The Term Premium Is Back

The disinflation arrived and the long end ignored it. The July CPI report, published on 12 August, put headline inflation at 3.4% on the year against 3.5% in June, and core at 2.5% against 2.6%. On the month headline added 0.1% and core 0.2%. That is a cooling print by any reading. The long end did not trade like one. Four multi-year highs bracket the summer, two set in the week after that CPI and two already on the board before it.

The United States set two of them. The 10-year closed at 4.75% on 31 July, twelve days before the CPI, the high of 2026 and the first close at or above that level since 14 January 2025, roughly eighteen months earlier. The 30-year peaked at 5.31% on 17 August, which Bloomberg dated as the highest since 2007, and CNBC marked an intraday 5.33% the next morning as a 19-year high. The Treasury close on 18 August was 5.28%.

Japan produced the sharpest of the four. On 18 August the 30-year JGB fixed at 4.096%, an all-time high in a Ministry of Finance series that begins on 2 September 1999, and the 40-year at 4.103%, also a record. The 10-year touched 2.934% the same day, its highest since 27 September 1996. The fourth high is the oldest of the set: Bloomberg dated the conventional 30-year gilt at its highest since 1998 on 5 May. The Bank of England nominal spot curve, a zero-coupon construction rather than a conventional gilt redemption yield, put the 30-year at 5.967% on 18 August and 5.864% on 25 August.

Year to date the moves are large and uneven. The US 10-year is 46bp higher, from 4.18% on 31 December 2025 to 4.64% on 25 August. The Bund is 38bp higher, 2.89% on 30 December to 3.27% on 25 August and 3.22% on 26 August, which leaves it 102bp over a 2.25% deposit rate. The 10-year gilt is 58bp higher, 4.4664% to 5.0513% on the 24 August fixing. The JGB 10-year is 83bp higher, 2.066% on 30 December to 2.897%, or 190bp over a policy rate of around 1.0%. Energy is the common input: CPI energy prices ran 14.7% above a year earlier in July. What is not common is the policy response, and that is where the decomposition has to start.

A nominal 10-year yield is two things added together. The first is the average overnight rate the market expects across the next ten years. The second is the extra compensation demanded for holding the bond through that decade instead of rolling bills, which is the term premium. The Adrian, Crump and Moench model published by the New York Fed splits the curve into exactly those two legs, and that split is the argument of this piece.

Year to date, the split does not flatter the title. The ACM fitted 10-year rose 52bp, from 4.245% on 31 December 2025 to 4.766% on 24 August. The expected-rate leg did 49bp of that work, moving from 3.491% to 3.978%. The term premium did 3bp, from 0.754% to 0.788%. On a full-year view 2026 has been a repricing of the policy path rather than of duration risk: a Fed that stopped cutting, an energy shock that lifted short-rate expectations across the G4, and almost nothing added for term.

The summer is a different data set. On 30 June the ACM term premium sat at 0.512%. By 17 August it was 0.895%, a rise of 38bp in seven weeks and the high of 2026. The expected-rate leg went the other way across the same seven weeks, from 3.965% to 3.900%, roughly 6bp lower. The fitted 10-year rose 32bp over that window, so the term premium accounted for the whole of the move and then some, against an expectations leg that subtracted from it. The cash 10-year tracked the same path, from a 4.44% close on 30 June to 4.72% on 17 August. That is the sentence the title rests on, and it is a seven-week sentence, not a year-to-date one.

The premium has since come off the high. It printed 0.804% on 13 August, 0.895% on 17 August and 0.788% on 24 August, the last value published, which is 11bp below the peak and 28bp above where the summer started. A move that hands back a quarter of itself inside a week is not a regime. It is a market that has begun charging for something it was not charging for in June, and has not settled on how much.

The reason to charge for duration begins with how much of it is coming. The Congressional Budget Office February 2026 baseline puts the fiscal 2026 federal deficit at $1.9tn, 5.8% of GDP and the third largest in American history, on outlays of $7.4tn, or 23.3% of GDP, against receipts of 17.5%. Gross federal debt is $38.6tn, 123.1% of GDP, on a path to $63.7tn and 136.4% by 2036.

The realisation is running ahead of the projection. Treasury's Monthly Statement puts the fiscal 2026 deficit at $1,798.8bn after ten months, on receipts of $4,485.4bn and outlays of $6,284.2bn. The full fiscal 2025 deficit was $1,775.4bn. Ten months of 2026 have already exceeded twelve months of 2025, and July alone contributed $432.3bn.

