The Treasury reopened $13bn of twenty-year bonds at 17:00 UTC yesterday and the result deserves a careful reading, because the two things it says point opposite ways. Demand beat the recent run: a bid-to-cover of 2.73 against a six-auction average of 2.62 and 2.57 at the last sale, primary dealers taking only 10.3 per cent against an 11.9 per cent average and 16.9 per cent last month, and direct bidders at 31.6 per cent against 22.9. Three of the four demand measures came in better than average. And it still tailed. The high yield was 5.469 per cent against a when-issued level of 5.460, which is 0.9 of a basis point of stop-through — small, but 4.5 times the 0.2 basis point average, and the first auction this desk has been able to call a tail using the subtraction our own rulebook demands. It cleared 34.9 basis points above the six-auction average yield, 4.9 above last month, and 5.6 basis points above the thirty-year. One reader grades it C. Another calls it in line with no directional read. The figures say better bidding at a worse price, which is not a contradiction but is the whole point.
The tail is real this time, and here is the subtraction
This desk refused the word “tail” two days ago. Japan sold twenty-year paper on 20 October at a record 5.84 times bid-to-cover, and one reader described a 0.07 figure as the tail; we declined to use the term, because that figure was a gap between average and lowest accepted prices — a dispersion measure — and we had no when-issued level to subtract from. The strict definition is that the auction high yield came in above the when-issued level, and without the when-issued level you cannot test it.
Here you can. High yield 5.469 per cent, when-issued 5.460 per cent, difference 0.9 of a basis point, auction yield on the high side. That is a stop-through by the definition, stated with its inputs so you can check the arithmetic rather than take the characterisation. It is also a small one. Against a 0.2 basis point average tail it is 4.5 times as wide; against last month’s 2.0 basis points it is less than half, a reduction of 1.1 basis points. So the concession demanded by the market shrank by more than half while the clearing yield rose 4.9 basis points. Those are different quantities and they moved in opposite directions, which is exactly the merge we warned about on Tuesday with the Japanese twenty-year: a trader who reads “demand improved” as “the price got better” has the sign backwards in both hemispheres this week.
Three of four demand measures beat the average, and the fourth is the one that matters least
Set the allocation out against the six-auction averages the same publisher gives:
- Bid-to-cover 2.73 against 2.62 average and 2.57 previous — 4.2 per cent above the average, 6.2 per cent above last month.
- Primary dealers 10.3 per cent against 11.9 average and 16.9 previous — 1.6 points below average and 6.6 below last month. Dealers taking less is the cleanest single read on real end demand, because dealers take what nobody else wanted.
- Direct bidders 31.6 per cent against 22.9 average and 30.7 previous — 8.7 points above average, the widest deviation on the board.
- Indirect bidders 58.2 per cent against 65.2 average and 52.5 previous — 7.0 points below average, and the one measure that looks weak.
The honest way to read the last two is together, because directs and indirects are the two routes non-dealers use and they add up. Directs up 8.7 and indirects down 7.0 is substitution between channels, not demand leaving the room. The quantity that is not substitution is the dealer share, and it fell 6.6 points from last month to its lowest reading in this comparison set. Note also that the three shares sum to 100.1 per cent, so the publisher is rounding each independently; a tenth of a point here is noise and we have not treated any single share as precise to the decimal.
Clearing 5.6 basis points above the thirty-year is the finding worth your attention
Strip the auction mechanics away and one number is left standing. The twenty-year cleared at 5.469 per cent. The same reader puts the thirty-year at 5.4133 and the ten-year at 5.106 at the time of the result, and the day’s closes elsewhere were 5.145 on the ten-year and 4.823 on the two, a 32.2 basis point two-to-ten slope.
So the twenty-year sold 5.6 basis points cheap to the thirty-year and 32.4 basis points above the ten-year close. The twenty-year point has a long history of being the ugly duckling of the curve, and this is what that looks like in a clearing level rather than in commentary: the government paid more to borrow for twenty years than for thirty. If you trade the yen crosses off the US long end, that is the number to carry, not the bid-to-cover. One reader calls this the highest twenty-year auction yield since 1986, which we report as that reader’s claim — we have not seen the series and have not checked it.
Three publishers, three verdicts, one set of numbers
The characterisations are more dispersed than the data. One publisher grades the auction C. A second summarises it as in-line demand with no directional read for bonds. A third publishes the comparison table with no verdict at all. And the reader we used yesterday for the offering amount described last month’s sale as drawing modestly below average demand — a description the figures support, since last month ran a 2.57 cover below the 2.62 average with dealers at 16.9 per cent.
What none of the three verdicts carries is the thing that makes yesterday’s result interesting, which is the combination: better bidding, a wider-than-average tail, and the cheapest point on the long end. A grade compresses four measures and a price into one letter, and the compression is where the information goes. We would rather you had the 2.73, the 10.3, the 0.9 and the 5.6 than our adjective for them. The settlement is today, 22 October, and the paper is a reopening rather than a new issue, which matters for anyone modelling the supply stack rather than the auction.
What this does not tell you
The issuing agency’s own result was not read. Every figure above is a secondary reading of the Treasury’s announcement by publishers who do read it, and the when-issued level in particular — the input the whole tail argument rests on — comes from a single reader. One number from one publisher is carrying the word “tail” in our headline, and you should discount it accordingly; a second reader on the when-issued would settle it and we did not find one.
We also cannot tell you whether the averages quoted as the comparison set are the same six auctions at both publishers, because only one of them says six. The 1986 claim is single-sourced. We have no bid-side detail, no non-competitive figure, and no view on what the auction did to price afterwards — the ten-year closed up about three basis points on the day, but the embassy headline four and a half hours later muddies any attribution and we are not going to pretend otherwise. And a 0.9 basis point tail on a $13bn reopening is a small number in absolute terms. It is a tail by the definition. It is not a failed auction, and nothing above should be read as one.
