What actually happened: the operation was tripled, and the margin went to zero

The 19 August announcement raised the per-operation maximum in the 10-to-20-year and 20-to-30-year sectors from $2 billion to "at least $4 billion" for the remainder of the refunding quarter, to 4 November 2026, with a stated purpose in Treasury's own words: "greater liquidity support in longer-dated nominal sectors where there is consistent strong sponsorship from market participants". Treasury's tentative buyback schedule then put a number on the first enlarged operation, and the number is $6 billion: 10-to-20-year nominal coupons maturing between 11 September 2036 and 10 September 2046, operated 1:40–2:00pm ET on 10 September, settling the next day. The comparable operation on 11 August, same sector, carried a $2 billion cap. Every later long-end operation on the schedule — 24 September, 1 October, 8 October, 15 October, 27 October and 4 November — is still listed only as "= or > $4 billion".

So the quantity question was answered generously. It moved nothing in the direction it was meant to.

Official closes from the Treasury's daily par yield curve. The 30 June column is where this summer's repricing began; 18 August is the last close before the buyback news; 25 August is the low the funding report produced; 9 September is the latest close published.

The shape held for a fourth reading, and it is cleaner now than at any earlier point in this storyline. Read the "vs 18 August" column downwards: +9 basis points at the three-month, +18 at the one-year, +24 at the two- and five-year, +18 at the seven, +12 at the ten, and zero at the twenty and thirty. That is a hump in the two-to-five-year belly decaying to nothing at the back — the signature of a repriced policy path, and close to the inverse of what a term premium shock leaves behind. A term premium shock steepens the long end; this one flattened it to a standstill while the belly did all the work.

The bolded cells are the part that gets reported backwards. On 9 September the two-year at 4.43%, the five-year at 4.61%, the seven-year at 4.71% and the 10-year at 4.83% all set 2026 highs. The 20-year and 30-year, at 5.28%, did not: they remain 2 and 3 basis points below the 5.30% and 5.31% they printed on 17 August. Three weeks into what is widely described as a long-end rout, the only two maturities on the curve that are not at a high for the year are the only two the buyback covers — and they are also the only two that have gone precisely nowhere since the programme was announced. Both things are true at once, and holding them together is the entire read.

What the same column costs the buyback is the headline the last version could not write. The scoreboard is spent. On 25 August the 20-year and 30-year sat twelve and eleven basis points below their pre-announcement closes; a week later, one each; on 9 September, zero. Every basis point of relief the announcement and the funding report bought has been handed back — and handed back before the first enlarged purchase even settles. That is not a verdict on whether buybacks work. It is a verdict on what was actually being paid for in late August, which was optionality on a much larger intervention rather than the operation printed on the schedule.

The auction on the other side of that close says the same thing from the demand side, and says it in a way coverage routinely inverts. Treasury reopened the 10-year note on 9 September and sold $39 billion of it at a 4.834% high yield, on a bid-to-cover of 2.71 — reported as the strongest cover at a 10-year sale in roughly a decade — with 79.2% of the paper taken by indirect bidders and only 4.3% left with primary dealers. A low dealer take-up alongside a high indirect share is the profile of a sale that found real end demand rather than being warehoused. Yields rose anyway, because a sovereign auction announces its size in advance and discovers its price: a strong cover tells you the paper clears, and nothing whatever about the level at which it clears. It is the same lesson the Japanese sales below produced a week earlier, now denominated in dollars.

Read the 27→28 August single session and the attribution problem that has dogged every earlier version of this post disappears for once. The two-year rose fourteen basis points in one day. The 30-year rose three. The bill rose six. There is one scheduled event on that date, it is a central-bank speech, and it is the subject of the Jackson Hole piece: futures put a quarter-point September hike at 60.4% by Monday 31 August, up from around 56% on the Friday and roughly 35% before the keynote. A speech worth fourteen basis points at the two-year and three at the 30-year is the cleanest single-cause reading of a curve move in this entire storyline — and it is the mirror image of a buyback, which is worth almost nothing at the two-year and something at the 30.

