The support arrived at the front end

The weekend's description of the bond market is that battered Treasuries drew support from falling oil. Treasury's own par yield curve says something narrower and more useful for the week that starts tomorrow. Support arrived on Friday. It reached the two-year, just about reached the ten-year, and did not reach the long end at all.

Friday's curve: the two-year at 4.81 per cent, the five-year 4.98, the ten-year 5.17, the twenty-year 5.54 and the thirty-year 5.49. Against Thursday, that is the two-year down six basis points, the ten-year down one, the twenty-year up one and the thirty-year up two, with both risers closing at their highs. On the week — measured off the previous Friday — the ten-year added sixteen basis points, the twenty-year sixteen and the thirty-year fifteen, against five at the two-year. A market ending the week sixteen basis points higher in its benchmark has not been rescued by anything.

The levels are worth stating precisely because they are being stated loosely. The ten-year traded as high as 5.23 per cent on Friday before closing at 5.17, and one widely syndicated account puts that at levels last seen in July 2006. On the Federal Reserve's own daily ten-year series the last close at or above 5.23 was 14 June 2007, and the last at or above Friday's 5.17 was 6 July 2007. Nineteen years.

At the back it is older than that. The twenty-year's 5.54 per cent is the first close at or above that level since 14 June 2004, when it printed 5.59 — this desk searched Treasury's annual files from 2004 forward and found seven such closes in 2004 and exactly one since, on Friday. The thirty-year's 5.49 is the highest since the bond was reintroduced in February 2006. As this desk set out on Thursday, Treasury's 2004 curve carries no thirty-year column at all, which is why the 2004 comparison belongs to the twenty-year and nowhere else. That inversion survived the week too: the twenty-year is paying five basis points more than the thirty.

Put the two ends together and the shape has turned over. The spread from two years to thirty closed Monday at 53 basis points, the narrowest of 2026; it closed Friday at 68. Fifteen basis points of steepening in four sessions, from a market that spent September flattening. On Monday the argument here was that September was a front-end event about the Federal Reserve. By Wednesday it was not, and by Friday the front end was rallying while the long end was not. Whatever is being priced now is duration, not policy.

The composition confirms it. The ten-year breakeven closed at 2.34 per cent on Friday against 2.33 on Thursday: inflation expectations did not move. Subtract it and the real ten-year yield fell from 2.85 to about 2.83 — the whole of Friday's rally, two basis points of real rate. That real rate is the variable that has repriced every zero-coupon asset this year, and it is still at an eighteen-year high.

Now the oil, which did fall. EIA's daily Brent spot series has the barrel at $130.80 on 16 September and $114.89 on 22 September — down 12 per cent in five sessions — while Cushing WTI went from $107.02 on the 15th to $96.41 on the 22nd. Both series were released on 23 September and stop there; the next publication is Wednesday. Friday's futures settlements, which this desk could not read at source because CME blocks automated requests, are reported at $92.41 for November WTI and $104.32 for November Brent. That second number deserves attention: the Brent front month settled Friday above the $103.08 it reached on Wednesday the 23rd, the day the ten-year rose fifteen basis points. Spot fell. The contract everybody quotes did not.

Nor has the physical market caught up. EIA's weekly on-highway diesel average was $6.529 a gallon on 21 September against $3.749 a year earlier, the highest in the series. A twelve per cent fall in the Brent print from a $130 peak is not disinflation arriving; it is a spike that has stopped widening, in a week when equities rose anyway, the S&P 500 by 1.2 per cent.

Which brings the argument to Wednesday, when it gets settled twice. At half past eight New York time the Bureau of Economic Analysis publishes Personal Income and Outlays for August — the date is on its own release schedule — with consensus at 3.7 per cent headline and 3.4 per cent core, an acceleration from 3.3 in July. The Federal Reserve's calendar has the next meeting on 27-28 October, and the market prices better than even odds of an increase. A core print that confirms the acceleration puts the front end back where it was on Thursday.

After the close on the same day, Micron reports its fourth quarter — consensus around $51bn to $52bn against guidance of $50.4bn to $51bn, next quarter's consensus already at $56.9bn. As this desk argued on Thursday, the headline is the least informative part of that release: what matters is the deposit book, whether the ceiling on the largest take-or-pay agreements has moved, and whether bit shipments grew at all. Micron's investor pages refused automated requests here, so the date rests on two secondary reports that agree. Payrolls follow on Friday.

Underneath all of it sits the arithmetic this desk set out on Thursday: an AI build-out whose flagship project financing priced at 6.58 per cent, against a thirty-year Treasury that closed Friday at 5.49 and rising. The cost of the long end is the cost of the build-out. It went up again on the day the bonds were said to have found support.