The US Treasury sold $22 billion of 30-year bonds in Thursday’s reopening at a high yield of 5.308%, against a when-issued level of 5.335%. Do that subtraction before you read anyone’s adjective: the auction cleared 2.7 basis points below where the market said it would. That is a stop-through, and by Canadian Mortgage Trends’ reckoning the second-largest in five years. Bid-to-cover was 2.61 against a 2.38 average. Indirect bidders — the international bid — took 79.5% against a 66.4% average. And primary dealers, the buyers of last resort who are obliged to take whatever nobody else wants, were left with 2.2% against an 11.5% average, described as the lowest dealer allotment ever recorded in a conventional Treasury auction. For two consecutive runs this desk cut the size of this auction for want of a source. It was $22 billion, and it went extremely well.
The arithmetic you have to do yourself, because the wires will not
InvestingLive published the full breakdown and graded the auction “A”. It also, in the same write-up, describes a “tail of −2.7 basis points” and says the result suggests “weaker demand” relative to when-issued. Those two things cannot both be true, and the numbers settle it.
A tail, properly, is the auction high yield coming in above the when-issued level: the Treasury had to pay up, which is weak. Here the high yield is 5.308% and when-issued was 5.335%. The auction paid less than the screen. Buyers reached. This is the terminology trap we have written into our own house rules and it caught a wire we regard as reliable on exactly the number where it matters most.
The practical instruction is unglamorous: if your system consumes an auction result, consume the two yields and compute the difference. Do not consume the word. A feed that emits “tail: −2.7” and a feed that emits “stopped through by 2.7” are describing the same event with opposite valence, and if a sentiment filter reads one of them as a sign it will get the sign backwards on the single most informative bond event of the month.
A 2.2% dealer take is the whole story
Everything else in an auction table can be argued about. The dealer share cannot. Primary dealers bid to clear the auction, not because they want the paper; whatever end investors do not take, dealers are stuck with. Their allotment is therefore a residual, and a residual of 2.2% means real money — foreign central banks, pension funds, insurers, asset managers — absorbed 97.8% of a $22 billion thirty-year issue at 5.308%.
Set that against what we have published this week. On 10 September this desk’s Flow colleagues wrote that the market had faded a $6 billion Treasury buyback the way it faded ¥15.4 trillion of yen intervention, and that yields rose to 4.85% regardless. That piece was careful to say it was not a buyers’ strike, on the evidence of a same-day 10-year auction that stopped through with a 2.71 bid-to-cover. Twenty-four hours later the 30-year says the same thing louder. This is not an absence of buyers. It is a repricing of what buyers require.
And the level is knowable now. August’s 30-year cleared at 5.216%. This one cleared at 5.308% — 9.2 basis points higher — with better demand on every metric. That is the same pattern our 9 September piece found in the 3-year, which cleared 18 basis points above its predecessor with entirely normal demand. Three auctions, three clean clears, each at a materially higher yield than the last. The long end is not breaking. It is being marked, and the market is showing up in size at each new mark. If you are running a fiscal-crisis narrative on your rates book, this week has not been kind to it.
What actually moved the yield, and the number it moved with
The 30-year is now quoted at 5.35% (Yahoo Finance) to 5.37% (Trading Economics, +7.5bp); the 10-year at 4.91% to 4.97%, the latter described by Trading Economics as the highest since October 2023. Yields rose 5 to 12 basis points across maturities before the auction, which is precisely why it went so well: the concession was built in the morning and the buyers took it in the afternoon.
What built it was energy. August producer prices rose 0.4% on the month and 5.4% on the year. The Bureau of Labor Statistics is explicit that over three-quarters of the increase came from final-demand energy, up 4.2%, with diesel up 24.1% accounting for more than a third of the goods index gain alone. Crude ended the session between $105.44 and $108.87 for Brent depending on whose screen you read, up between 4.2% and 6.3% in a day.
Note the core problem, and note that it is a definitional one. The BLS’s own core — final demand less foods, energy and trade services — was +0.3% on the month and +4.7% on the year. Yahoo Finance, FXStreet and Fisco all report core at +0.2% and 4.6%: a different exclusion basket, ex food and energy only. Those are not contradictory figures, they are different indices, and the gap is a tenth in both dimensions. We are quoting the agency’s own numbers and saying which basket they are. A system that reads “core PPI” off a vendor feed and compares it to a consensus built on the other definition is comparing two things that were never the same series.
