The Marginal Buyer Is a Carry Trade

Chinese banks are buying Treasuries. Nobody knows how much. The interesting part is who stopped.

Subscribe
MarketsMacroeconomicsGeopolitics
Listen to this post
AI Summary Chinese commercial banks are buying U.S. Treasuries by raising dollar deposit rates to attract customer dollars that already exist from trade surpluses, not by converting yuan into dollars, which means this is primarily a currency operation to prevent yuan appreciation rather than a Treasury demand story. …
  • Chinese commercial banks are buying U.S. Treasuries by raising dollar deposit rates to attract customer dollars that already exist from trade surpluses, not by converting yuan into dollars, which means this is primarily a currency operation to prevent yuan appreciation rather than a Treasury demand story.
  • The composition of China's Treasury holdings is shifting from price-insensitive government reserve managers to price-sensitive commercial bank carry trades, which means the marginal foreign buyer now holds duration only when the spread works and will disappear when funding leaves or the trade stops paying.
  • Even if all of China's $121.2 billion in foreign-exchange deposit growth over seven months went into Treasuries, it would be smaller than the $207.1 billion foreigners bought in June alone, so this flow cannot explain current Treasury yields of 4.8% on the 10-year and 5.34% on the 30-year.
  • The author's compression thesis predicts that sustained high Treasury yields will eventually compress equity valuations, but after three years of being right about rates and wrong about the consequence, the author now believes credit spreads will widen first—before equity multiples compress—as corporate debt refinances at higher rates and hyperscalers issue record amounts of debt to fund AI infrastructure.
  • The author provides specific falsifiable tests with dates through 2027, including whether Chinese buying continues when the yuan weakens (which would prove it's a yield trade rather than an FX operation) and whether credit spreads widen while equity multiples hold (which would confirm the compression mechanism).

Reuters reported this morning that Chinese commercial banks have been buying U.S. Treasuries after lifting their dollar deposit rates.1 The reporting is honest about its own limits: the amounts are unascertained, and whether they move China's overall holdings is unknown.1

That gap is the whole story. It is also the reason the story is being read wrong.

Start with the size, because it settles the question everyone is asking

In June alone, foreign residents bought $207.1 billion of long-term U.S. securities.2 Chinese foreign-exchange deposits grew $121.2 billion across the entire first seven months of the year, reaching $1.18 trillion at the end of July.1

Even if every marginal dollar of that deposit growth had gone into Treasuries — it did not — the total would be smaller than one ordinary month of foreign long-term buying.

Meanwhile, the 10-year sits near 4.8%, and the 30-year touched 5.34% in August, its highest since 2007.34 Those levels were not set by Chinese commercial banks and will not be unset by them.

Important

This is not a yield story. Anyone using it to explain the long end is reaching for a number nobody has.

What is actually being reported

Read the mechanism instead of the headline.

The banks are not converting yuan into dollars. They are bidding for dollars their customers already hold — export receipts and trade-surplus balances parked in FX deposits — by paying up on deposit rates. Yuan deposits at major state banks pay 0.95%. Negotiated dollar rates for large accounts moved above 3% starting in June.1

Those dollars never touch the FX market. They move from a customer account onto a bank balance sheet, and the bank matches a dollar liability with a dollar asset.

That is a currency operation with a Treasury purchase attached. Not a Treasury trade with a currency side effect. The purpose is to stop dollars from being sold for yuan.

Which means most of the commentary, including a version I wrote down myself before I checked it, has the trigger variable wrong. The loop is not:

yields rise → Chinese demand appears → yields moderate

It is:

yuan appreciation pressure rises → dollar retention is encouraged → Treasuries get bought

Right now those two conditions coincide. They do not have to. A stabilizer that only appears when the yuan is strong is not a Treasury-market stabilizer. It is an FX operation that occasionally resembles one.

That distinction is testable, and I have put a date on it below.

The finding is compositional, not directional

China's recorded Treasury position has fallen all year — $695.3 billion in January, $653.3 billion in March, $633.4 billion in June, a multi-year low — with long-term net transactions dominated by sales.5

Over the same window, SAFE's banking-sector external accounts show foreign bond assets jumping from roughly $547 billion at end-December to $633 billion at end-March.6

But here is the rub — both can be true. Commercial banks buying inside a shrinking aggregate is not a contradiction. It is a handoff.

And the handoff is the finding.

