“The problem with making assumptions is that we believe they are the truth. We could swear they are real.”— Don Miguel Ruiz
Researchers in Wurzburg have built robots roughly fifty times smaller than the width of a human hair. Small enough to work in water, among cells and bacteria. They can pick a bacterium up, carry it somewhere else, and put it down (uni-wuerzburg.de).
Collect. Transport. Release.
Sixty years ago that was a film. A submarine crew shrunk to microscopic scale and injected into a man’s bloodstream, because the clot sat somewhere a surgeon’s hands could not reach. They had sixty minutes before the miniaturization wore off and the ship expanded inside the patient. The clock was the whole plot.
Yesterday the Treasury did the same thing to the bond market.
It did not raise taxes. Did not cut spending. Did not change the auction schedule in any way a primary dealer would call ordinary. It shrank itself down, went inside the ten-to-thirty year sector, and started removing the obstruction by hand. Collect the bonds nobody wants. Transport the duration. Release bills in their place.
The operation begins 9 September. It runs through 4 November.
Election Day is 3 November.
Everyone spent yesterday arguing about the wrong word. Is it yield curve control? No. Yield curve control is a central bank promising unlimited quantity at a stated price, funded by printing. The Fed did it from 1942 to 1951. The Bank of Japan did it from 2016 to 2024. The promise is the machine; the buying is almost decorative once the market believes you.
What Treasury announced: a borrower raising per-operation purchases from $2bn to at least $4bn, funded by selling bills. Bounded. No price target. No promise. Roughly $128bn a year at that pace, which is close to 30% of expected issuance in those maturities and 2.4% of the stock outstanding (WSJ, citing Natixis). Enormous flow number. Trivial stock number.
So the label is wrong and the label does not matter.
Here is what matters. The 20-year auction yesterday afternoon stopped at 5.204% against a 5.163% prior. It tailed. Real money did not turn up for a bond at the highest yield of the modern tenor’s life, on the same morning the Treasury announced it would double its bid in exactly that sector.
And the 20-year still closed 11 basis points lower in yield.
A failed auction, absorbed. That happened yesterday and nobody wrote it down.
**THE OPERATION HAS A CLOCK.**Buybacks run 9 September to 4 November. The midterms are 3 November. Every long-end position now carries a dated distortion. Trade the expiry, not the announcement.**THE BELLY IS MANUFACTURED.**The 10s20s30s fly sat at 56 to 57bp for four sessions, then richened 7bp in one. Sell the belly. Only trade on the sheet that needs no direction.**BANKS PAID FOR THE RESCUE.**A bull flattener is the worst curve shape for net interest margin. Regionals worst but one of fourteen. Treasury just shelved the regional bank trade until November.**I’M WRONG IF.**The 10s20s30s fly closes above 58bp. Then the belly was cheap and the official bid is not the thing moving it.
The Treasury went inside its own bond market yesterday and started operating. Long-end yields fell hard, the 20-year fell hardest at 11 basis points, and a 20-year auction tailed four basis points into that same session without anyone caring. The curve bull-flattened: 2s10s from 52 to 46, 2s30s from 109 to 100. Which is the single worst configuration for a bank balance sheet, and financials duly took the beating. Regional banks minus 2.37%, investment banking minus 2.68%, diversified banks minus 2.08%, life insurance minus 2.36%, on a session the long end rallied nine basis points in their favor. Two-day dispersion inside the sector ran 557 basis points, the widest reading I have logged. Life insurance is now last of fourteen over two sessions, having lost on rising yields, falling yields and flat yields in succession, which retires the available-for-sale explanation and leaves the general account’s fifteen-year AI credit as the surviving one. Brent finally paid, up 2.58% to $93.98, and US energy equities fell anyway. The program expires the day after the midterms.
Start with what the intervention actually did to the shape of the curve, because the level is the part everyone reported and the shape is the part that pays.
The 2-year did not move. Not a basis point. Everything from five years out rallied, and the rally got bigger the further out it went until it hit the 20-year, then got smaller again.
Tenor18 Aug19 AugChange2Y4.19%4.19%0bp5Y4.37%4.35%-2bp7Y4.53%4.48%-5bp10Y4.71%4.65%-6bp20Y5.28%5.17%-11bp30Y5.28%5.19%-9bpLook at the 20-year against its own neighbors. Eleven basis points, against six at the ten and nine at the thirty. The middle of the target band outperformed both of its wings by a wide margin on the day the buyer of last resort announced itself.
