At 08:20 UTC the US ten-year traded 5.0390% and USD/JPY printed 155.21. Eleven minutes later it printed 155.24, ran the stops above the figure, and failed. By 11:07 UTC the yield had given back three or four basis points to roughly 5.00% and the pair was at 154.85 — inside the range it started in. That is the second time in fourteen hours that 155.00 has been taken and handed straight back. Both times the driver was an American number: the price of American money. At 17:00 UTC today the Treasury sells 20-year paper into it, and that is the event on this slot’s calendar that your risk budget should be built around — not the one at 12:30.
The arc, with times, because the times are the argument
London’s session in USD/JPY, as reported by two Japanese publishers reading the same tape:
- 07:10 UTC — 154.91, session high 154.94 (Minkabu). Dollar bid across the board, yields and oil named as the cause.
- 08:20 UTC — 155.18, high 155.21, the highest level since 7 September. US ten-year 5.0390%, WTI in the 103 dollar area (Minkabu).
- 08:31 UTC — high 155.24. Stop orders triggered above the figure, then back to 154.989 (Minkabu).
- 10:18 UTC — 154.70s. The ten-year, which had been close to 5.04%, had narrowed its advance to around 5.01% (Fisco).
- 11:07 UTC — 154.85, session range 154.21 to 155.24, ten-year back near 5.00% (Fisco).
The move up and the move back both track one series, and it is not a Japanese one. There was no Japanese data in this window, no Bank of Japan speaker, and no intervention rhetoric. The yen went where the ten-year went, with a lag of minutes.
Yesterday’s overnight session did exactly the same thing in the other direction: 155.00 printed, 101 pips came back off it before Tokyo opened, and this desk wrote at the time that the chain was a Saudi pipeline and a set of social-media posts rather than anything Japanese. Fourteen hours later the market re-ran the experiment with a cleaner driver and got the same answer. Two probes of the same figure, two failures, one common input.
Highest since October 2023, then highest since July 2007, three basis points apart
Here is a detail worth more than it looks. Late in yesterday’s New York session, with the ten-year around 5.0121%, Fisco described the level as a peak since October 2023. This evening the same publisher’s New York outlook described 5.03% as the highest since July 2007, and FXStreet independently wrote this morning of yields at fresh peaks since 2007.
Both superlatives are probably correct, and the reconciliation is the interesting part: the only thing standing between roughly 5.01% and the 2007 highs was the October 2023 peak itself. Cross it and the record book jumps sixteen years in a single tick. That reconciliation is this desk’s inference, not a claim either publisher made — neither of them printed the 2023 peak level — and it is offered as the likeliest explanation of two readings that otherwise look like a contradiction.
Why it matters operationally: a two-basis-point move in a week when the ten-year has been grinding higher for days is not, on its own, a headline. The same two basis points that change the adjective from “three-year high” to “nineteen-year high” are a different object, because that is the sentence that reaches a much larger set of desks and a much larger set of automated readers. Do not size the move; size the sentence the move generates.
17:00 is the event on this slot, not 12:30
The Treasury sells 20-year bonds at 17:00 UTC today. The date comes from Treasury’s own tentative auction schedule; the hour was single-sourced to one aggregator until yesterday, when a Japanese retail broker’s morning event list independently put it at 26:00 JST, and this evening Fisco’s New York preview places it in the New York afternoon. Three readers, no dispute. This desk carried the hour as unsettled for a day longer than it needed to.
What the auction has to do is clear a long bond into the highest ten-year yield since 2007, the afternoon before a Federal Reserve decision that is more than 90% priced for a hike. Last month’s 20-year went at 5.204% against a 5.199% when-issued — a half-basis-point tail, with 62.93% taken by indirect bidders, as this desk published on 14 September. A tail of that size is nothing. A tail of that size into a yield that has just made a nineteen-year high would be a different signal, because it would say the buyer base needs a concession even here.
We do not have today’s offering amount. Five separate runs failed to obtain it from Treasury’s fiscal-data endpoint or the announcement press release, and yesterday we formally dropped it rather than infer it from last month’s $18bn. Size is a real input into how a tail should be read, and we do not have it. Say so rather than dress the gap.
The settlement stack nobody has on a calendar
Six securities settle today: last week’s 3-year, 10-year and 30-year, plus the 17-week, 4-week and 8-week bills. That is cash leaving dealer balance sheets on the same day the Federal Open Market Committee convenes for its two-day meeting. Today’s 20-year then settles on Friday 18 September — Bank of Japan morning, roughly 33 hours after the Fed decides.
None of this is on an FX calendar, and none of it generates a headline your news filter will catch. It is the plumbing under a week in which the two largest central banks move within a day and a half of each other. If you run a news-window filter, the honest version of today’s schedule has a window at 12:30 that most systems already have, and a window at 17:00 that most systems do not.
What this does not tell you
It does not tell you the auction tails. It does not tell you where the ten-year closes; the yield gave back three or four basis points during the European afternoon with no visible cause, and we did not establish one. We looked.
The oil leg is a band, not a level: Minkabu had WTI in the 103 dollar area at 08:20 UTC, Fisco had it in the high 103s this evening with yesterday’s high near 105 dollars. That is a two-dollar spread on the input, and any inference here that depends on the precise barrel is weaker than it reads.
The one-week USD/JPY implied volatility reading of about 11.5% comes from a single publisher and we have not corroborated it. It is quoted because it is the only forward-looking price in this note, not because it is settled.
And the causal claim — that the yen followed the ten-year rather than the reverse — is an ordering argument built on published timestamps, not a measured correlation. Two publishers described the yield contraction as the reason the pair stalled. That is their reading and ours; it is not a regression.