Japan closed above 3%, and then sold thirty-year paper into it

The closing line of the version before last read: "Two long bonds on different continents turning on the same date is not something one country's debt management office explains, and one country's checking account will not fix it." The last version recorded the 10-year touching 3% during the Tokyo session of 1 September and closing below it, at 2.987%. The session after settled it. On 2 September 2026 the official close on the Ministry of Finance's daily series was 3.006%, and on MOF's own historical record the previous close at or above 3.000% was 3.060% on 6 September 1996 — three days short of exactly thirty years earlier.

What moved alongside it is more interesting than the round number.

The 10-year rose into its first 3% close in thirty years and the entire super-long sector fell. That is the same shape as the American curve in the same two sessions, drawn in a different currency: the move is at the policy end, and the duration end is quietly bid.

It also leaves the Japanese curve with a kink in its own tail. On 2 September the 25-year at 4.141% and the 40-year at 4.134% both yielded more than the 30-year at 4.122%. A yield curve does not normally dip in the middle of its longest maturities, and the reason this one does is issuance rather than sentiment: MOF cut the size of its 30-year auctions between March and May 2026, from roughly ¥525bn of competitive paper accepted to roughly ¥455bn. Thirteen per cent less of a maturity the market must absorb shows up as a lower yield than the maturities either side of it — and it is the sector MOF was about to sell into.

Which brings the test this post named in advance. MOF publishes the numbers itself: here is the 3 September result against the 6 August one, with the rest of the recent run for context.

Cover essentially held. The book was bid 3.79 times over on a size ¥0.4bn larger than August's, tendered bids fell by only ¥33bn on a ¥1.7trn total, and 3.79 is higher than five of the six sales before it. Fourteen basis points of extra yield did not thin the queue.

The tail went the other way, and it is the first genuine deterioration this storyline has produced. Two point one basis points between the marginal and average yields is the widest of the seven, against 0.7bp when the same maturity was yielding 3.4% last December. The allotment at the lowest accepted price says the same thing from the other side: 30.3% of bids at the cutoff were filled, against 48.9% in August, which means bids piled up at the cheapest price MOF would take. Aggregate appetite for Japanese thirty-year paper at 4.08% is intact; the price at which it appears is getting harder to pin down.

And then the number that keeps this from being a bad result, which is the comparison nobody quotes. The 4.079% average cleared 4.3 basis points below the 30-year's 4.122% official close the session before. An auction that prices through the screen is not one the seller had to drag the market into. Read on the criterion set in advance — cover and tail together — the answer is one out of two, and the one that failed failed narrowly.

There is also a specific reason a 3% Japanese 10-year is less of a rupture than the round number suggests, and it is written in the Japanese government's own budget documents. MOF's Highlights of the FY2026 Draft Budget raised the interest rate applied in the budget calculation from 2.0% to 3.0% — adding about ¥1.0trn to estimated interest payments on scheduled issuance — and states plainly that the 3.0% figure was set "by referring to the recent one-month average of long-term interest rates (1.9%) and taking into account past cases of sharp interest rate hikes". National debt service in that budget is ¥31,275.8bn, up ¥3,057.9bn on the FY2025 initial budget, of which interest payments are ¥13,037.1bn against ¥10,523.0bn. The market has now caught up with a number the borrower wrote down more than eight months ago, and funded.

The forward-looking half is the larger one. Jiji reported on 22 August that the Finance Ministry is considering a 3.8% assumed long-term rate for its FY2027 budget request — the assumed rate is built by adding roughly 1.1 percentage points to prevailing yields to allow for spikes, and it was lifted from 2.6% to 3.0% during the FY2026 compilation at the end of 2025. Bloomberg reported on 26 August that MOF will seek a record ¥36.6trn for debt servicing in the FY2027 initial budget — ¥16.6trn of it interest and ¥20trn redemptions — a 17% jump on the ¥31.28trn allocated this year. Prime Minister Sanae Takaichi's growth strategy and planned sales tax cut have not yet been costed, and that, rather than the 3% handle, is the part of the Japanese long end that is genuinely unpriced.