Put a number on that. The 10s20s30s butterfly, twice the 20-year less the two wings, closed at 56, 57, 57 and 57 basis points on the four sessions running into Wednesday. It closed at 50 yesterday. Seven basis points, one session, off a plateau it had held all week.
No market re-rated the relative value of the 20-year sector overnight. A mechanical bid landed where the cheapest, most off-the-run-heavy paper sits, because that is what a liquidity-support buyback is designed to hoover up. And the program that produced it has a published end date.
Which gives me the cleanest expression on the sheet. More on it in the book.
Now the part I think consensus has genuinely mislabeled, and it is not the YCC argument.
The consensus read is that lower long yields are good for financials. Lower discount rate, lower funding stress, marks come back on the securities book, the regional bank case gets its life back. Every strategist note in circulation this morning takes that as read.
Financials were negative on both exchanges yesterday. Minus 0.52% on one composite, minus 0.39% on the other, on a session when the thing they are supposed to want fell nine basis points.
Go inside the sector and it is worse than negative. Carnage, in exactly the cohorts the theory says should have been celebrating.
The mechanism is not subtle once you say it out loud. Banks fund short and lend long. They are paid the spread between the two. A bull flattener rallies the long end, which is the asset yield, and leaves the front end untouched, which is the funding cost. Asset yields down. Funding costs unchanged. The spread narrows.
Bessent bought the asset side of the bank income statement and left the liability side exactly where it was. He financed it by issuing bills, which is the sector that sets deposit competition.
If you wanted to design a single policy action to compress net interest margin, you would design this one.
Second chart, and this is the one I keep going back to.
The Citi US surprise index prints 15.30. It ran above 60 in June. The index it is supposed to inform is at an all-time high.
Fine, divergences happen. Now impose the lag.
Take the ten-year yield, invert it, push it forward three months, and lay it over the surprise index. The fit is good and the direction of causation is the obvious one: borrowing costs rise, activity disappoints roughly a quarter later. The projected line has fallen hard at the right-hand edge, which is the April-to-May rate rise arriving. It runs down through September, October and November.
The economic disappointment is already in the post. It was mailed in the spring, at the yields the Treasury is now trying to undo, and it does not stop being in the post because the postman changed his mind.
Jim Paulsen has been making this point with a longer lens and I would rather quote his chart than paraphrase it (PaulsenPerspectives). His question mark is doing real work: during this bull market, when momentum slowed, the index struggled. Momentum is slowing now.
And the long version of the same divorce. Two series that moved as one instrument from 1960 through 2022, and have not since. Paulsen asks whether they ever reconnect. I would put it differently: one of them is measuring the economy and one of them is measuring nine hundred billion dollars of capital expenditure by six companies, and they stopped measuring the same country somewhere around the launch of the first frontier model.
Which brings the intervention back into focus. If growth surprises are queued to deteriorate into the autumn, and the administration’s own polling on the economy is 33% approval with gasoline above four dollars, then a program that ends on 4 November is not a debt management decision that happens to have a date on it.
I will state the inference plainly and label it as inference, because I cannot read anyone’s mind.
**Fact:**the buyback runs 9 September to 4 November and the election is 3 November.**Inference, moderate confidence:**the calendar is the point. Byzyka at Credent said the same thing in fewer words, that the move appears political and meant to drive rates down ahead of the midterms, and that it puts the validity of the bond market in question (WSJ).The counter-argument deserves a hearing. Buyback windows are routinely set to quarterly refunding cycles, and 4 November sits close enough to one that the date could be entirely mechanical. If the November refunding announcement extends the program at $4bn or better, the political reading weakens considerably and I will say so. If it lapses, it did its job.
The bill is not hidden and the people presenting it are not cranks.
Jay Barry’s team at JPMorgan made the argument that matters: absent genuine fiscal consolidation, a market that watches the borrower intervene in its own long end reads that as a credibility problem, and credibility problems show up as term premium (Bloomberg). Which is the opposite of the intended effect, on a lag.
Forty years of American debt management ran on two words, regular and predictable. The doctrine is the single largest reason the US term premium has sat structurally below what the fiscal position deserves. You are not compensated for guessing what the Treasury will do, because the Treasury told you in advance and then did it. That doctrine was suspended yesterday morning.
Total public debt crossed $40 trillion, $40.05trn at Tuesday’s close, up by a third in under five years (Bloomberg). The country now spends more servicing that than it spends on defense. The deficit is running near 6% of GDP against Bessent’s own stated 3% target.
