On Friday the US economy was reported to have added 29,000 jobs in September, 61,000 below the median of 27 analysts. The two-year Treasury yield — the tenor whose entire job is to price the Federal Reserve — closed the day 3.3 basis points higher, at 4.825 per cent. The ten-year closed 3.0 basis points higher at 5.271 per cent. Both had been sharply lower within minutes of the release. That is the whole story of the session, and it is a reaction-function story rather than a data story. There is a second one underneath it, and it is ours: on Saturday morning this desk published that three feeds were giving three different October Fed probabilities and called the spread the widest and most decision-changing it had recorded. It was not three feeds. It was one feed, read at five times of day.
The policy tenor closed higher on the weakest payrolls print of the cycle
Take the two-year on its own, because it is cleaner than the ten-year. It entered Friday around 4.78 per cent. On the release it fell to 4.6913 per cent — 8.9 basis points lower, a perfectly orthodox response to a 61,000-job miss. It then recovered to approach 4.85 per cent and settled at 4.825, up 3.3 basis points on the day. From the low to the close is 13.4 basis points in one afternoon.
The ten-year did the same shape with a wider amplitude: a low of 5.1528 per cent, a rebound to 5.296 in the New York afternoon, a close of 5.271, up 3.0 basis points. The thirty-year closed up 0.7 basis points at 5.620. The two-to-ten spread finished at 44.6 basis points.
If you run a system that trades the dollar off the front end of the US curve, Friday was a day when the input went one way for twenty minutes and the other way for five hours, and the official daily change — the number your end-of-day feed stores — has the opposite sign to the headline. That is not a rare event, but it is a specific hazard: a strategy conditioned on the daily change would have read Friday as mild hawkish repricing. A strategy conditioned on the release-window move would have read it as a dovish shock. Both readings are faithful to the data. They disagree because they are sampling different windows, and nothing in the print tells you which window is the right one.
One named attempt at the mechanism, and we are reporting it as one bank’s view rather than as established: TD Securities argued on Monday morning that the September weakness came mainly from seasonal adjustment rather than from deterioration in the labour market, pointing to rising participation and a rising employment-to-population ratio. If that is how the market read it, the recovery in yields is not a puzzle at all. We have one source for that interpretation and we are not going to pretend it is consensus.
Our three-feed probability spread was one feed on a clock. We retract it.
On Saturday this desk published the following as a finding: three feeds now give three different October Fed probabilities — 77 per cent priced for a hold, “the 20 per cent range”, and a fall from 28 per cent to below 13 per cent — and we said that the question “did payrolls reprice the Fed” therefore had two opposite answers in the same afternoon. We refused to adjudicate between them on the grounds that we could not.
Every one of those readings is CME FedWatch. So are the two more we found this morning. Laid out with their timestamps, the series reads: about 28 per cent before the data; 12.9 per cent at the trough immediately after it; 17 per cent later that day; 21.59 per cent at 18:06; and the 20 per cent range, and 100 minus 77, at the close. The same number, five times, as it moved.
Note what that does to the arithmetic. The intraday round trip — 12.9 to 21.59, 8.69 percentage points — is 1.36 times the net repricing on the day, which was 28 to 21.59, or 6.41 points. The recovery was bigger than the move. One reader published the trough and one published the close, and we treated their disagreement as evidence about vendors when it was evidence about the hour.
We are not going to dress this up. The desk has a standing rule that says to read at least two sources and name the feed in text whenever a probability is load-bearing. We followed it, and it did not save us, because the rule conflates two different things. Naming the publisher is not naming the feed, and neither is worth anything without the time. A probability is a price. It has a timestamp or it is not a datum. The rule becomes: record the hour alongside the figure, and before calling two readings a disagreement, check whether they could be the same series sampled twice.
What survives is smaller and duller than what we published. October hike pricing fell about six points on the print, from the high twenties to the low twenties, and the amount of tightening priced into the whole of the rest of 2026 fell 3.3 basis points, from 25.5 to 22.2. December is priced at a hold with no meaningful probability of anything else. That is a market that trimmed its expectations and did not change its mind.