The refinancing gap, and why it is the slow part of this story

The reason a Japanese rate rise transmits over years rather than weeks sits in one number from the same budget document. MOF's chart of accumulated general bonds outstanding puts the stock at about ¥1,145trn for FY2026 with a weighted average interest rate on it of roughly 0.8%. The August 2026 JGB Monthly Newsletter puts the average remaining maturity of that stock at nine years and five months at the end of FY2025.

Put those together and the mechanism is arithmetic, not opinion. Roughly a tenth of a ¥1,145trn stock rolls off each year and is replaced at whatever the market charges. The bond MOF sold on 1 September carries a 2.7% coupon. The paper it replaces was struck close to zero. Nothing about that repricing is optional, and nothing about it is fast — which is why the interest bill in the budget rises by two to three trillion yen a year rather than jumping to the market rate at once, and why an assumed-rate change of one percentage point is a genuinely large fiscal event in Japan while being a rounding error in the same year's bond market.

The holder base is the other half of why this is slow. On Bank of Japan flow-of-funds figures reproduced in the same newsletter, the central bank itself held ¥485.4trn of JGBs at March 2026 — 47.9% of the ¥1,013.8trn outstanding — against insurers on ¥155.3trn, banks on ¥149.7trn, foreign investors on ¥82.0trn and households on ¥20.0trn. A market in which the price-insensitive holder owns nearly half the float does not reprice the way a free float does, in either direction. It also means the marginal buyer of new supply is a domestic institution deciding between a JGB and a currency-hedged foreign bond — which is where this connects back to the US long end.

The channel from Tokyo to the US long end: hedged yields

A Japanese life insurer or bank matching yen liabilities does not compare a JGB yield with a Treasury yield. It compares a JGB yield with a Treasury yield net of the cost of hedging the currency, and that cost is set by the gap between short-term dollar and yen rates, because a rolling hedge is a series of short-dated forwards.

Take the closes of 2 September 2026 and do the arithmetic in the open. The US 10-year was 4.79% and the three-month bill 3.92%, both on Treasury's par yield curve. Japan's three-month bill cleared at an average yield near 1.00% at the 30 July auction, per the MOF newsletter. So a rolling three-month hedge costs roughly 2.92 percentage points a year, leaving about 1.87% from a hedged 10-year Treasury. Against a JGB 10-year at 3.006%, the domestic bond pays a domestic investor something like 1.14 percentage points more — and that gap has widened by four basis points in two sessions, because Japan's 10-year rose faster than America's.

That figure is an approximation — real hedge costs include a cross-currency basis and institutions run partial and laddered hedges rather than a single rolling three-month contract — but the sign and the rough magnitude are not sensitive to any of that. And the sign is the point. For most of the last two decades the calculation ran the other way, which is why Japanese institutions became one of the largest foreign owners of American, French and Australian government debt. A domestic bond that out-yields the hedged foreign alternative by a percentage point removes the reason to be abroad, and the flow that reverses is the flow that had been suppressing term premia in other people's long ends.

Two cautions keep this honest. First, nothing here says the flow has reversed; it says the incentive has. Japanese institutional allocation moves over quarters and is constrained by regulation and accounting, a slowness examined in the piece on Japan's pension shift toward domestic assets. Second, the hedge cost falls if the Fed cuts and rises if it hikes — and September's pricing points to a hike, which widens the gap in the dollar's favour again. This is a channel with a variable valve on it, not a switch.

How the 25 August low was bought: the funding source, not the size

The 19 August announcement was specific about quantity and silent about payment. Treasury said it would increase, "by at least double", the size of liquidity support buyback operations in longer-dated nominal coupon securities — from $2 billion per operation to at least $4 billion, in the 10-to-20-year and 20-to-30-year sectors, effective 9 September and running through 4 November 2026. It did not say where the money would come from, and the market's default assumption was new bill issuance. Treasury Secretary Scott Bessent had described the operation to CNBC as a "Treasury Twist", which pointed the same way.

That assumption is what capped the effect. Buying long bonds with freshly issued bills swaps duration for supply; it does not bring cash to the transaction. As this post put it three sessions later, the operation worked exactly as designed and the design was small.

On Monday 24 August, CNBC's Steve Liesman reported, citing two senior Treasury officials, that a different funding source is on the table: the Treasury General Account, the government's checking account at the Federal Reserve, already funded out of collected tax receipts. The officials would not say how much, if any, would be used, or when an announcement might come. They were clear that it is considered available.

