Since the first of September the US three-year note has repriced 51 basis points and the thirty-year bond has repriced 13. That is not one move in the curve; it is four moves in the part of the curve that prices the next two years of Fed policy, and almost nothing in the part that prices everything after. The twenty-year still yields more than the thirty-year — fourth session running, and the gap widened yesterday. If you have been reading the September selloff as an inflation-premium story, the issuer’s own table says otherwise, and it says so in a way that changes where your rate risk actually sits.
The whole move is in the belly
These are par yields from the US Treasury’s own daily curve, 1 September against 23 September, the latest date published. Basis-point changes are ours.
- 1-month 3.85 to 3.99, up 14
- 6-month 4.00 to 4.31, up 31
- 1-year 4.18 to 4.49, up 31
- 2-year 4.39 to 4.85, up 46
- 3-year 4.46 to 4.97, up 51
- 5-year 4.55 to 4.99, up 44
- 7-year 4.66 to 5.05, up 39
- 10-year 4.79 to 5.11, up 32
- 20-year 5.27 to 5.45, up 18
- 30-year 5.27 to 5.40, up 13
The three-year moved 3.9 times as far as the thirty-year. The one-month bill moved 14 basis points, which is to say it moved on funding and not on expectations at all. Between those two poles the curve has grown a hump: the 1-month to 3-year segment steepened 37 basis points, from 61 to 98, while the 3-year to 10-year segment flattened 19, from 33 to 14. The three-year now sits 2 basis points below the five-year and 14 below the ten-year. A month ago it was 33 below the ten.
That shape has one clean reading. The market has added roughly half a percentage point to where it thinks the policy rate goes over the next two to three years, and has added very little to what it thinks about the decade after that. Term premium does not behave this way. A repriced hiking path does.
One print delivered a third of it
Take the same table across a single session, 22 to 23 September, the day the US flash PMIs printed 58.7 on services and 57.0 on manufacturing against consensus figures in the mid-fifties. The 2-year added 14 basis points, the 3-year 16, the 5-year 16, the 7-year 16, the 10-year 15, the 20-year 12, the 30-year 11. The 1-month added 2.
Sixteen of the three-year’s 51 basis points arrived in that one session. Thirty-one percent of a month’s repricing, in one release, in the sector that is supposed to be the most heavily anchored by the Fed’s own guidance.
The operational point is not that the PMI was strong. It is that a survey with no revision discipline and a two-week collection window is currently moving the three-year more than the FOMC’s own statement did. If your risk model treats a flash PMI as a second-tier event and an FOMC as a first-tier one, September has been making that ranking look wrong, and it has been making it look wrong in the tenor where most carry trades actually live.
The twenty-year still pays more than the thirty. That is now four sessions.
This desk flagged the US 20s/30s inversion as an observation on 23 September and said three sessions was not a pattern. Here is the issuer’s table, and the answer is that it survived.
- 1 September: 20-year 5.27, 30-year 5.27 — exactly flat
- 18 September: 5.38 and 5.34 — 4 basis points inverted
- 21 September: 5.33 and 5.29 — 4 inverted
- 22 September: 5.33 and 5.29 — 4 inverted
- 23 September: 5.45 and 5.40 — 5 inverted
Four consecutive published sessions with the twenty-year above the thirty, and the gap one basis point wider on the most recent. It started from dead flat three weeks ago. Note that both long points rose 12 and 11 basis points yesterday while the three-year rose 16 — so the inversion is widening not because the thirty-year is rallying but because the whole long end is being dragged along behind a front end that is moving faster.
Germany, for comparison, is the right way up: Monday’s two reopenings cleared at 3.78 percent on the 2047 and 3.80 on the 2056, a 2 basis point positive slope, from the issuing agency. Two sovereigns, same segment, opposite signs. If you run relative-value across long sovereigns, that is the number to check before London, not after.
The yen’s defence is denominated in the thing that just repriced
USD/JPY traded up to somewhere between 158.27 and 158.50 across the Japanese holiday window — three readers, three figures, and this desk publishes the band rather than picking one. It has since come back. Trading Economics has spot at 157.917, down 0.26 percent on the day; one Japanese desk had 157.80 at 02:55 UTC with the pair having retreated from near 158.50. Both readings are below 158.05, which is where one Japanese desk puts last Friday’s official rate check. Tokyo reopened for the first time in four days and the yen took 35 to 70 pips back without anyone doing anything.
One Japanese desk — single source, and we are labelling it as such — argues additional intervention has become difficult for three reasons: speculative yen short positions have not expanded and the broader book is net yen-long; the move is dollar strength rather than autonomous yen weakness; and selling held US Treasuries to fund intervention would cut directly against the US authorities’ interest in keeping rates down. The first of those is consistent with our own file, which last had the speculative community 103,023 contracts net long yen as of a Tuesday close in early September — stale, and Friday’s COT is the first snapshot that will contain the hike and the rate check.
The third reason is the one that matters here, and it is why this is a rates article and not a yen article. Japan’s intervention capacity is held in Treasuries. Deploying it means selling into a market where the three-year has already moved 51 basis points in three weeks. Every dollar of yen bought that way pushes the differential that caused the problem a little wider. That is not an argument that intervention will not happen. It is an argument that the size at which it works has gone up while the size at which it is politically tolerable has probably gone down, and those two lines move apart every time the belly sells off.
So the question for today is not where USD/JPY goes. It is whether your position sizing is keyed to a level — 158.05, 160, pick one — or to the speed of the three-year. September says the second one is doing the work.
What this does not tell you
The Treasury’s published curve stops at 23 September. We have no 24 September par yields, and secondary readers had the on-the-run 2-year at 4.895 against the par curve’s 4.85 — those are different objects, an interpolated constant-maturity par yield and a specific security’s close, and we are not treating the 4.5 basis point difference as a disagreement. We have no when-issued levels and are asserting no tails.
A caution about our own process, because it nearly cost us this article. The first read of the Treasury page came back reporting the 20-year and 30-year columns as N/A for every September date. Had we published that, we would have told you the issuer had stopped printing its long end. A second read requesting the row verbatim returned 5.45 and 5.40, correctly. The columns were there the whole time. When a primary source appears to have dropped a field, read the row verbatim before you write the sentence.
We have no option strikes for today’s New York cut — the usual expiry board had not published a 24 September edition when we looked — and no order-book information, so nothing here is informed by where resting interest sits. We have no intervention data; Japan’s figures are monthly and the relevant month is not out, and we assert no intervention. The three-reason intervention argument above is one house through one publisher and we have not second-sourced it. And the causal claim that the yen recovered because Tokyo reopened is ours, not a source’s: we looked for a specific driver in the Tokyo session and did not find one.
Related
- FX events calendar — the SNB at 07:30 UTC, German ifo at 08:00, US claims and the current account at 12:30
- Signals
- EA presets
