The Treasury sold sixty-nine billion dollars of two-year notes at 17:00 UTC yesterday and the result is soft in exactly one respect and unremarkable in every other. The high yield was 4.708 per cent against a published when-issued level of 4.705, a tail of 0.3 basis points — larger than September’s 0.2 and, more to the point, larger than a six-auction average of minus 0.2, which is to say this tenor normally stops through. Bid-to-cover 2.59 against 2.63 last time. None of that is the finding. The finding is the allocation: indirect bidders took 45.7 per cent against 57.8 last month, direct bidders took 41.3 against 29.0, and the dealer take moved by two tenths of a point. A quarter of a sixty-nine billion dollar book changed hands between two bidder classes and the one number most risk rules actually watch did not move.
Our own derivation method got its first real test, and it passed
Yesterday morning this desk published a preview of this auction which had to derive the September when-issued level by subtraction, because the only source carrying a tail published no when-issued. September cleared at 4.787 with a stated tail of 0.2 basis points, so we published an implied when-issued of 4.785 and labelled it ours twice, with the rounding caveat attached.
Yesterday’s result is the first time that method could be checked against a published figure. Two readers give the when-issued at 4.705 and the high yield at 4.708. Subtract and you get 0.3 basis points, which is the tail both of them print. The derivation reproduces exactly. That is one observation, and it is one more than we had.
What it does not do is rescue the distribution problem the preview raised. The tail field on the source we used is populated for three of twelve auctions, and a single confirmation of the arithmetic does not create a history to size against. You can now trust the subtraction. You still cannot tell a system what a normal tail looks like from three numbers, and the six-auction average of minus 0.2 that a different publisher supplies is a fourth window on the same question rather than an answer to it.
The dealer take is the number that did not move, and it is the one most rules read
Set the three bidder shares side by side. Indirect 45.7 against a prior 57.8 and a six-auction average of 58.3. Direct 41.3 against 29.0 and 30.4. Dealers 13.0 against 13.2 and 11.3. All three columns sum to a hundred, which we checked, so this is a genuine rotation and not a definitional artefact.
Twelve point one points of indirect share is about eight and a third billion dollars at this size. Twelve point three points of direct share is about eight and a half billion. Those two numbers are the same trade seen from both ends, and the class the market treats as a proxy for foreign and real-money accounts stepped back by roughly an eighth of the auction while domestic direct bidders stepped into it.
Now consider what a risk schedule usually keys on. Dealer takedown is the standard distress indicator: dealers are the residual buyer, so a jump in their share means the auction did not find natural demand. Dealers took 13.0 per cent against 13.2 last month. Two tenths of a point. On that field, and on that field alone, nothing happened. If your rule for “was the auction absorbed” reads dealer share, yesterday was a non-event, and it was a non-event on a day when a quarter of the book rotated between bidder classes. That is not an argument that dealer share is useless. It is an argument that it answers a narrower question than people ask of it, and the broader question needs the other two columns.
Four publishers read identical numbers four different ways
Every figure above is agreed. What is not agreed is what it means, and the spread of characterisations is wider than the spread of numbers by a long distance. One publisher calls the auction in line and says explicitly that it offers no directional read for bonds. A second headlines it as attracting slightly below average demand. A third calls the 0.3 basis point tail modestly soft rather than weak, notes that front-end auctions usually behave well, and says the signal only matters if the five-year and seven-year show the same pattern later this week. A Japanese morning note simply calls the result weak and attributes the subsequent rise in yields and the bid for dollars to it.
There is no vendor error in that set. It is four honest readings of a 0.3 basis point tail, and a trader who stores “the auction was weak” as a fact has stored somebody’s adjective.
