# Why Yields Jumped: When the Last Buyer Is Already Full

> Notes from Bloomberg Odd Lots' conversation with Stanford finance professor Darrell Duffie: the surge in long-dated Treasury yields may have less to do with inflation than with supply landing on investors who already own all they want. On marginal buyers, Treasury buybacks, the liability side of the Fed's balance sheet, and how the idea transfers to your own judgment. Educational, not investment advice.

Published: 2026-09-04
Locale: en
Tags: bonds, yields, macro, federal-reserve, podcast-notes


![A long pier at dawn, cargo crates stacked away into the mist while a handful of dockhands watch another shipment arrive](/covers/oddlots-2026-09-03-what-s-behind-the-big-surge-in-us-government-bond--cover.png)

> Today you give away five cities, tomorrow ten, and buy yourself one night of quiet sleep. You rise, look to the borders — and the armies of Qin have come again.
>
> — Su Xun, "On the Six States" (Northern Song, c. 1060; my own translation)

## What this episode is about

Odd Lots published this one on 3 September 2026. Tracy Alloway and Joe Weisenthal were at Jackson Hole, recording outdoors, and the sun came out halfway through. The official theme of the symposium was payments innovation. The thing everyone was actually talking about was why long-dated Treasury yields kept climbing, with the 30-year above 5%.

Their guest was Darrell Duffie, professor of finance at Stanford. He works on the macro picture and on the plumbing of the Treasury market at the same time — dealer balance sheets, buyback operations, where liquidity clogs — so he can connect what the conference is discussing to how the market actually clears. What stayed with me wasn't his view on rates. It was where he chose to start: not with inflation, not with policy, but with who is supposed to buy.

## The main points

**The thought experiment carries the whole episode.** Duffie put it to Tracy: suppose I convince you there is no inflation risk and no default risk. You run a macro hedge fund holding $20 billion of 10-years. Treasury calls and says there's room on your balance sheet, take another $10 billion. Why wouldn't you? His own answer: because at 5.3% you already hold the amount you chose to hold. To get you to take more, you need a higher yield as compensation. The price isn't set by what everyone thinks of Treasuries. It's set by the one who already owns enough and is being asked for one more bite.

![A horizontal dashed line marked 5.3%; beneath it a wide, low gray block for the $20 billion already held, and to the right a taller block for the $10 billion he is asked to take on, the extra height being the compensation he demands](/figures/marginal-buyer-one-more-bite-en.svg)

**The buyer base has changed hands.** Foreign central banks have had what they need for a long time and aren't adding. Foreign investors as a group aren't keeping up with the growth of the market. What's left to absorb the increase is domestic discretionary money — mutual funds, hedge funds, banks, insurers, pension funds. They have one thing in common: they're yield-sensitive. Ask them to hold more and they'll want to be paid for it.

![A tall bar on the left for the Treasuries added each year, with three arrows of different thickness pointing to three buyer groups on the right; the first two arrows are thin and the one to domestic discretionary investors is the thickest](/figures/who-absorbs-the-new-supply-en.svg)

**The crowding runs the opposite way from the popular story.** A common line is that hyperscalers are issuing so much long-dated debt that they're pushing Treasury out. Duffie would describe it the other way. The Treasury market is $31–32 trillion and growing by roughly $2 trillion a year; hyperscalers might reach a trillion of debt over the next couple of years, which is small by comparison. The entity stuffing bonds into the same pension funds and insurance companies is government — finance ministries and legislatures, the US above all.

![A very large square for the $31 trillion Treasury market, a thin strip along its base for the $2 trillion added each year, and a much smaller square beside it for the $1 trillion of debt data-center operators might issue](/figures/scale-of-the-squeeze-en.svg)

**What changed was the volume, not the story.** Joe pressed the right question: you could have told this story ten years ago — demographics, no political appetite to cut spending, debt-to-GDP marching upward. People told it about Japan for decades while rates fell. So what changed? Duffie's answer is quantity. The IMF once treated 60% debt-to-GDP as a red line you crossed at your own risk; major governments have walked past it one by one, France now at 100% too. Ten years ago the Treasury market was around $18 trillion. Today it's $31 trillion. He cited Ken Rogoff's remark at lunch that long-term yields show a lot of mean reversion, that things come and go — but what keeps coming is more bond debt.