The seller is a reserve manager: price-insensitive, mandate-driven, slow, and effectively permanent capital in the Treasury market. The buyer is a spread book: it holds duration because the carry works, it is funded by deposits that can leave, and it unwinds when the trade stops paying.

What this does to the compression thesis

For those who have been following my Compression thesis, it supports it, but through a channel I had not been using.9

The thesis has been that structurally higher long-end yields eventually compress elevated equity and credit valuations even if nominal growth and AI-driven productivity hold up. The usual telling is a discount-rate argument: higher risk-free rate, higher hurdle, lower justified multiple.

The Chinese bank story adds something better than that. It says the composition of the marginal foreign holder of U.S. duration is changing — from price-insensitive official capital to price-sensitive carry.

A price-sensitive buyer does two things:

  1. It caps the upside tail, because there is a level at which the trade turns on, and demand appears.
  2. It evaporates on the downside, because the same buyer is gone the moment the spread closes or the funding leaves.

That produces the specific rate environment the thesis actually needs. Not a spike to 6% that forces a policy response and resolves. A durable 4.5–5% with a floor under it made of foreign spread books that will not be there in a stress. Higher base, thinner defense.

Which is worse for equity multiples than a crisis, and considerably harder to write about, because it never produces a date.

Why the thesis keeps not arriving

Checkmate
As most professionals will acknowledge, even if we still have the more visible characters predicting when the "event" occurs and seemingly never does, there are a lot of moving chess pieces to the global economy, and it takes time to arrive at a checkmate.

Ten-year yields have been at or above 4.5% for most of three years. Index multiples did not compress. The S&P's forward earnings yield has sat below the risk-free rate for a sustained stretch — investors accepting less current yield from equities than from Treasuries — and the market kept paying anyone who ignored it.

A thesis that has been right about the rate and wrong about the consequence for three years owes an explanation or a retirement.

The explanation I actually believe is that the discount-rate channel is the weakest of the three transmission channels, and it is the one everybody watches. It operates on sentiment, and it can be overridden indefinitely by a growth story. The other two are mechanical, and they run on a schedule:

Refinancing. Corporate America termed out its debt in 2020 and 2021 at coupons that no longer exist. The coupon stack has to roll before the rate matters. That is a maturity calendar, not a market event.

Credit as the AI financier. Five hyperscalers have issued $220 billion of debt this year funding data centers and models, more than double last year, and global corporate issuance has hit a record $4.9 trillion.7 The productivity boom is being paid for in the same bond market that is repricing. That is the connection the discount-rate framing misses entirely.10

Important

Both of those transmit through credit before they touch equity multiples. Which means I have been watching the wrong instrument.

The dates

A thesis that cannot be killed is a mood.

So here are the tests, with what confirms and what falsifies, in the order they arrive.

September 16, 2026 — July TIC.2 If China's headline holdings rise materially and long-term net Treasury transactions turn positive, the compositional story is wrong: this is simply China buying, and the handoff framing goes in the bin. If holdings fall again while FX deposits keep climbing, the handoff holds.

  • The next SAFE banking-sector release — Q2 external assets. If foreign bond assets do not extend meaningfully past $633 billion, the Reuters flow is small, and the entire economic reading of this story collapses to anecdote. I would rather find that out than keep the framing.
  • Monthly PBOC FX deposits, against CNY. This is the clean test of my correction to the feedback loop. If deposit growth stalls and reported buying stops while the yuan weakens — with yields still above 4.5% — the trigger is FX, as I have argued. If buying continues through a period of yuan weakness, I am wrong, and it is a yield trade after all.

February 26, 2027 — benchmark survey preliminary data.8 The only clean read on custody-obscured Chinese holdings, and it is six months away. No article published this month gets to claim it.

And for the compression thesis itself, which is the one that has been deferring:

By the end of 2027, if the 10-year has held above 4.5% and the S&P 500 forward earnings yield still has not converged toward it — a negative equity risk premium sustained across a fourth consecutive year, with index multiples flat or higher — then the discount-rate channel does not operate at the index level, and I should stop writing as though it does.

  • Through the 2027 maturity wall, if corporate debt reprices into 5%-plus coupons and aggregate interest coverage outside the mega-caps does not deteriorate, the refinancing channel is dead too.
  • If credit spreads stay tight through a full year of 5%-plus 30-year yields while hyperscaler issuance continues at this pace, the credit channel is not transmitting either, and the thesis was three arguments that all failed for the same reason: I mistook a price level for a mechanism.