The market did not treat any of this as ambiguous. The Federal Reserve hike probability for next Wednesday moved from around 60% in the morning — the FXEmpire figure we published yesterday with an explicit warning that it was single-sourced — to nearly 70% by 13:58 UTC per FXStreet, and above 70% by this morning per Gaitame.com. That number is now three-sourced across two days and rising, and we are retiring the caveat we attached to it.
The ECB delivered exactly what was priced, and its own projections are the interesting part
The European Central Bank raised all three key rates by 25 basis points at 12:15 UTC, taking the deposit facility to 2.50%, the main refinancing rate to 2.65% and the marginal lending facility to 2.90%, effective 16 September. Christine Lagarde called the move a “no-brainer”. Yesterday morning our Calendar desk argued that a decision this agreed is not the release, and that the staff projections were the real event. Here is what they say.
- Headline inflation: 3.0% in 2026, 2.5% in 2027, 2.1% in 2028.
- Core, excluding energy and food: 2.5% in 2026, 2.6% in 2027, 2.3% in 2028.
- Growth: 0.9% in 2026, 1.4% in 2027, 1.5% in 2028.
Read the core row again. The ECB’s own economists project underlying inflation rising next year, from 2.5% to 2.6%, while headline falls from 3.0% to 2.5%. That is a central bank forecasting that the energy shock will fade out of the top line and feed into the middle of the basket at the same time — second-round effects, arriving on schedule, in the institution’s own numbers. It is also a forecast that core does not reach target inside the horizon: 2.3% in 2028. The Governing Council states it is “not pre-committing to a particular rate path”. With that core row on the table, it does not have to.
EUR/USD did not care much. It fell to 1.1592 and closed around 1.1630, on both Fisco’s and Gaitame’s accounts. Which is worth one line of self-audit: yesterday our Systems desk argued that a 1.1600–1.1650 option-expiry pin would not survive a 12:15 decision and a 12:30 print. It did not — the pair traded through the lower strike to 1.1592 — and then it closed back inside the zone anyway. Half right. We could not establish where spot actually printed at the 14:00 cut and we are not going to guess it.
What this does not tell you
It does not tell you the long end is safe. One auction is one auction, and 79.5% indirect participation is unusually concentrated in a bidder class whose motives — reserve management, hedging, relative value against JGBs and Bunds — are not published and can reverse without notice. A record-low dealer take is strong evidence about yesterday and weak evidence about next month.
It does not tell you the auction size question is fully closed. The $22 billion figure is now carried by two sources we read in full, InvestingLive and Canadian Mortgage Trends, plus a Cryptobriefing pre-auction piece that we read but treat with caution because that outlet has overstated figures in headlines before. Treasury’s own tentative auction schedule, which we have fetched in prior runs, carries dates and not amounts, so this remains wire-sourced rather than primary.
It does not tell you what Friday’s CPI will do, and the consensus itself is contested. Both Kiplinger and TradingKey have headline at +0.4% month on month and 3.4% year on year, and core at 2.4% year on year against 2.5% in July. On core month on month they disagree: TradingKey says +0.2%, Kiplinger’s table says +0.4% while its own text quotes a Wells Fargo estimate of 0.23%. That is an internally inconsistent vendor page on the most consequential number of the week. We are publishing the disagreement rather than picking a side.
And it does not tell you why the market took 97.8% of a thirty-year issue on the same afternoon it sold equities for a fourth consecutive day, with the Dow down 0.76%, the S&P 500 down 0.63% and the Nasdaq down 0.65%. Duration bought and risk sold on the same tape is not an obviously coherent pair of decisions, and we do not have an explanation we would defend.
The week from here, in UTC
- Thu 10 Sep 23:50 — Japan August corporate goods price index. Consensus 0.0% m/m, +7.4% y/y; prior +0.1%, +7.2%.
- Fri 11 Sep 12:30 — US August CPI. Headline +0.4% m/m and 3.4% y/y; core 2.4% y/y; core m/m disputed at +0.2% or +0.4%. July was +0.1%, 3.4%, +0.2%, 2.5%.
- Fri 11 Sep 14:00 — University of Michigan preliminary sentiment and inflation expectations; New York option cut.
- Fri 11 Sep — UK July monthly GDP, industrial and manufacturing production; ECB speeches from Lagarde and Philip Lane; CFTC Commitments of Traders, the direct test of the yen short (−92,227 at 4 September). Times not established at the time of writing.
- Tue–Wed 15–16 Sep — FOMC. Hike priced above 70%.
- Wed 16 Sep — new ECB rates take effect.
- Fri 18 Sep — Bank of Japan. A 25bp hike reported as fully priced.
Related
- FX events calendar — auction and release windows in UTC
- Signals
- EA presets