So the arithmetic underneath the intervention got worse on the same day the intervention was announced. And the buyback is funded by bills, which shortens the weighted average maturity of the debt, which increases sensitivity to the front end, which is set by a committee that just recorded three dissents in favor of a hike.
My good friend PauloMacro ( a must-follow on substack and twitter) put the trade-off better than I would have:
*Bessent can save the dollar or the bonds. He is choosing the bonds, and just told you so, clear as day... Treasury will rain Tbills to do more and more buybacks even as the issuance mix moves in that direction already, which is very Brazil 90s of us and highly inflationary over time.*Metals are smelling exactly that. Gold at $4,486 is off two thirds of a percent this morning and up better than four percent on the week. Silver at $66.78. The dollar index at 98.60 and the broad dollar gauge at an eleven-week low, with the yuan at its strongest since early 2023. Metals firm, dollar soft, long end bid. Coherent set of prices, and the thing they are pricing is not disinflation.
Paulsen’s question on this one is the driest thing I read yesterday: if the US ever has another recession or crisis, would anyone notice? Policy uncertainty above the level that used to mark a crisis, sustained, for years, with no recession shading underneath it. The gauge has stopped discriminating. Worth remembering the next time someone tells you the market is complacent, because a market cannot be complacent about a signal that no longer separates anything.
I owe an update on this because it moved against me.
I have argued for three weeks that crude refuses to pay for the war and that the premium sits in refined product. Yesterday Brent settled up 2.58% at $93.98, with a session high of $94.08. My stated kill print is a settle above $95. $1.02 away.
The framing is not dead. One session from retirement, and I am not going to pretend otherwise.
But look at what happened underneath, because it is stranger than the headline.
Brent up 2.58%. And on the US tape: refining and marketing minus 1.08%, integrateds minus 0.34%, midstream minus 1.21%, oilfield services minus 1.37%, drilling minus 0.66%. Exploration and production plus 0.15%, which is a rounding error against a 2.58% move in the underlying. The only energy subsector that genuinely paid was coal, plus 5.36%.
The barrel paid. The equity refused.
Two readings. Either the equity market thinks $94 is a spike that reverses, in which case it is fading a chokepoint that has run six months with no negotiation scheduled. Or the equity market is telling you that a $94 barrel is a cost, not a revenue, for most of the complex, and the marginal energy business in America is a margin business that just watched its input cost jump 2.58% in a session.
The second reading is why my refining pair lost 123 basis points on day one, and I should have seen it coming. A crude spike compresses the crack mechanically before capacity scarcity expands it. The input repriced first. Timing error in the entry rather than a failure of the thesis, and the distinction only matters if the crack behaves from here.
And there is a third possibility that I had not considered until I read Mark Leonard this week, which is that the screen price and the clearing price have come apart.
Leonard’s argument is that we are not moving from an American order to a Chinese one but into a period with no order at all, and that its signature is the weaponization of chokepoints and the fragmentation of prices. His example: in early April, Brent futures traded around $113 while physical cargoes changed hands as high as $144 (Foreign Affairs). Futures and physical diverge all the time, by a dollar or two. Thirty-one dollars is a different phenomenon.
Identical barrels now trade at wildly different prices depending on where they are sold, through which payment system, to which buyer, under which sanctions regime. If that is even half right, then “Brent settled at $93.98” is a statement about one instrument in one venue, not about the price of oil, and I have been running a kill condition off a screen print that may no longer describe the market it is named after.
I am not changing the kill condition today, because a rule you move when it gets uncomfortable is not a rule. Brent settles above $95 and I say the framing is dead. But I am flagging that the instrument itself is degrading, and that is a bigger problem for everyone’s oil model than it is for mine.
Back to financials, because the sector produced the single most informative print of the session and it is not the one on the sector line.
Life insurance closed minus 2.36%, last of fourteen subsectors, on a day the 30-year fell nine basis points and the 20-year fell eleven.
Life insurers hold long-dated assets against long-dated liabilities. A long-end rally is a mark-to-market gain on the available-for-sale book. Falling yields hurt the reinvestment story over a decade, and help the balance sheet today.
I built a short leg on the available-for-sale mechanism ten days ago. Then on Tuesday life lost while yields fell, and I said out loud that my mechanism was incomplete. Yesterday life lost again, harder, while yields fell much harder.
Three consecutive sessions. Up days, down days, flat days. Same result.
A mechanism that predicts losses when yields rise, applied to…
