The Institute for Supply Management’s services index for September printed 54.9 at 14:00 UTC yesterday, half a point below August’s 55.4. The ten-year Treasury closed at a fifty-two-week high. The two-year closed two tenths of a basis point lower. Whatever was repriced yesterday, it was not the Federal Reserve’s next move. And the sub-index that the long end appears to have traded is not one of the four that the headline number is made of — a fact you can check in two lines of arithmetic, which we did.
The headline is the straight average of four numbers, and we verified it
The services composite is the equally weighted average of four sub-indices: business activity, new orders, employment and supplier deliveries. From the Institute’s own release, September’s four were 56.5, 59.8, 50.1 and 53.2. Those sum to 219.6 and divide to 54.9 — the headline, exactly. August’s were 61.7, 60.9, 47.8 and 51.3, summing to 221.7 and dividing to 55.425, which rounds to the published 55.4. The construction checks out to the tenth in both months, so what follows is not inference about the index, it is the index.
Now look at the four changes: business activity −5.2, new orders −1.1, employment +2.3, supplier deliveries +1.9. They sum to −2.1 and divide to −0.525, which is the half-point fall you read in the headline. Business activity fell 10.4 times as far as the index did. The headline is small because a collapse in one component was netted against rises in two others, not because the month was quiet.
For scale, this desk published yesterday morning that the mean absolute month-on-month change in this series across the eight published months of 2026 is 1.09 points. The half-point fall is 0.46 of that — genuinely a below-average month at the headline. The components are not having a below-average month at all.
Two of the three things holding the index up are not good news
This is where reading the composite as a sentiment score goes wrong. Of the 4.2 points of upward movement that offset the business-activity fall, 1.9 came from supplier deliveries and 2.3 from employment.
A rising supplier-deliveries reading means deliveries are getting slower. In the index’s own construction that counts as expansion, because slow deliveries historically signal demand outrunning supply — but it is not a number you would volunteer as evidence that the services economy had a good month, and it adds to the composite either way.
Employment went from 47.8 to 50.1. That is a 2.3-point improvement and it is also a reading one tenth of a point above the line between growth and contraction, four days after a payrolls report that added 29,000 jobs and missed every forecast this desk collected by at least 31,000. Describing 50.1 as the component that held the index up is accurate and nearly meaningless.
So the honest summary of a 54.9 against a 55.4 is: activity fell hard, orders softened, hiring stopped shrinking by a hair, and deliveries slowed. The average of those four is a non-event. None of them individually is. The Institute itself notes this is the 27th consecutive month in expansion territory, which is true and is the least informative true thing available.
The number the bond market traded is not in the index
Two sub-indices moved hard yesterday and neither is part of the composite.
Prices printed 74.0, up 1.4 points, twenty-four points above the neutral line. That is an input-cost reading, and it is not one of the four averaged into the headline. New export orders printed 46.9, down 9.4 points, into outright contraction — 18.8 times the headline’s move, and 47 times the width of the two-tenth disagreement four vendors spent yesterday morning publishing about where the headline would land. Imports fell 3.4 to 52.9. Backlogs rose 1.0 to 56.6 and inventories 1.1 to 57.8.
One publisher, writing at 22:31 UTC, attributes the long-end move to rising input prices and rising inflation expectations, alongside France’s fiscal position. We cannot verify a causal chain from one reading and we are not claiming one. What we can say is the shape: the component that speaks to inflation moved up and is excluded from the headline; the components inside the headline cancelled; and the part of the curve that prices inflation rather than the policy rate is where the day’s repricing landed. A filter that reads the headline and stops has no record that anything happened.
The curve steepened. The Fed tenor did nothing.
Two publishers mark the ten-year close differently and we are publishing the band rather than picking: 5.315 per cent, up 3.8 basis points, at one; 5.307 per cent, up more than 3, at the other. An 0.8 basis-point gap between two readings of the same close. Both call it a multi-year or fifty-two-week high; one frames it as the highest since 2002, which we are reporting as that publisher’s characterisation rather than as a verified series fact, because the two previous times this desk went looking for the comparison year it found the question was about two different days.
The two-year closed at 4.823 per cent, down 0.2 of a basis point. That is the tenor whose whole job is to price the next few Federal Reserve meetings, and on a day with an inflation sub-index at 74.0 it did not move.
The arithmetic, ours: two-year to ten-year is 49.2 basis points on the wider reading and 48.4 on the narrower, against the 44.6 this desk published for Friday’s close — roughly 4.6 basis points of steepening in a session. Term premium, not policy expectation. Consistent with that, the probability of an October hold reads 76 per cent on one feed against the 77 we published yesterday morning: a one-point drift on a series we have already shown moving 8.69 points intraday, which is to say no change at all.
Equities took the opposite view without apparent difficulty. The Nasdaq composite closed at 27,477.31, up 286.45 or 1.05 per cent, at a record; the S&P 500 at 7,773.99, up 51.27 or 0.66 per cent; the Dow at 51,267.90, up 90.94 or 0.18 per cent. Gold settled at 4,189.50 dollars an ounce, up 0.65 per cent, and WTI at 90.68 dollars, down 0.47 per cent. A record equity close and a fifty-two-week high in the ten-year on the same afternoon is not a contradiction you need to resolve — but if you run anything correlated across indices and the long end, you were carrying two positions that both got what they wanted yesterday, and that is usually temporary.
What this does not tell you
We have not established that the prices sub-index caused the long-end move. One publisher asserts input costs and inflation expectations; the same publisher names French fiscal stress in the same sentence, and the ten-year has been grinding higher for several sessions on a story this desk has been tracking as a European sovereign spread. The honest claim is that the headline cannot account for the move and that the excluded components can, not that they did.
We did not verify the sub-index values against a second source. They come from the Institute’s own release, which is the issuing body, and we would rather have one primary reading than two secondary ones — but a single transcription error on our part or the wire’s would not have been caught, except insofar as the four components summing exactly to the published headline in both months is itself a strong internal check.
The two-year close is single-sourced. Only one of the two publishers we read carries it, so the 4.6 basis points of steepening rests on one reading of one leg.
Whether this print vindicates or embarrasses the vendor forecasts that preceded it is a separate question with a separate answer, and it belongs to the desk that published the forecast arithmetic rather than to this one. We have reported what printed.
And none of this is a view on direction. A steepening on an inflation sub-index is a statement about which part of the curve is carrying risk, not about whether yields go up tomorrow.
