The United States sold 22 billion dollars of thirty-year bonds at 17:00 UTC yesterday at a high yield of 5.618 per cent, one tenth of a basis point above the when-issued level. That is a tail — a real one, by the strict definition, for the first time this desk has had cause to use the word correctly. Primary dealers were left with 6.79 per cent of competitive awards against a stated average of 10.3 per cent. A month ago the same publisher gave that average as 11.5 per cent. We went and worked out what the average is, and it is the last six auctions of this security — which means “below average” is a statement about five numbers you already had plus one you just learned.
The result, and the word used correctly for once
The numbers first, because two readers agree on the load-bearing ones. High yield 5.618 per cent. When-issued 5.617. Bid-to-cover 2.54 against a stated average of 2.41. Dealers 6.79 per cent, directs 20.89, indirects 72.32. A second reader independently gives 5.618, 2.54 and 72.3 with a 22-billion-dollar size and dealers at 6.8, and grades demand average where the first reader grades the auction B.
Do the subtraction and show it, because this desk has a standing rule about the word: 5.618 minus 5.617 is 0.1 basis points, the high yield came in above the when-issued, and that is a tail. One month ago the same publisher labelled a result “tail: minus 2.7 basis points” when 5.308 against a 5.335 when-issued was a stop-through — a phrase meaning the opposite of the word it started with. This month the sign is positive and the label is right. We note that without charity: the field is correct when the number is positive and was wrong when it was negative, which tells you the label is probably generated from the arithmetic rather than chosen, and that a reader who only ever sees one month will never find out.
The repricing is the bigger number and nobody should need a desk to point at it. September cleared 5.308. Yesterday cleared 5.618. That is 31.0 basis points in one month on the same security. One reader calls it the highest thirty-year clearing yield since August 2000, which is editorial and is that reader’s description, not a fact we have checked. What we can check: a preview filed on 5 October put the thirty-year secondary at 5.61 per cent as of 1 October, so the auction cleared 0.8 basis points above where the cash market had been a week earlier. The supply was absorbed essentially at the market, not through it.
The average moved 1.2 points in a month, so we reconstructed it
Here is what set this article off. The 6.79 per cent dealer take is quoted against an average of 10.3 per cent. In September, the same publisher, the same security, the same field, quoted that average as 11.5 per cent. The bid-to-cover average went from 2.38 to 2.41. The indirect average went from 66.4 to 69.1. The direct average went from 22.1 to 20.6. Four stated averages, all four moved, inside one month, on a security that is sold once a month. A single new observation cannot move a long average by 2.7 points, so either the window is short or the averages are computed on something other than what they appear to be.
The window is short, and we can show it. Take the last six thirty-year auctions before September — March through August — and average each field. Bid-to-cover: 2.383, against a stated 2.38. Dealers: 11.500, against a stated 11.5. Indirects: 66.433, against a stated 66.4. Now roll the window forward one auction, April through September, and do it again. Bid-to-cover 2.410 against a stated 2.41. Dealers 10.300 against a stated 10.3. Indirects 69.117 against a stated 69.1.
Eight reconciliations across four fields and two windows, four of them exact to the decimal place the publisher prints. The direct-bidder field misses by 0.05 and 0.07, which is what you would expect from averaging figures we hold to one decimal when the publisher holds them to two — our source gives dealers as 6.8 where the publisher gives 6.79. The average is a six-auction rolling average of the same security. The publisher does not say so anywhere on the page.
That is ours, derived, and we will not dress it up further than it deserves: eight reconciliations make coincidence implausible, but they do not exclude some other rule that happens to produce the same two sets of numbers. What they do establish is that the denominator is small and that it rolls.
What that does to the only figure anyone quotes
Now the consequence, which is the reason this is a flow article and not a trivia one. September’s auction left dealers with 2.2 per cent of competitive awards, a fifth of the then-stated average, and this desk published that as the payload of the month. That 2.2 per cent is now inside the window. It is one of the six numbers that produced yesterday’s 10.3 per cent benchmark, and it is what dragged the benchmark down 1.2 points.
So the sentence “dealers took 6.79 per cent against an average of 10.3” contains the same event twice. The extraordinary month that made September worth writing about is now part of the yardstick that makes October look only mildly unusual. Measured against the average as it stood before September — 11.5 per cent — yesterday’s take is 4.7 points light. Measured against the average that now includes September, it is 3.5 points light. Same auction, two different degrees of strangeness, and the difference is entirely which of your own previously published observations has rolled into the denominator.
If a system keys off a “dealer share versus average” signal, this is a specification problem rather than a data problem. A rolling six-observation benchmark on a monthly security has a memory of half a year and no more, so a genuinely abnormal month becomes normal to the signal within six prints, and two abnormal months in a row will hide each other. The fix is not complicated: hold the raw shares and compute your own baseline over a window you chose, rather than reading the publisher’s. The point is that until this morning we did not know we were reading somebody else’s window at all.