The dollar made its new high in Tokyo on a French bond story
Dollar-yen traded up to 158.17 in the Tokyo session, taking out Friday’s New York high by about 26 pips and closing to within 34 pips of the 200-day moving average at 158.51. Nothing Japanese did it. The move was broad dollar strength driven from euro-dollar, which fell to 1.1167 from Friday’s 1.1255 close, with the dollar index around 102.50.
The cause sits in French government bonds. The France-Germany ten-year spread reached 130 basis points on 1 October, described by one reader as the widest since 2012, after a 2027 French budget aimed at holding the deficit to 5 per cent of GDP failed to settle the market. Euro-dollar fell to its lowest since May 2025 on it.
The point for a system trader is narrow and worth the discomfort: if you are long dollar-yen this morning, the thing that is paying you is a sovereign credit story in a third currency. Your position is sized against yen volatility and your profit and loss is being driven by an instrument you do not monitor. That is correlated exposure arriving from a direction your risk model does not have a column for — and it is a much better reason to cut size than any yen-specific argument available today.
There is also an inflation print inside the London morning that bears on it. Euro-area producer prices for August are due at 09:00, and the consensus is 7.9 per cent year on year against a prior of 5.8 — a 2.1-point acceleration. We note in passing that our own Saturday schedule carried the 5.8 in a way that reads as today’s expected figure. It is the prior.
Three coupons this week, and they all settle the day after CPI
The US Treasury auctions a three-year note on Tuesday, a ten-year reopening on Wednesday and a thirty-year reopening on Thursday, all announced on 1 October. All three settle on Thursday 15 October. So do the six-week, thirteen-week and twenty-six-week bills auctioned on 13 October: six securities settling on one day. A seventeen-week bill on Wednesday and four-week and eight-week bills on Thursday settle earlier, on 13 October.
US September CPI is on 14 October. The coupon settlement lands the day after it. We are not claiming that is causal — the settlement date was fixed when the quarter’s schedule was published and has nothing to do with the inflation calendar. We are claiming it matters operationally, because the single largest cash movement of the month in Treasuries falls into the day after the month’s largest scheduled repricing event, and a dealer balance sheet that has to absorb three coupons is a worse shock absorber on 15 October than on any other day of the month.
On the Japanese side the Ministry of Finance auctions roughly 2.6 trillion yen of ten-year JGBs on Tuesday and roughly 600 billion yen of thirty-year debt on Thursday. Both figures are approximate as published and come from a single reader; treat them as the right order of magnitude rather than exact. The Finance Minister and the Bank of Japan governor both speak at the National Securities Convention on Tuesday, which is the next scheduled opportunity for anyone to say something about the 3.0 per cent Tokyo core-core print that the market has conspicuously not repriced.
What this does not tell you
We do not know why yields recovered. The seasonal-adjustment reading is one bank’s argument, named as such, and we have not found a second house making it. The alternative — that nothing was learned and the morning move was simply over-sold — is equally consistent with the price action and cannot be distinguished from it.
Our Friday close for the ten-year, published on Saturday as 5.296 per cent, was the New York afternoon rebound level and not the close. The close was 5.271, from the same publisher’s own bond summary — a page we did not check before publishing. That is 2.5 basis points, which changes nothing in the argument and is still a figure we got wrong, from a source we already had. The low we published as 5.1549 reads as 5.1528 on that page, 0.21 of a basis point apart; we have no basis for preferring either and both are the same publisher.
We also have no post-print October figure for the Bank of Japan. That is now four consecutive slots without one, and the 3.0 per cent core-core argument remains an argument this desk keeps making about a repricing that has not visibly happened.
And the retraction above is not a claim that CME FedWatch is correct. It is a single feed, derived from fed funds futures, and we have no second source for the October probability at all. Having spent Saturday wrongly describing a vendor disagreement, we are now in the position of reporting a figure with no corroboration whatsoever. The honest statement of our October knowledge is: one feed says low twenties, we cannot check it, and the number moves nine points inside an afternoon.
Related
- FX events calendar — the auction and release times above, in UTC
- Signals — how we treat a release window as distinct from a daily change
- EA presets — news-window and session filters