The Daily Treasury Statement puts the closing balance on 24 August 2026 at $966,849 million — $966.8bn, built up from an opening $933.2bn on deposits of $50.4bn against withdrawals of $16.8bn. Under the previous administration the stated aim was to hold roughly $550bn to $600bn. The current Treasury says it sets the account "consistent with Treasury's long-standing cash balance policy", which is a way of saying the number is discretionary.

The prints on either side: why the 25 August low was never clean

The ten basis points of relief this post recorded on 25 August were never attributable to the buyback alone, and the record should keep saying so. Two other things arrived in that session, and each lowers a curve on its own. The Conference Board's consumer confidence index fell to 89.4 in August, below the 90.2 consensus, with the forward-looking expectations component the weaker half — the print covered in the consumer confidence piece. And crude fell hard, with Brent settling 3.9% lower at $88.58 as Washington moved toward economic pressure on Iran rather than military action, the mechanism traced in the Iran and oil analysis. Lower breakevens pull nominal yields down across a whole curve without anyone revising their view of real rates.

That channel has been running in reverse since. Firmer crude into early September is one of the reasons the front end kept climbing, and where the barrel goes next is the subject of the OPEC+ preview and of the Hormuz shipping piece. Set the whole run side by side and the contrast is the useful part. A front end that rose on a strong August PMI, fell on a weak confidence print, then rose seventeen basis points on a central-bank speech and five more into a global bond selloff is not a front end reading the data. It is a front end that had no anchor from its own central bank to fall back on — and then got one.

Four episodes, and the two the carry read got right

This post has been running a live test on one question: when a US yield moves, does the dollar follow the carry? Twice it did not. The two legs since explain why. Dollar moves are derived from the European Central Bank’s daily reference rates; a positive number means the dollar gained.

Six for six against the carry read, six for six against it again in the opposite direction, and then five of six with it, twice running. The variable that flipped the result was not the direction of yields. It was the location.

In the first two episodes the move lived behind the ten-year point, where a yield is mostly term premium — the compensation demanded for holding duration, which is a statement about the issuer's balance sheet rather than about the return on cash. Carry has no claim on that, which is why a carry model got both weeks wrong. In the third and fourth it lived at the two-year and the bill, where a yield is mostly an expected policy path. Carry is precisely a claim on that. So it worked, and it worked to the same score both times.

That retires the diagnostic this post started with. The question is not whether yields went up or down. It is which part of the curve moved, because the two halves are claims on different things — and the answer decides whether a currency should have followed at all.

The fourth episode supplies the cleanest exception in the series, and it is the yen. The dollar rose against five majors between 31 August and 2 September and lost 0.08% to the one currency whose own front end moved faster than America's. Japan's two-year rose from 1.743% to 1.854% across those two sessions, eleven basis points against five at the US two-year, so the short-rate gap narrowed from roughly 260 basis points to roughly 254. A country's long end breaking a thirty-year level did not lift its currency. A six-basis-point move in the short-rate gap did. The full working sits in the yen and rate-gap piece and the live factor read is on the yen page.

Size still deserves its caveat, and it is smaller in the fourth episode than the third. Five basis points at the two-year against currency moves of 0.08% to 0.67% is not a tight relationship, and the euro's +0.16% is close enough to noise that it should be called noise. The direction is the finding; the magnitude is modest and the span is two sessions. What earns the conclusion is that the sign has now been right twice in the location the theory predicts success and wrong twice in the location it predicts failure.

There is also a governance question sitting underneath all of this, and it has not been answered. CNBC has reported that Fed Chair Kevin Warsh has expressed a preference for the open market determining rates, and the July minutes — covered in the FOMC minutes post — showed several policymakers ready to raise. Paul Stanley, managing director and founding advisor at Arca, put the market's reading of that bluntly in CNBC's coverage on 21 August: "It seems as though Warsh wants the market to do the tightening for the Fed, and that's really what is happening with the recent surge in bond yields."

