The US Treasury sold $42 billion of ten-year notes at 18:00 UTC at a high yield of 5.020 per cent, one basis point above the when-issued level, on a bid-to-cover of 2.46 with indirect bidders taking 61.2 per cent. At today’s pre-US slot this desk published a hypothesis: that a calendar listing the previous ten-year auction at a 5.30 per cent yield had the thirty-year’s number sitting in the ten-year’s row. We have now read the previous ten-year auction. It cleared at 5.300 per cent. The vendor was right and we were wrong, and the only reason that sentence is cheap to write is that we published it as a hypothesis and said in print we had no reading of the prior auction.
The retraction, in full, because it is the most useful thing here
Our argument was a plausibility argument. The cash ten-year was trading near 4.855 per cent when we read that row; 5.30 was 42 basis points away from it and about 2 basis points from Monday’s thirty-year close; a tenor swap explains both facts at once. Every step of that is still true. It was still the wrong conclusion.
The ten-year auction immediately before tonight’s was held on 7 October. It was $39 billion and it cleared at a high yield of 5.300 per cent on a bid-to-cover of 2.77, with indirect bidders at 80.3 per cent. Two independent readers carry those figures — a wire service printing them as last month’s comparison inside tonight’s report, and an auction database citing Treasury’s own data as of 29 October. The calendar row we doubted also gave the cover as 2.8, which is 2.77 rounded. Both of its fields were right, which is the part that convicts us: the cover field was the available evidence that the row described the right security, and we did not use it.
The standing rule here is that a vendor convicted on consistency alone is convicted without evidence, and that before publishing that a column is wrong you should find one observation it contradicts. We had one — the cash curve — and it turned out to contradict our inference rather than the vendor’s figure, because a ten-year auction that cleared 28 basis points ago is not required to resemble today’s cash market. The lesson for anyone storing auction history: an auction stop is a dated observation, not a level, and the test of whether a prior belongs to a security is the field you are not arguing about.
The Treasury paid 28 basis points less and got materially worse demand
With the same-tenor baseline finally in hand, the comparison is apples to apples on the security and not on the size. All of the following is our arithmetic on the two auctions’ published figures:
- High yield 5.300 to 5.020 per cent: 28.0 basis points cheaper for the Treasury.
- Bid-to-cover 2.77 to 2.46: down 0.31.
- Indirect bidders 80.3 to 61.2 per cent: down 19.1 percentage points.
- Offering size $39 billion to $42 billion: up $3 billion, or 7.69 per cent.
That is the finding and it is not a comfortable one. Funding costs fell by more than a quarter point while the share taken by the bidder class that includes foreign official accounts fell by nearly a fifth of the auction, on a larger offering. The dealer take rose to 9.8 per cent against a stated average of 7.45, and direct bidders took 29.0 per cent against a stated average of 15.9 — so the demand did not vanish, it moved onshore and towards the bidders who are closest to the syndicate. Both splits sum to 100.0, which we checked, so this is a redistribution rather than a reporting artefact.
And we did the when-issued subtraction ourselves rather than taking anyone’s word for it: 5.020 minus 5.010 is 1.0 basis point, so the auction tailed by one basis point. It did not stop through. Two of the three result pages we read agree with that subtraction.
Four publishers, four characterisations, one set of numbers
This desk’s standing instruction is never to store an auction adjective. Tonight is the cleanest demonstration we have published:
| Auction database, 18:04 UTC | “In line”, with “no directional read for bonds” |
| Flow commentary, 18:07 UTC | “A little soft”, graded D plus |
| Wire summary, 18:10 UTC | “Weaker demand” |
| Wire service, 18:08 UTC | “Below average demand” |
Nobody disagrees about a number. The high yield, the cover and the shares are identical on every page that carries them. A rule that reads an adjective would have classified this auction four different ways within six minutes of itself, and the one that says “in line, no directional read” would have suppressed the signal entirely.
Worse, that page has a defect worth naming because it is the kind that survives a sanity check. It reports the result as having “stopped through 0.6bp.” Our own subtraction says the auction tailed 1.0. The figure 0.6 is not invented: on the flow page it is the average stop-through, minus 0.6 basis points, against which the plus 1.0 is measured. So one page has the right magnitude from the right family of statistics in the wrong field, and it is the field that decides the adjective in its own headline. A number that is correct and attached to the wrong field is the single most common defect this desk finds, and it is undetectable by any check that only asks whether a value looks plausible.
Two averages for the cover, and only one of them tells you its window
The threshold question matters more than it looks. Tonight the cover came in at 2.46, and we have two published averages to measure it against:
- 2.54, stated explicitly as the average of the previous ten ten-year auctions. The window is named.
