sit together here. The Fed's Treasury holdings fell 143840000.0bn between 2026-01-07 and 2026-10-21, to 1895180.000tn: the runoff of paper that would otherwise have been rolled at auction. And the private sector absorbed the difference at the tenors where the add-on had been largest. A 10s30s at 40.0 basis points is the price of that, not the price of a deficit forecast.⁶

The mechanism, the instrument, the falsifier

The mechanism, stated plainly. The 3 November result closed the deficit channel: a sweep that cannot pass a budget through a 60-vote Senate does not add duration supply, and the 4 November refunding duly held the 3yr, 10yr and 30yr sizes flat for a fourth consecutive quarter. With no new supply to price, the political fact reaches the bond market through the funding mix instead — who holds the paper that already exists. The Desk is shrinking. Its add-on went from 28.3% of the February offering to 2.4% in October. Every point of add-on that disappears hands a point of the auction to private balance sheets, at a price set in the 10s30s and paid in dealer inventory.

The instrument is the 10s30s at 40.0 basis points, with the 30yr tail as confirmation. Across 10 nominal 30yr auctions this year the tail averaged +0.48bp with a median of +0.70 and a range of -6.2 to +7.0. A 25bn 30yr on Thursday that clears inside 2bp of its close with dealers under 14.7% of private take is a long end absorbing the runoff without complaint.

Where this is wrong

The falsifier is a 30yr on 12 November clearing more than 2.0bp above its same-day par close with primary dealers above 15% of private take. That combination would say the buyer base is genuinely thin, not merely remeasured: no 2026 30yr auction has put dealers above 14.7% of private take, and the median is 10.7%. A tail that wide with the Street holding that much inventory is a demand problem, not a denominator problem.

The second falsifier is the spread. A 10s30s close above 63.0 basis points, the 90th percentile of the year, would say the long end has already absorbed the transfer and begun charging for it. That kills the mechanism rather than merely damaging the position, because the price of duration would have moved ahead of the supply that was supposed to create it.

The third is quiet and it is the one that would embarrass this piece most. The announcements for all three auctions this week carry `somaInsideOfOfferingAmount` as `No`. That field read `No` before the May and August blocks too, and both carried add-ons. The denominator for this week is therefore not knowable from the announcement. It is knowable at 13:00 ET on each auction day, which is precisely the point.

Notes

1. Auction results from Treasury Fiscal Data, `auctions_query`, settlement dates from 1 January 2026; offering amount, total accepted and SOMA accepted as reported per auction. 10yr auctioned 2026-02-11: offering 42.00bn, total accepted 53.90bn, SOMA accepted 11.90bn. 30yr auctioned 2026-02-12: offering 25.00bn, total accepted 32.08bn, SOMA accepted 7.08bn.

2. Indirect share of total accepted is indirect accepted over total accepted. Indirect share of private take is the same numerator over total accepted less SOMA accepted. The private take equals the offering in this sample by construction, because the Desk's purchase is added to the offering rather than subtracted from it. Gaps: 10yr 14.19pp, 30yr 15.42pp.

3. The identity is indirect/total = indirect/private ÷ (1 + SOMA/offering). Checked on all 75 nominal coupon auctions; the largest residual is 0.000pp. Sample excludes inflation-linked auctions, whose real-yield prints are not comparable to a nominal par curve. Ordinary least squares of indirect share on add-on-as-share-of-offering: slope -0.578, r -0.602 on total accepted; slope -0.064, r -0.078 on private take.

4. Refunding blocks are the largest 3yr, 10yr and 30yr of each month. February, May and August: 96.72bn of add-on on 375bn offered and 471.7bn accepted, 20.5%. March, June, September printed 0.03bn, 0.00bn and 0.00bn at the same three tenors.

5. 30yr reopenings (n=7): median indirect of total 63.3%, of private 66.7%. 30yr new issues at or above 13bn (n=3): 53.7% and 66.6%.

6. SOMA holdings of Treasury securities, FRED `WSHOMCB`: 2039020.000tn on 2026-01-07, 1895180.000tn on 2026-10-21, a fall of 143840000.0bn. Par yields from the Treasury daily par yield curve; spreads are differences of par closes on the same date, in basis points.

7. ACM 10yr term premium, FRED `THREEFYTP10`: 0.9089% on 2026-10-30, the 83th percentile of 209 prints in 2026 inside a range of 0.4627% to 1.0965%. Quoted for context. The argument here does not depend on it.