A quarter-point went in and the ten-year came out at exactly five per cent — the long end is…

## The lender left in July

*A quarter-point went in and the ten-year came out at exactly five per cent — the long end is clearing without the buyer that used to set it*

Three central banks moved inside five sessions: the Federal Reserve to 4.00% on 16 September, the Bank of England holding at 3.75% on the 17th, the Bank of Japan to 1.25% on the 18th. Friday settled with the S&P 500 at 7,650.50 and the Nasdaq at 26,522.55 — +0.85% and +2.08% against their closes the session before the Fed decided — and the VIX at 14.81, its lowest since 4 September. The ten-year Treasury settled at 5.00%: the same yield, to the basis point, that it carried before the increase. The five-year added 3 basis points and the thirty-year gave up 3. A tightening went in and nothing came out of the long end at all.

The long end has stopped trading policy and started trading its buyer. On 16 September the July capital-flow data showed foreign investors were net SELLERS of US long-term securities — minus $27.9bn against a consensus of plus $146.3bn and plus $174.4bn the month before. That is a sign flip, not a miss, and it is the sharpest claim in this issue: a ten-year that does not move one basis point across a rate rise is not a market arguing about inflation, it is a market whose marginal price-setter has changed hands. It also marks the call this desk made on 7 September, that official money was relocating out of US paper; July is the first hard print of it rather than an inference from the gold bid. We part from the benign reading here. A long end that absorbs a hike without flinching is usually credibility, and the thirty-year falling 3 basis points on the week is exactly what credibility looks like. But credibility bought at five per cent from price-insensitive domestic hands is a different security from the same yield cleared by a standing foreign reserve bid — it holds until the price moves, and then there is no one underneath it. Conviction is MEDIUM: one month of a series that swings this hard is a signal, not a trend.

### What matters

**July's capital-flow data printed on the other side of zero.** Net long-term Treasury International Capital flows for July, released on 16 September, came in at minus $27.9bn against a consensus of plus $146.3bn — a swing of roughly $174bn against expectations — with the prior month at plus $174.4bn. Foreign investors did not merely buy less US long-term paper than forecast; on the net measure they sold it. A consensus miss tells you forecasters were wrong. A sign flip tells you the behaviour changed. The series is volatile and one month proves little on its own, which is why this belongs in the argument rather than at the end of it — but it is the only release from the week that explains a ten-year which declined to react to a rate rise. Here is the observation that would look different if this were an inflation story instead of a demand story. An inflation repricing is global and steepens curves: it shows up in Bunds and gilts too, and it lifts the long end hardest. Over the same five settled sessions the German ten-year fell 1 basis point and the Swedish ten-year fell 2, while the US ten-year rose 2. And the American curve did the opposite of steepening — the thirty-year gave up 3 basis points against its pre-decision close while the five-year added 3. Prices are not what is being repriced here.

**A hike delivered, and the currency fell anyway.** The Bank of Japan raised to 1.25% on 18 September in a split decision taken under acknowledged US pressure. The yen finished the session 0.45% weaker at 156.85 and 2.21% weaker across the week. The Nikkei rose 1.38% on the day; Europe fell, with the Euro Stoxx 50 -1.07%, the DAX -1.38% and the Swiss market -1.15%. This desk named the Bank of Japan decision as a catalyst on 12 September and wrote that a hike delivered into an already-firming yen would raise the cost of the cheapest money in the system. The hike arrived on schedule and the second half did not: the yen weakened instead. We were right about the policy and wrong about the price, and the gap between those is the informative part. A currency that will not rise on its own central bank tightening is telling you the exchange rate is being set somewhere other than the policy differential — by flow, not by carry arithmetic. That is the same conclusion the capital-flow data reaches from the other end, and it is why the yen sits in this issue's Signal rather than in its own paragraph.

**Conditions eased into the tightening, and the equity market sorted by balance sheet.** Two sessions after the first Federal Reserve increase since 2023, the S&P 500 was +0.85% against its pre-decision close, the Nasdaq +2.08% and the VIX -13.90%. Underneath, Friday sorted the index by who funds what: the semiconductor index rose 2.69% and Micron 3.92%, while Oracle fell 1.99% — 1.78% on the week — Meta 2.43% and utilities 3.04% across the week. Small caps fell 0.50%. Easier conditions after a tightening are the standing argument for the next tightening, which this desk made on 18 September and which the front end has since confirmed by moving three basis points above its pre-decision yield. The dispersion adds the part that was missing: the market is not de-rating duration evenly, it is charging the companies that must borrow to build and still paying the ones that sell them the equipment. The honest caveat is that Friday was a quarterly options expiry, when single-session dispersion between large stocks is partly mechanical. The weekly figures carry the claim better than the daily ones: Oracle down on the week, utilities down more, chipmakers up.

### What we see

The engine's positioning series on the yen — managed-money and other speculative futures accounts — read across the two reports either side of the Bank of Japan's decision. The speculative net long moved from +10.8 thousand contracts in the report dated 2026-09-11 to +120.4 thousand dated 2026-09-18 — a swing of +109.6 thousand, and the largest long of the past three months from -163.4 thousand on 2026-07-31. Over the window those positions were measured, 2026-09-08 to 2026-09-15, the yen WEAKENED 1.20% against the dollar. It weakened again to 156.85 on 2026-09-18, the session the Bank of Japan delivered its increase to 1.25% — 2.21% weaker across the five settled sessions.

The yen is the funding leg of most things that are levered. A crowd that positions for dearer Japanese money, and is then proved right by the central bank within the week, should be paid — and was not: the currency fell through the build and again after the decision. The inference we draw, and it is an inference rather than an observation, is that money stepping back from dollar government paper is not being repatriated to Japanese government paper either; it is standing aside from sovereign duration generally. That is the same withdrawal the July capital-flow release measured, visible on the funding side a month earlier and at weekly rather than monthly frequency. The reading has a stated limit: it covers one reporting category of futures positions, not a census of holders, and it stops on 2026-09-15 — three sessions before the decision it is being read against.

### The wider frame

Step back and the week describes an institution losing an instrument rather than an economy losing its balance. A central bank sets the price of overnight money; it cannot conjure a buyer for thirty-year paper. For two decades the American long end was priced as though the second followed from the first, because a reliable official bid made the assumption free. What this year keeps testing is whether policy credibility and funding demand are the same asset. They are not, and the distance between them is where the term premium now lives — which is why the dollar can hold at 100.21, the front end can reprice, and the long bond can sit through a tightening without an opinion. The view is wrong if the August release, due in mid-October, reverses July and foreign buying resumes above the consensus it missed; a single negative month in a series this volatile is a warning, not a verdict. It is also wrong, in a different direction, if the coming week's data pushes the ten-year decisively off five per cent — a long end that starts moving again with policy is a long end policy still governs.

### Method

Every figure in this issue is taken from the settled session of 2026-09-18 unless it carries its own date; the physical-flow series settle a day later, on 2026-09-19, and are labelled where used. Two dating notes apply this morning. The WTI daily series rolled from the October to the November contract on 2026-09-18, so its headline change compares two different contracts; the like-for-like moves are stated above and were verified against the individual contract months. And the weekend's strikes on Riyadh fall after the last settled session, so no price quoted here contains them — they are carried in the Radar as a forward risk, which is where an unpriced event belongs.