The bid-to-cover average makes the same point with a number instead of an adjective. One publisher gives the average as 2.64 over six auctions. Another gives 2.62 over ten. Both are correct and the entire two-hundredths gap is the window — which is precisely what we published yesterday morning about two pivot ladders 46.6 pips apart, where all ten levels reproduced from their inputs and the whole disagreement was whether the inputs were last week’s or Friday’s. We wrote then that a pivot carries a window the way a print carries a vintage. One trading day later the same structure turns up in a Treasury auction statistic, in a different asset class, with a different unit. A rolling average has a window and nobody labels it.
The yen’s high was ninety minutes before the auction, and the concession stayed in the front end
The preview asked, as its fourth pre-registered question, whether a front-end auction forty-nine hours before a Federal Reserve decision would move dollar–yen at all. It did not.
One reader puts the session high at about 157.64 near 11:30 New York time, which is 15:30 UTC — ninety minutes before the auction closed. The pair finished around 157.52, twelve pips below that high, on a day that ran 99 pips from a low near 156.65 in the Asian morning. The move that produced the high is sourceable to something else: the Dallas Fed manufacturing index printed 13.4 against 9.8 prior at 14:30 UTC, a beat of 6.4 points against one reader’s forecast of 7.0, and a second publisher describes the figure as the highest since January of last year against expectations of a decline. The ten-year traded near 5.01 per cent at midday, fell to about 4.96 by 14:15, and bounced to near 4.99 after that release. The dollar’s high came an hour later. The auction came two and a half hours after that and the pair was already done.
Where the concession did land is the curve. By the late session the two-year was 4.7159 per cent, up 3.39 basis points on the day and 0.79 above the auction clear, while the ten-year was 4.9863, down 0.17, and the thirty-year fell 1.65. So the two-year to ten-year spread narrowed to 27.0 basis points from 31.0 at Friday’s close, a flattening of about four. The auction repriced the sector it was in and nothing else. For a system that holds correlated exposure across the front end and the long end, that is the useful shape: this was a supply event with a local footprint, not a duration event.
One correction against ourselves while we are here. The preview said the note came to auction roughly ten basis points below September’s clear — 10.7 on one reader’s Friday close of 4.68 and 9.7 on another’s 4.69. It cleared 7.9 below. The gap narrowed because the level rose into the auction by 1.8 to 2.8 basis points between Friday’s close and the stop, so our published figure was a statement about Friday being used as a statement about Monday.
What this does not tell you
- It does not tell you why indirect bidders stepped back. We have no bidder-level data and nobody publishes any. A Federal Reserve decision in forty-nine hours, a settlement date of 2 November that falls the day after the American clock change and the day before the midterms, and five consecutive weeks of Japanese net selling of global bonds reported by one reader are all available as explanations, and we are not choosing between them.
- The six-auction averages come from one publisher and are not independent of the figures they are compared with. The ten-auction average comes from a second. We have no source that publishes both windows.
- The 7.0 forecast for the Dallas Fed index is single-sourced. The 13.4 print is two-sourced.
- The session high and the time it printed come from one reader’s intraday narrative. Four publishers agree the pair finished between 157.42 and 157.52 by the close, which is a 10-pip band and not a disagreement worth adjudicating, but only one of them timestamps the high, so the ninety-minute figure has one source behind it.
- We read two closing prices for crude a dollar and 41 cents apart, and one publisher’s close sits below the other publisher’s stated session low. Oil bears on nothing in this article, so we are recording the discrepancy rather than resolving it. Equities and gold agreed across the same two readers to within a point and two dollars respectively.
- Nothing here is a view on whether the five-year today or the seven-year on Thursday tails. One publisher says explicitly that is where the signal would be, and it is a reasonable thing to say and completely untested.
Related
- Sixty-Nine Billion for a Thirteenth Straight Auction, 49 Hours Before the Fed — the preview this article scores, including the implied when-issued we derived and the four-security settlement stack.
- Two Publishers Named a Pivot 46.6 Pips Apart — the window argument, one trading day earlier and in a price rather than a statistic.
- FX Events — the week’s scheduled releases and auctions in UTC.
- Signals
- EA Presets
