Treasury triples its long-end buyback
Treasury tripled its long-end buyback to $6 billion and yields rose anyway, with the thirty-year at 2007 levels. The same move ran through munis and agency mortgages, alongside fresh numbers on where AI in the investment process stops.
Treasury triples its long-end buyback and yields rise anyway
Treasury said on 9 September that it would buy up to $6 billion of 10- to 20-year notes at the following day's operation, three times the size of its last long-dated buyback and well above the at-least-$4 billion floor it set on 19 August for the 9 September to 4 November window. Long yields rose regardless. The 10-year reached 4.8528%, its highest since November 2023, and the 30-year touched 5.29%, a level last seen in 2007. The backdrop did most of the work: Brent above $100 on US-Iran tension, August payrolls at 162,000 against roughly 55,000 expected, and CME FedWatch pricing better than a 60% chance of a hike at the 16 September FOMC. Ira Jersey of Bloomberg Intelligence told Axios that Treasury would need "maybe $10 billion or more per operation" to matter; Deutsche Bank's Steven Zeng said "it's like Treasury created this monster that it now has to keep feeding"; BNP Paribas called the operations a "mere 'Band-Aid' solution".
The step from $2 billion to $4 billion, reported here last month, was plumbing; the step to $6 billion was an attempt at price-setting, and it failed in a single session. Long-end execution assumptions and duration overlays built on an implied official bid now need re-parameterising, and the operation results are the cleanest free read on where real long-end liquidity sits — worth capturing at CUSIP level rather than inferred from a screen.
Treasury · Reuters via Investing.com · Euronews · Axios · FOMC odds · SWBC
Licence terms, not model quality, cap AI at one step
Nasdaq published its 2026 Global Voice of the Issuer study on 10 September: 406 senior investment product professionals — 356 asset managers and 50 insurers, all at firms with $25 billion or more in assets, across seven countries, fielded with Escalent. Eighty percent use AI somewhere, but only 26% run it across more than one stage of the workflow; 69% use it for idea generation and 61% in product design. Non-adopters named compliance concerns (53%), lack of internal expertise (53%) and doubts about accuracy and transparency (50%). Asked where the infrastructure is missing, 58% pointed at private-market benchmarks and 41% at digital-asset benchmarks. Two days earlier SIX launched Market Signal, an API service carrying SIX Swiss Exchange, BME and 16 European markets via Aquis, built on the former IEX Cloud assets and — the notable clause — licensed for training trading models and AI agents. SIX did not publish pricing. On 10 September Waters reported Databento, profitable on 24 staff with revenue up 6.65x year on year after a $97 million Series B led by NEA in July, extending its pay-as-you-go model from historical into real-time data.
What stops AI at step one, on managers' own account, is permission and provenance rather than capability — and the week's supply-side moves competed on licence terms, not content. That is the opposite direction from LSEG, which added an AI use-case licence plus usage-based charges for API and MCP access on the argument that agents consume roughly ten times what humans do, and from CME, which reclassified end-of-day data as delayed at $304 per exchange per month; SIX's price is undisclosed, so a like-for-like comparison is unavailable, but the real number is never the subscription line — it is the legal review, the audit trail and the entitlement plumbing each new source drags behind it, which is precisely the cost that keeps a working pilot from reaching a second desk.
Nasdaq · SIX · WatersTechnology · Databento · LSEG · CME licensing
Munis hit 2011 yields as supply outruns retail demand
The holiday-shortened week carried a calendar that grew from $15 billion to $17 billion against $7.7 billion in Labor Day week a year ago. By 9 September the 10-year MMD-to-Treasury ratio was 74% and the 30-year 91%, both the cheapest of the year, and New York City sold $1.36 billion of general obligations with 5s of 2041 at 4.52%, Fort Worth $497.12 million of convention centre bonds at 5.07% for 5s of 2051, and Minnesota $1.03 billion. A $3.8 billion Alabama toll road financing accounted for much of the rest. Thursday broke it: the 10-year benchmark rose 15 basis points to 3.69%, the highest since April 2025, and the 30-year rose 14 to 4.89%, the highest since February 2011, while JPMorgan counted $460 million leaving muni funds in a single day, the largest since April 2025, on top of $947 million of ICI-reported weekly outflows to 2 September. Friday recovered up to 5 basis points, and next week's calendar drops to $10.3 billion led by $1.67 billion from the New Jersey Transportation Trust Fund Authority.