Net interest is the line that compounds. CBO has it at $1.0tn in fiscal 2026, 3.3% of GDP and the third largest item in the federal budget behind Social Security and Medicare, exceeding defence spending in every year of the outlook and running at almost double defence by 2036. It absorbs 19% of tax receipts this year against 9% in 2021. The realised figures agree with the projection: net interest was $931.4bn through July, and gross interest on the debt reached $1,169.6bn against $1,012.9bn in the same ten months a year earlier, $156.7bn more and 15.5% higher. In fiscal 2018 net interest was $325bn. In fiscal 2025 it was $970bn.

The issuance follows mechanically. On 3 August Treasury raised its July to September net marketable borrowing estimate to $739bn, $68bn above the May figure, or $87bn above it excluding a higher assumed opening cash balance, with an end-September cash target of $950bn. October to December is estimated at $628bn against an $850bn year-end cash balance. The second quarter, for scale, needed $190bn and closed with $919bn of cash. Two days later the August refunding was set at $125bn: $58bn of 3-year notes, $42bn of 10-year notes and $25bn of 30-year bonds, refinancing roughly $96.3bn of maturing paper and raising about $28.7bn of new cash. The guidance attached to it was flat: Treasury "anticipates maintaining nominal coupon and FRN auction sizes for at least the next several quarters." Supply is not the variable here. It is the constant everything else has to price around.

Someone has to absorb that supply. The Treasury International Capital data for June 2026, published on 17 August, puts total foreign holdings of US Treasuries at $9,299.0bn, down $72.1bn from $9,371.1bn in May. Foreign official holders, meaning central banks and sovereign funds, account for $3,778.1bn of that total, or 40.6%.

The two largest holders are both smaller than a year ago. Japan holds $1,116.7bn against $1,154.8bn in June 2025, a reduction of $38.1bn. China holds $633.4bn against $731.4bn, a reduction of $98.0bn, or 13.4% of the position in twelve months. The United Kingdom, at $939.9bn, is the second largest holder.

The chart carries the shape of it. The aggregate peaked at $9.489tn in February 2026 and has fallen in three of the four months since, ending June $29bn above where it closed December. Japan peaked at $1.239tn in that same February and has shed $122bn since, $26bn of it in June alone. China has not been above $700bn in any month of the window bar the first, and set the low of the thirteen months in June. Strip the two Asian blocks out and the picture inverts: they hold $136bn less than they did in June 2025, while everyone else added $341bn. The aggregate grew $205bn over the year only because the rest of the world covered what Tokyo and Beijing sold.

The Japanese leg has a mechanical explanation that the weekly briefs have tracked since early August. A 30-year JGB at a record 4.096% and a 10-year at 2.897% change the arithmetic for a Japanese life insurer or bank that spent a decade reaching into foreign duration for yield unavailable at home. Domestic paper now clears the hurdle with no currency risk and no hedge cost attached. Every basis point of JGB yield weakens the case for holding a Treasury or a Bund unhedged, and that withdrawal never arrives as a headline. It arrives as a monthly TIC line that stops growing, which is precisely what June delivered.

None of this is a buyers strike. Net TIC inflows were $133.5bn in June and net foreign purchases of long-term securities $207.1bn, so demand for US assets in aggregate was present. The composition is what changed: less official, less Japanese, less Chinese, more of everyone else. A buyer base rotating from price-insensitive reserve managers toward price-sensitive private funds is a buyer base that asks for a term premium. The next reading, covering July, lands on 16 September.

Every one of these central banks is working the same shock. The Strait of Hormuz remains closed to normal traffic behind a US naval blockade. On 10 August Iran's foreign ministry spokesman Esmail Baghaei said that "as long as the U.S. naval blockade continues, the necessary conditions for the reopening of the Strait of Hormuz do not exist," with no agreement in place to resume shipping. Brent settled at $87.72 that Monday, up about 5%. On Monday 24 August, as Washington launched what it called Operation Economic Outcast, the heaviest sanctions campaign ever mounted against Iran, Brent settled at $92.06, down 2.5% on the day but 4.9% above the level of a fortnight earlier. Energy prices in the July CPI ran 14.7% above a year earlier.

The Federal Reserve held its target range at 3.50% to 3.75% on 29 July, with Beth Hammack, Neel Kashkari and Lorie Logan dissenting for a quarter-point increase. Three dissents on the hawkish side of a hold is not a committee preparing to ease. At 4.64% the 10-year sits 89bp above the top of that range. Kevin Warsh took the oath as chair on 22 May, so the market is pricing a reaction function it has never seen tested.