![A horizontal dashed line running across the top for the story that has not changed in ten years, and below it two bars growing from $18 trillion to $31 trillion](/figures/same-story-bigger-stock-en.svg)

**The buyback program was built for one job and is now being used for another.** Its original purpose was housekeeping: sweep up the leftover bits of old off-the-run notes and bonds that no longer trade, clog dealer balance sheets, and price below what a smooth yield curve would suggest. Buy them back on a regular and predictable schedule, replace them with liquid new issues, and incidentally make a little money for the taxpayer by buying low and selling high. Joe admitted he had never understood why the program existed until now; Tracy noted that crypto has a word for this — dust, the fragments too small to be worth the transaction fee to move. Which is where the show gets its name, and the hosts noticed. This time, though, the stated reason for expanding buybacks was liquidity, while spreads, market depth and dealer balance sheets all looked normal. From Bessent's own remarks, Duffie reads it differently: he thinks yields are too high, and the word he used was signal.

![An upward-sloping yield curve with a few low-priced old-issue dots scattered below it, boxed inside a dealer balance sheet, and an arrow sweeping them back onto the curve](/figures/buyback-sweeps-the-dust-en.svg)

**Governments are smaller than bond markets, and someone in this story has paid tuition on that.** A few billion dollars against tens of trillions earns a name: a micro twist. Beyond the signalling, it doesn't move the needle. (Tracy said it sounds like an early-2000s alcopop; Joe said he'd try one.) The market reacted fast to the announcement and then reversed itself just as fast. Duffie then reached back to 1992, when Bessent was working at Soros's fund during the attack on sterling: the British government couldn't defend the pound and shouldn't have tried, and burned a lot of firepower doing it. If markets decide yields belong at 6%, the US Treasury doesn't get a strong say — not without taking on a lot of risk. He drew the boundary clearly, though: in a March 2020 situation, where the market itself is dysfunctional, stepping in with buybacks is entirely legitimate, and could be done on the scale of hundreds of billions rather than the four to eight billion under discussion.

**Shrinking the balance sheet is trapped on the side nobody looks at.** The new chair wants a smaller Fed balance sheet. Tracy went straight to the constraint: you can't shrink assets without shrinking liabilities. Duffie told her she got there faster than almost anyone he'd talked to. Go through the liabilities. The Treasury General Account — the Fed isn't going to call and ask Treasury to withdraw money. Paper currency — it isn't going to run ads asking the world to hand its banknotes back. That leaves reserves, the deposits commercial banks hold at the Fed. And 2005 is gone: back then reserves paid no interest, so banks had no appetite for them. Today, to control inflation, the Fed has to pay roughly the market rate, and reserves also satisfy liquidity regulation and settle payments. Duffie called them the Swiss army knife of finance — banks won't give them up easily. At Jackson Hole in 2017, Acharya and Rajan described the ratchet: every time the Fed expands and adds reserves, banks get more attached to holding them, and if you try to force the issue, markets get volatile and the Fed backs off.

![A staircase rising to the right, each step with a short downward arrow bounced back to where it was, and the floor a step higher each time](/figures/reserves-ratchet-en.svg)

![On the left an asset bar that has to shrink; on the right the liability bar is cut into four blocks, three marked immovable, with only the reserves block circled](/figures/shrinking-the-liability-side-en.svg)

## Where my thinking went

### "The numbers look fine, so why is the price still falling?"

This is the version most of us actually live with. The results are in line, the trend is intact, the story holds — and the price grinds lower while you start wondering what you missed.

Duffie's thought experiment offers another reading: maybe nobody disagrees with you. Maybe everyone who wants it at this price already has it. He stripped inflation and default risk out of the setup on purpose, to show that even when everyone agrees the asset is safe, growing supply meeting a fully-allocated buyer base still pushes the price down until a new price persuades someone.