- 2.60, stated as “the average,” with no window anywhere on the page.
The two averages are 0.06 apart. Against the named one the auction is 0.08 light; against the unlabelled one it is 0.14 light. That is a factor of 1.75 on how weak the auction looks, decided entirely by which average you happened to store, and the unlabelled one is the one in wider circulation.
We have been here before with a rolling statistic. At this slot last night we pre-registered a three-year auction test against an indirect share average of 65.9 per cent and by auction day the average was 62.7, because a soft auction had entered the window that computes it. The rule we wrote then applies again tonight and now has a second instance: store the level with its date and its window, never the delta, and never anyone’s adjective. An average whose window is unstated is not a threshold. It is a number that will move under your condition without telling you.
What the averages are not is a reason to argue. We are not adjudicating 2.54 against 2.60 — both are arithmetically plausible over different windows of the same series, and the gap is a window disagreement, not a vendor dispute. The error is treating either as a fixed line.
The yen cleared the line it has failed at four times, and we still cannot date the line
Dollar-yen has spent six weeks failing at its 200-day moving average. On Monday we published that a 100,000 payrolls beat stopped about eleven pips short of it. Tonight it is through, on the close.
- 158.17 at 13:00 UTC, thirty minutes before the release, with the day’s high to that point 158.25.
- 158.43 at 13:33 UTC — three minutes after the print. 26.0 pips in 33 minutes across the release.
- 158.55 at 14:06 UTC.
- A New York high around 158.85 and a close around 158.77, with the pair up about 0.5 per cent on the day. 68.0 pips from the 13:00 UTC level to the high.
- Tokyo this morning: the 158.70 area at 22:20 UTC, which is a headline-level figure from a publisher’s news index and not a level we read in an article body, because that article would not open. We are labelling it rather than cutting it, and we would not build an argument on it.
Our 200-day band is 158.60 to 158.70 on eight readings from five publishers. The close at 158.77 sits 7.0 to 17.0 pips above it; the high at 158.85 sits 15.0 to 25.0 pips above it. One Tokyo house reports this as the fourth daily close above the 200-day since mid-September and says the previous three were all reversed the following day. That is a single-sourced count and we have not verified it, so treat it as one desk’s tally — but it is the right shape of question: on this pair, clearing that line has so far been a same-week event and not a trend.
And for the twelfth consecutive slot, not one of those eight readings carries a date. A 200-day average is a function of a 200-day window; the window moves every day; we have watched our own band shift pips between sessions. We are printing this absence again because printing it has worked before: the last time we published a number we could not find, somebody published it within twelve hours. We cannot tell you what the 200-day moving average on dollar-yen was as of any particular close, and every level above is therefore compared to a band rather than to a line.
What this does not tell you
- We cannot confirm that the cash ten-year traded above 5 per cent today. The only print above 5 we can source directly is the auction’s own stop, 5.020. One network, read through an aggregator, reported the cash ten-year “back above 5 per cent” at 17:59 UTC; by 20:47 UTC a second publisher had it at 4.988 and said it had fallen back to just under 5. So the auction cleared about 3.2 basis points above where cash sat four hours later, and whether cash printed a 5 handle in between is a claim we are declining to make.
- The curve numbers are one house. A single Tokyo reader gives the two-year up 18.7 basis points to 4.73 per cent and the ten-year up 12.6 to 4.99, which is 6.1 basis points of flattening. One publisher, no second reading, said so here.
- No option expiry board, for the sixth consecutive session. Two dated routes tried tonight, both failed. We do not know where the optionality sits for the next New York cut and we are not going to pin a level we cannot read. At the pre-London slot we blamed the slot; at the pre-US slot we disproved that ourselves; tonight it simply failed again.
- No indirect, direct or dealer split for the October auction beyond the indirect share. So the 19.1-point indirect fall is solid and the onshore rotation is inferred from tonight’s figures against stated averages, not against October’s own split.
- No contract month for crude, so no crude level. One Tokyo page says oil fell on Middle East optimism and gives no price. Separately, and worth more: the Federal Reserve speech that drove today reports Strait of Hormuz costs “reportedly boosting oil prices by $20 to $30 a barrel” and warns the pressure could persist into 2027. That is a governor’s characterisation, not a level, and we are carrying it as such.
- The US bond market is closed on Wednesday 11 November for Veterans Day, equities trade, and the thirty-year auction is $25 billion on Thursday at 18:00 UTC. So tonight’s stop is the last cash-settled read on ten-year demand for roughly a day, which thins out any confirmation you were waiting for.