The European Central Bank is described in the weekly briefs as finished, and that word needs a date attached to it. It is finished since June. On 11 June the Governing Council raised all three key rates by 25 basis points, reasoning in its own words that "the war in the Middle East is generating inflation pressures," which took the deposit facility to 2.25% with effect from 17 June. On 23 July it held that level with no pre-commitment to a rate path. The June projections carry HICP at 3.0% this year, 2.3% next and 2.0% in 2028.

The Bank of England held Bank Rate at 3.75% on 30 July on a 6-3 vote, with Greene, Mann and Pill preferring 4%. Its own dashboard carries UK inflation at 2.9% against a 2% target, and the next decision is 17 September. The Bank of Japan raised to around 1.0% on 16 June by a 7-1 majority, effective the following day, and held on 31 July by 8-1, with Takata Hajime voting for 1.25%.

No G4 central bank is easing. Two raised in June. Two are holding with a hawkish minority already dissenting. That is the regime: a synchronised energy shock, no policy cushion anywhere, and a long end left to price whatever is left over.

The curve has been doing this arithmetic in public. On 18 June the 2s10s spread was +27bp, the low of the summer, and on 30 June it was +30bp. By 17 August it reached +53bp, closing 25 August at +47bp. That is 23bp of steepening across the same seven weeks that carried the 38bp term premium move.

The mechanism changed three times while the direction did not: a long-end selloff, then a parallel drift higher, then a front-end rally on the tame CPI. Three different causes producing the same widening is the clearest statement the market has made all summer. The week to 21 August broke the pattern with a bear flattening to +50bp as the front end cheapened into Jackson Hole, a pause in the mechanism rather than a reversal of it.

The level is pinned. Since 24 July the 10-year has not closed outside 4.61% to 4.75%, through a hawkish hold, a soft CPI and a record week in JGBs. The top of that band, the 4.75% of 31 July, is the number that matters: no close has been at or above it since 14 January 2025. A break above with the 2-year steady confirms the term premium thesis. A break below with the 2-year leading is the policy path reasserting itself. Gilts sit outside the argument, charging 130bp over Bank Rate on the 24 August fixing.

The year-to-date picture cuts the other way. On 31 December 2025 2s10s was +71bp, with the 2-year at 3.47% and the 10-year at 4.18%. It is +47bp now: 24bp flatter across 2026, because the 2-year added 70bp against the 10-year's 46bp. The steepening is a summer phenomenon sitting inside a year of front-end repricing, the same split the ACM decomposition found.

The Japanese spread shows the same tension from the other side. The 10-year JGB traded 187.6bp through the 10-year Treasury on 3 August and 174.3bp through it on 25 August. Thirteen basis points closed in three weeks, the JGB adding 7bp and the Treasury giving back 6bp. That convergence comes from both ends at once, which is what makes it a standing pressure on the offshore bid.

Three ways this resolves, each with a trigger attached rather than a probability.

Path one: the premium sticks.

This needs the supply and buyer arithmetic to keep doing what it did in June: refunding sizes held flat by guidance, a TIC total that stops growing, official holders that keep shrinking. The 10-year then breaks the 4.75% of 31 July with the 2-year steady and 2s10s widens past the +53bp of 17 August. The triggers here are fiscal and foreign rather than monetary: the 16 September TIC release covering July, the next refunding statement, and the UK autumn budget in November.

Path two: the path was repriced, not the premium.

Warsh delivers his first Jackson Hole keynote on 28 August, and the FOMC meets on 15 and 16 September with a Summary of Economic Projections. Three members wanted 25bp more in July. If the projections validate them the front end does the work again, the curve bear flattens back toward the +27bp of 18 June, and the term premium story is postponed rather than disproved. The year-to-date decomposition, expectations 49bp against the premium's 3bp, is the evidence for it.

Path three: Japan sets the price.

The Bank of Japan raised in June and holds with Takata voting for 1.25%. Its September meeting, sitting behind a 30-year JGB that printed a record 4.096% on 18 August, decides whether the largest foreign holder of Treasuries keeps shrinking its book. A further leg higher in JGB yields moves repatriation from thesis to flow, and it registers in the TIC line first.

What would change this view: a term premium falling back toward the 0.512% of 30 June while the 10-year holds inside its 4.61% to 4.75% band would make the summer move a supply-calendar effect rather than a repricing of duration risk. So would a reopening of Hormuz that takes Brent well below the $87.72 it settled at on 10 August, since energy is the one input common to all four blocks. So would a July core PCE, released 26 August, printing well below the 3.3% year-on-year reading June delivered. The positioning tell sits in the weekly COT tables, where a crowded gold long and an unwinding dollar-index net would register a reversal first.