There's a direct equity analogue: secondary offerings, convertibles, lockup expiries, index rebalancing. None of it says anything about whether the business is good. It says something about whose hands the new supply has to land in. I used to treat "fundamentals unchanged" as enough to file price weakness under noise. Now I try to ask one more question: over the next three to six months, where does the new supply come from, and who is the one being asked to take an extra bite. It doesn't make me right. It does help separate "I was wrong" from "it isn't time yet" — and those two call for completely different responses.

### "The authorities stepped in — should I follow?"

Two questions from this episode are worth asking before you do.

First, size. A few billion against tens of trillions. Micro twist settles it. Judge an intervention by its order of magnitude, not by the confidence of the press release. Second, and this was Joe's contribution: classical interventions work better when they're level-bound than volume-bound. Announce that 5% is the line and you'll take the other side of anything through it, and in theory you may never spend a cent. But once the market can count your ammunition, it will start counting. That's what happened to sterling in 1992.

![On the left half a thick baseline blocks a downward arrow and bounces it back; on the right half a row of ammunition cells empties one by one and a long arrow passes straight through](/figures/hold-a-line-versus-spend-a-pile-en.svg)

Duffie also pointed to a way to answer this after the fact rather than in advance: watch the reversal. The market moved quickly on the news and then gave it back. That reversal is the market's own verdict — it decided this was a statement, not a line. Reading the follow-through beats forecasting who wins, and it costs nothing, because you were never obliged to act in the first hour.

### "Everyone has known this for years — why now?"

Joe's question was the sharpest thing in the episode. Same deficits, same demographics, same political gridlock — all sayable ten years ago, while rates fell.

Duffie's answer was accumulation, not new information. That distinction seems worth keeping: some risks aren't waiting for a trigger, they're waiting for a stock to grow past what the existing absorbers can take. Until then the bears are right every year and lose money every year, and when the turn comes there's no headline you can point to as the cause.

So "everyone knows" isn't evidence that nothing will happen. It only means the information is public. The thing to track is the quantity that keeps building, and how much room the people absorbing it have left. That's harder and slower than following the news, and it will probably have you nervous two years early. But at least you're watching something that moves, instead of re-reading the same old story every morning.

## Worth following up

- Bloomberg Odd Lots, episode of 3 September 2026, with Darrell Duffie of Stanford, recorded at Jackson Hole.
- Reading mentioned in the conversation: John Cochrane on the fiscal theory of the price level; the Acharya–Rajan paper on the reserves ratchet, presented at Jackson Hole in 2017; Cochrane and Piazzesi on decomposing the term premium.
- Duffie mentioned work in progress with Michael Fleming and Or Shachar of the New York Fed and his Stanford PhD student Sam Wicherle, on the effects of the buyback program.
- The closing banter earns its place: Joe's science-fiction premise where debt-crushed governments give way to AI companies as the new sovereigns, and the risk-free rate becomes the Claude yield and the Gemini yield; plus the detail that Bessent's apparently well-timed buyback trade doesn't exist yet, because the purchases don't start until 9 September.

## The one thing to take away

Prices are set by the most reluctant buyer, not by the average opinion. That's the only line I'm keeping from this episode. A room full of people agreeing something is good is one thing; that same room willing to hold more of it at today's price is another. The second one sets the price, and it usually comes down to the handful of people who still have room.

![Five containers side by side, the first four filled to the top and the last only half full, with an arrow pointing at the one that still has room](/figures/the-one-with-room-left-en.svg)

Here's something I've tried, and it has nothing to do with markets. Pick something you've been trying to move and can't — persuading family to relocate, getting a team to adopt a new process, asking friends to join you in something inconvenient. Don't ask whether people think the idea is good. Write down one name: the person who isn't convinced yet and without whom it can't happen. Then write one line underneath — what they'd have to receive before saying yes. Time, standing, one less responsibility, something you handle for them first.

The first time I did this, I found I'd spent two months adding arguments for people who had already agreed, and had never seriously thought about what belonged on that one line.