Eighty-five pips, two pips through the twenty-day, and no timestamp to pin either
The price action around all this was substantial and we cannot connect it to the auction, which we would rather say than imply. Two publishers give the New York session high as 158.37 and 158.36 — one pip apart, which is a strong two-reader confirmation — and the low as 157.52. That is an 85.0-pip range on the day. The pair was in the 157.80s at 22:34 UTC, 28.0 pips off the low, and one morning note puts the day’s move at about minus 0.1 per cent.
The low matters for one specific reason. A technical note filed yesterday morning put the twenty-day exponential moving average at 157.54 and named it the immediate downside support, with the then-current 158.19 area as the pivot and the 24 September high at 159.04 as the main resistance. The session low of 157.52 is two pips below that published average, and the pair closed back above it. Two pips is not a break by any standard a sane system would use, but it is a clean, checkable score on a level published hours before it was tested — which is more than this desk can say for most of the levels it reads.
What caused the reversal is attributed, by one reader, to politics rather than to supply: the dollar firmed on the claims print and on concern about renewed strikes on Iran, which lifted oil and yields, and then sold off when the president described talks as constructive and ruled out strikes before the midterms, taking oil back down. We have no timestamp for either the 158.37 high or the 157.52 low, so we cannot place either against the 17:00 auction, and we are not going to assert a causal chain we cannot time. One reader does say yields turned lower into the close. A thirty-year that cleared 31 basis points higher than last month and a long end that rallied afterwards are both true; which one the yen was trading is not established here.
One morning forecast range for today runs 157.200 to 158.600, 140 pips wide, with spot entering it 26.0 pips above the twenty-day average and 124.0 pips below the September high.
19:30 today: two publishers, one Tuesday, and a prediction of ours that gets scored
The positioning report lands at 19:30 UTC today, measured as of Tuesday 6 October — the first reading that contains the payrolls week. Yesterday morning this desk published the state of the file and it is not flattering: two readers of the same report for the same earlier Tuesday put leveraged funds net short 14,161 and net short 16,809 contracts, one of them puts asset managers net long 51,961 in the same sentence, and our own carried figure of 55,440 net long has no category written next to it at all. Four numbers, one Tuesday, 72,249 contracts apart.
We also published a prediction, and today is the day it is testable. The claim was that the leveraged-fund net short widens, and that the two publishers stay roughly 2,600 contracts apart on the level while agreeing within a few hundred contracts on the change. That second half is the part worth watching, because it is the one that says the disagreement is a definitional offset rather than a measurement error. If they agree on the change again, the gap is a fixed difference in what each one counts, and either figure is usable for direction as long as you never mix them. If they disagree on the change, neither is usable for anything.
The sizing note, since that is what this is for: the report is measured at Tuesday’s close and both publishers label it by Friday’s release date, so what arrives tonight is already three days old and will be four by Monday’s open. It is a slow-moving constraint on how crowded a position is, not a trigger. Treat a 19:30 figure as a reason to resize on Monday, not as an event to trade at 19:30.
What this does not tell you
The six-auction reconstruction is our arithmetic. Four of eight reconciliations land exactly and four land within 0.07, and the misses are confined to the one field where our inputs are rounded more coarsely than the publisher’s. We have not found the publisher stating a window anywhere, and we have not tested the hypothesis against a longer window because we do not hold enough history to do it. A rule that is not six auctions but produces these two sets of averages is not excluded.
The per-auction history we averaged comes from a single reader. It agrees with independent readings on three separate months — September’s 79.5 per cent indirect share, August’s 66.8 and the 5.046 per cent May clearing, which one other publisher gives as 5.050 — so it has been corroborated where corroboration was available, and it is still one table.
We have no when-issued figure from a second source for yesterday, so the one-tenth-of-a-basis-point tail rests on one reader giving both legs of the subtraction. The direction is what matters and the magnitude is inside the noise of almost any secondary quote, so do not build anything on the sign of a tenth of a basis point.
On the price: the high, the low and the morning level come from three different publishers at three different minutes, and the 85.0-pip range is therefore our arithmetic across sources, which is precisely the criticism this desk has made of other people’s session ranges. The two-pip pierce of the twenty-day average is as good as the published average was, and that average was quoted as of yesterday morning rather than at the moment of the low.
And the honest limit on the whole piece: nothing above tells you which way the pair goes. A thirty-year tail of a tenth of a basis point is not a yen signal. What it is is a reason to stop quoting somebody else’s rolling average as though it were a fixed benchmark.
Related
- FX events calendar · the 19:30 UTC positioning release and today’s 12:30 and 14:00 prints
- Signals
- EA presets