Credit goes first. If compression is real, spreads widen while equity multiples are still holding — that is the leading indicator, and it has not happened yet.

So the near-term claim is narrow and it is checkable. Chinese commercial banks are running a funded carry trade that quietly replaces permanent official demand for U.S. duration with reversible private demand.

It does not move yields today. It changes who is standing there when yields next try to move.

The first number that tells us anything lands at 4 p.m. on September 16.

Footnotes

  1. Exclusive: Chinese banks purchasing Treasuries after wooing dollar deposits, sources say — Reuters, September 4, 2026 — The report this piece is built on, including its own stated limits — amounts unascertained, effect on China's overall holdings unknown — plus the deposit-rate mechanism (yuan deposits at 0.95% against negotiated dollar rates above 3% from June) and Chinese FX deposits growing $121.2 billion over seven months to $1.18 trillion at end-July. https://www.investing.com/news/stock-market-news/exclusivechinese-banks-purchasing-treasuries-after-wooing-dollar-deposits-sources-say-4888904 ↩
  2. Treasury International Capital Data for June — U.S. Department of the Treasury, August 17, 2026 — Foreign residents bought $207.1 billion of long-term U.S. securities in June — the scale against which the Chinese deposit growth is measured. The July release, due September 16, is the first falsification date named in this piece. https://home.treasury.gov/news/press-releases/sb0606 ↩
  3. Global bonds sell off as Middle East conflict escalates, further stoking inflation fears — CNN Business, September 1, 2026 — Where the 10-year sits near 4.8% — the level this piece argues Chinese commercial banks did not set and will not unset. https://www.cnn.com/2026/09/01/investing/global-bond-market ↩
  4. Global bond markets are getting hammered — CNN Business, August 18, 2026 — The 30-year touching 5.34% in August, its highest since 2007. https://www.cnn.com/2026/08/18/investing/global-bond-market ↩
  5. Major Foreign Holders of Treasury Securities — U.S. Department of the Treasury (TIC), monthly series through June 2026 — China's recorded position falling through the year — $695.3 billion in January, $653.3 billion in March, $633.4 billion in June, a multi-year low — with long-term net transactions dominated by sales. This is the seller side of the handoff. https://ticdata.treasury.gov/Publish/slt_table5.txt ↩
  6. SAFE Releases Data on External Financial Assets and Liabilities of China's Banking Sector as at the End of March 2026 — State Administration of Foreign Exchange, June 25, 2026 — The buyer side of the handoff: banking-sector bonds investment of $632.9 billion at end-March 2026, against roughly $547 billion at end-December 2025 (safe.gov.cn/en/2026/0326/2401.html). The $86 billion increase is a ceiling on Treasury purchases, not a measure of them. https://www.safe.gov.cn/en/2026/0625/2425.html ↩
  7. Explainer: What's behind the selloff in world bond markets? — Reuters, September 2026 — Five hyperscalers issuing $220 billion of debt this year to fund data centers and models, more than double last year, against a record $4.9 trillion of global corporate issuance — the productivity boom being financed in the bond market that is repricing. https://finance.yahoo.com/economy/policy/articles/explainer-whats-behind-selloff-world-141230628.html ↩
  8. Report on Foreign Portfolio Holdings of U.S. Securities at End-June 2025 — U.S. Department of the Treasury, April 30, 2026 — The benchmark survey that gives the only clean read on custody-obscured holdings; the next covers end-June 2026, with preliminary data expected February 26, 2027. https://home.treasury.gov/news/press-releases/sb0482 ↩
  9. Valuation Gravity: Why 2022–2028 Could Mirror Historic Compression Cycles — The Compression Series entry where the multiple-compression argument is made directly — a century of compression cycles, the 2022–2028 roadmap, and the Dollar Trap section warning that foreign holders of Treasuries face a choice between accepting debasement and leaving. That warning is what this piece revisits: not whether they leave, but who replaces them. ↩
  10. The Boom Is the Fuse — The same connection worked out from the macro seat, in MarketAlly: the AI buildout is financed in the bond market that is repricing, the debt is serviced by revenue that assumes a productivity dividend gets broadly monetized, and the collateral behind it depreciates faster than the loans mature. Where this piece asks who holds the duration, that one asks who is left holding the paper. ↩
Back to the Journal