Ratios at the year's cheaps with retail money leaving is the separately managed account problem in its purest form: the marginal long-end buyer is a price-taker on ratios and a forced seller on redemptions, and both show up in the odd-lot bid-offer exactly when accounts need to raise cash. Few SMA books can measure what that cost them — the trades are there in the tape, but reconciling a thousand small fills against same-day comparable prints is a data exercise most muni desks have never been resourced to run.
Bond Buyer, 8 Sept · Bond Buyer, 9 Sept · Bloomberg via Briefs · Bond Buyer, 11 Sept
Agency MBS undershoots as home equity and non-QM take share
BTIG's Douglas Harter told National Mortgage News on 9 September that agency MBS issuance across his five-company coverage universe is tracking $134.6 billion against a $141.1 billion consensus, and that holding the July-August pace through September would leave the quarter 6% below his own forecast and 12% below consensus. August agency issuance was $116.2 billion, up 5% on July, but the misses were concentrated: PennyMac 25% below estimates with August MBS volume of $5.8 billion, down 19% month on month, and loanDepot 17% below, narrowing to 3% once HELOCs are counted. The growth sat elsewhere. Home equity issuance was $3.3 billion in August, running 19% faster than the prior three months and 10% above 2025, with Bank of America Securities putting year-to-date volume at $30 billion to 28 August. Non-QM slowed to $8.4 billion in August, 17% below the second-quarter pace, but still reached $82 billion year to date, up from $68 billion at 31 July.
Every unit of origination that migrates from agency to home equity or non-QM moves from a market with a TBA benchmark, published prepayment data and cheap hedges into one where each deal needs its own model and its own loan-level licence. The spend that follows is not the coupon — it is analyst headcount, loan-tape subscriptions and a valuation stack that cannot be reused, and the loanDepot figure that only makes sense once HELOCs are added back is a small warning that even the issuance series a desk tracks are drifting out of alignment with what is actually being produced.
MarketAxess sells liquidity into other firms' front ends
IMTC said on 9 September that it has integrated MarketAxess into its portfolio and order management platform, letting portfolio managers route orders and reach Open Trading without leaving the construction workflow, aimed explicitly at SMA and model-driven custom portfolio managers. Spencer Lee, MarketAxess head of products for the Americas, described it as "embedding liquidity directly into the investment workflow"; IMTC chief executive Russell Feldman said portfolio managers "need optimisation, compliance, and execution to operate in one connected workflow". It follows MarketAxess integrations with OptimX in June and FactSet a year ago. The contrast is Tradeweb's 4 September report: fully electronic US credit average daily volume up 32.5% year on year to $8.9 billion, a 19.0% fully electronic share of high-grade TRACE and 26.8% including all protocols, with global cash credit portfolio trading ADV up 50.4%. MarketAxess suspended its monthly volume releases ahead of the ICE deal, so there is no comparable August print to set against it.
A venue that is losing share on its own screen and is about to be owned by an exchange chasing distribution has an obvious answer: stop selling the screen and start selling the liquidity into whatever screen the client already uses. Neither IMTC nor MarketAxess disclosed commercial terms and Tradeweb's equivalent OEMS arrangements could not be established from public sources, so the cost comparison is unavailable — but the governance point does not need pricing: once the construction tool decides which venues are reachable, best-execution analysis has to be re-based on what was reachable that day rather than what existed, and the switching cost of the platform quietly becomes the switching cost of the liquidity.
Jefferies shuts outsourced bond trading ten months in
Jefferies is winding down the outsourced fixed income trading desk it launched in the US in November 2025, and Joram Siegel, who joined in April 2024 to head the business, is leaving with it. The firm declined to comment. In February it had selected TS Imagine's order and execution management platform specifically to scale the operation, which means the technology build was barely finished when the business was closed.
Renting a bond desk only works when the provider's fixed costs are already sunk somewhere else, and fixed income outsourced trading does not share much with the equities version — it needs its own dealer connectivity, its own evaluated pricing and its own execution stack, all of which Jefferies was still paying for. Anyone currently weighing outsourced fixed income execution should price provider durability as a real line item, because the switching cost of an abandoned desk lands entirely on the client, and it lands at the worst possible moment in a market that has just repriced the long end.