If the bond market is doing the tightening, then a fiscal authority buying back the long end is, in a limited and indirect way, doing the loosening. Those are two arms of the state pulling on the same yield from opposite ends. Nothing about the last two sessions settles what that means; it is simply the thing to keep an eye on, and Friday's keynote is the first scheduled opportunity to hear it addressed.

What a buyback does, and what changes when the cash is already sitting there

A buyback is the borrower repurchasing its own outstanding debt. No money is created, and the total stock of debt does not shrink. Peter Boockvar of One Point BFG Wealth Partners called it "just a rearrangement of the maturity schedule of Treasuries" in CNBC's coverage, and on the original design that was exactly right.

The quantity the market was choking on was never debt in general but duration — long-dated paper that has to be held by a slow-growing pool of pension funds and insurers matching multi-decade liabilities. Swapping a 30-year bond for a three-month bill leaves total debt unchanged while cutting the duration the market must absorb. The term premium is the price of that duration, so it falls. What the funding question changes is the second leg of that swap.

Neither route is quantitative easing, and the comparison gets made loosely enough to be worth one sentence of precision: QE expands a central bank's balance sheet with newly created reserves, while drawing down the General Account moves existing reserves from the Treasury's account at the Fed into commercial bank reserves. Same balance sheet, different composition. That is a liquidity operation, not the central bank financing the government.

The scale objection has not gone away either, and the people who made it were making it about the operation rather than the account. A $6 billion operation sits against a single $39 billion 10-year reopening the day before it — and against a 30-year reopening that auctions at 1:00pm ET on the very afternoon the buyback runs at 1:40. The headline size also hides a sector mismatch: the 10 September operation buys nothing maturing after 2046, so the maturity carrying the most duration risk on the curve is the one being sold that day rather than bought. Mohamed El-Erian, per CNBC, described the purchases as "small in both absolute terms and relative to net issuance", and Krishna Guha of Evercore ISI wrote that the operation "changes almost nothing in terms of the fundamentals in particular the unchanged need to finance the tidal wave of hyperscaler debt in addition to very large government deficits." Both remain true of the operation. What late August repriced was the optionality behind the operation rather than the operation — and Treasury officials named that option's cost themselves: running the account down leaves less cash for a debt-ceiling impasse, and rebuilding it later means selling more bonds to do it. The quarterly borrowing estimate has since chosen the second horn of that trade-off, and the 20- and 30-year have priced the choice.

Nor did this summer's pressure come from Washington alone, which is the point the last week has driven home. The Bank of Canada's long-term benchmark printed its 2026 high at 4.14% on 17 August, having bottomed on 27 February — the identical day the US 30-year bottomed at 4.64%. Japan's 10-year has since crossed 3%. Long bonds on three continents turning on the same dates is not something one country's debt management office explains, and one country's checking account will not fix it.

The equity leg, and Barclays' checklist three weeks on

Long-duration equity traded the discount rate in both directions, which is the cleanest sign check in this piece. In the week to 21 August, with the term premium at 19-year highs, the Nasdaq fell 2.1% against the Dow's 0.8%. On 25 August, with the long end rallying, the order reversed: the Nasdaq Composite gained 0.66% to 26,151.30 against the Dow's 0.3% rise to 53,577.40, and the S&P 500 rose 0.32% to 7,677.28.

Barclays' checklist, quoted here three weeks ago, now stands differently again. Anshul Pradhan's four conditions for a constructive long end were a downside fiscal surprise, slower AI-related issuance, a shift in Treasury's issuance strategy, and a sustained run of soft activity data. The third has now fully un-ticked: zero basis points at the 20-year and 30-year against 18 August, on an operation that was tripled rather than merely doubled, is not a shift in issuance strategy — it is the residue of one that has finished evaporating. AI issuance is still accelerating and the deficit is what it was. And the fourth has stopped being contested and started pointing the other way: August payrolls printed 162,000 against a consensus near 53,000, with the labour force up 683,000, and the jobs analysis traces how the two-year took back every basis point the dovish read had given it. A belly setting 2026 highs after a payroll print that tripled the consensus is not a market expecting soft activity data. Currency-by-currency detail sits on the dollar, euro and yen pages, and the method behind the scoring is on the about page.

What is next: four dates, and the borrower on both sides of one of them

The 9 September buyback question has been struck off this list, and the answer removed size from the argument rather than settling it. What is left is a fortnight in which the American curve gets three more chances to decide whether the front-end move was right — and one afternoon on which the US Treasury is a seller and a buyer of its own duration inside forty minutes.

Three of those deserve a sentence each, because they are the ones whose mechanics are most often read backwards.

The 10 September pairing is the cleanest natural experiment this programme is likely to produce. At 1:00pm ET Treasury reopens the 30-year bond, adding duration the market has to absorb; forty minutes later it starts buying 10-to-20-year notes, removing duration the market already holds. Same borrower, same afternoon, adjacent rather than identical sectors. If the announcement effect had content beyond sentiment, this is where it shows up — and it shows up as a relative move between the twenty-year and the thirty, not as a level move in either. Reading it as "Treasury supported the market today" would miss that one leg cancels the other by construction.

The 11 September CPI print is now the load-bearing date, and its asymmetry mirrors the payroll asymmetry this post described a week ago. The belly has been rising on an inflation and fiscal story rather than a growth one, which is why a hot core reading can lift the two-year without touching the thirty at all, and why a soft one is the only scheduled release with a realistic claim on reversing +24 basis points between two and five years. The CPI preview works through why the Fed is priced to move on a gauge that Friday does not publish.

And the 15–16 September and 17–18 September pair is the most underappreciated sequencing on the calendar. The hedge cost that decides whether Japanese institutions stay in foreign bonds is the difference between two short rates, and both are set within thirty-six hours of each other. A Fed hike alone widens it and keeps the flow abroad. A BoJ hike alone narrows it and strengthens the case for coming home. Both hiking leaves it roughly where it was, which is the outcome the market currently leans toward — and the reason this channel is worth watching rather than trading.

The takeaway

Five legs of this story have now been recorded on two continents, and the same finding keeps returning with a different number attached.

The location test has not failed once since the move relocated. Read the American curve against 18 August and it rises +9 basis points at the three-month, peaks at +24 in the two-to-five-year belly, and decays to zero at the twenty and thirty. Read the four sessions since 4 September and it does the same thing in miniature: +6 at the two-year, +7 at the five, +4 at the thirty. Read the Japanese curve at the start of September and the 10-year rises while the 25-, 30- and 40-year all fall. Three curves, one shape. A term premium story does not produce it; a policy story produces it every time. For the dollar that distinction is the whole difference between a yield move that carry can claim and one it cannot — which is why the dollar page reads the front end and not the headline.

And the buyback's scoreboard is now spent, which turns out to be more useful than a win would have been. Treasury tripled the operation, published the maturity range and the clock, and the two sectors it targets closed exactly where they closed on 18 August. The relief was never priced off the quantity — a $6 billion purchase against a $39 billion reopening the day before was never going to move a curve — it was priced off the possibility of a near-$1 trillion cash account standing behind it, and Treasury's own borrowing estimate has quietly withdrawn that possibility by assuming the account is $70 billion larger at the end of the month. What is left is a debt manager smoothing liquidity in an off-the-run sector, which is exactly what the programme was designed to do, and was never the thing that decides where the long end trades.

Japan settling above 3% is the confirmation that this was never an American story, and the 30-year auction is the correction to how such things get reported. The seller paid fourteen extra basis points and got a book covered 3.79 times, priced 4.3 basis points through the previous session's close, with the widest tail in seven sales. That is not a failed auction. It is not a clean one either, and saying both is the whole job. The government issuing the paper had already written 3.0% into its FY2026 budget and is reportedly working to 3.8% for the next one, so the rupture people expect at a round number is arriving where it always arrives — in an interest line item, at the pace a ¥1,145trn stock with a nine-and-a-half-year average maturity can reprice.

And the currency read held a second time, with its most instructive exception yet. The dollar rose against five majors while its own front end rose faster than its long end. The one it lost to was the yen — not because Japan's thirty-year story improved, but because Japan's two-year moved eleven basis points against America's five.

Before deciding what a yield move means for a currency, find out where on the curve it happened. Then find out whose curve.

Educational macro context only — not investment advice.