The question is not “what is capital?”

Book capital is an accounting residual, the difference between recognized assets and liabilities. The more interesting question for risk managers and regulators is what is behind that residual and how much real loss-absorbing capacity it represents.

In insurance, it’s even a little more complicated as there is non-GAAP accounting, specific assets earmarked to back capital, and there regularly are granted exceptions to the rules by the multitude of regulators.

Regulatory Arbitrage

I’m on record that all things described as ‘regulatory arbitrage’ are not per se bad. Often times the regulatory rules are uneconomic, and it is the responsibility of CFOs to optimize their returns on capital deployed, subject to proper risk constraints.

The recent reporting on Mark Walter, Guggenheim and the insurance companies around them has begun to move beyond the original question of whether Delaware Life and Clear Spring properly disclosed their related-party exposures. The broader question is becoming whether the insurance/private-credit model itself has exploited gaps between the economic form of risk and the regulatory form in which that risk appears

There has been some extremely good forensic work here. Nick Nemeth’s The Guggenheim Universe reconstructs overlapping investments and reinsurance relationships across eight insurers with historical ties to Guggenheim. His admonition is exactly right, that

all capital should be tested for hardness.

Peter Dziedzic’s The Dodgers Are on Security Benefit’s Balance Sheet explains one particularly large capital arbitrage at Security Benefit: collateral loans. And Rod Dubitsky’s Security Benefit Life (SBL) Bet It All on a Risky Asset, Then the NAIC Closed the Loophole extends the inquiry to reinsurance, Security Benefit’s Vermont captive and the SkyRidge sidecar.

The FT has now placed these questions inside the much larger private-capital/insurance trade in Mark Walter’s unravelling empire tests the insurance trade behind private credit’s rise (FT), while the Journal reports in Mark Walter’s Troubles Are Disrupting the Insurance World’s Hottest Trade (WSJ) that the fallout is already complicating Aquarian’s proposed acquisition of Brighthouse Financial.

I think Nemeth, Dziedzic and Dubitsky are onto something important.

But before putting everything under the heading of “regulatory arbitrage,” it is worth separating several quite different things: some involve taking advantage of an inadequately calibrated national rule, some involve an explicit exception granted by a state regulator, some are regimes deliberately created by state legislatures, and some are conventional reinsurance structures. And those difference may matter.

And some come down to, paraphrasing Bill Clinton, what the word affiliate actually means.

A very short insurance-regulation primer

It feels strange for me to write this. Because after even a decade in insurance as a risk manager things can get complicated in ways that still make my head hurt.

For readers who sensibly spend their time on other things, the NAIC is not a federal insurance regulator. In the US, insurance remains primarily state regulated. The NAIC develops model laws, statutory accounting rules and risk-based-capital formulas that states generally adopt, sometimes with modifications. We will also likely have to discuss Bermuda, which has an entirely different regime based on the Solvency II framework used in much of the rest of the world.

Two concepts matter here.

First, unlike banking, where the regulators standardized on GAAP accounting, insurers report under statutory accounting principles, or SAP, rather than ordinary GAAP for regulatory purposes. SAP is deliberately solvency-focused. Among other things, as referenced above, an asset generally needs to qualify as an admitted asset before it counts toward statutory capital and surplus.

Second, insurers are subject to risk-based capital, or RBC. Different assets require different amounts of capital according to their perceived risk. A 30% RBC factor doesn’t mean the insurer expects to lose 30%; it means the regulatory capital formula requires substantially more capital against that asset than against one carrying, say, a 6.8% factor. And it can differ materially from a short-term economic capital view of the world.

Yet another wrinkle is that a jurisdiction can have

• a prescribed practice, a law, regulation or rule under which its insurers account differently from ordinary NAIC SAP, and a

• permitted practice where a regulator can allow a particular insurer to depart from the normal rules. A bespoke treatment.

And these distinctions again can matter.

Five Mechanisms Worth Seperating

There are five mechanisms worth discussing that come up in the various articles discussing the Water (and now Sammons) insurance and reinsurance entities:

1. Asset-side RBC arbitrage; collateral loans.

2. Bespoke permitted practice; SARC/Vermont XOL asset.

3. Prescribed captive financing; Sammons/Iowa LPS.

4. Reinsurance sidecar capital; SkyRidge.

5. Affiliate/control classification; whether structurally close relationships remain legally “unaffiliated.”

Asset-side RBC arbitrage; collateral loans.

I think this is the easiest one to understand.

Security Benefit has an extraordinarily large collateral-loan portfolio. A collateral-loan portfolio is just a collection of loans that are secured by pledged assets (equity interests, joint ventures, real estate or structured securities) rather than by the borrower’s general credit. Dziedzic reported that in 2024 Security Benefit held $12.9 billion of collateral loans, almost all of which were backed by assets affiliated with Eldridge. Eldridge, the investment group founded by former Guggenheim president Todd Boehly and owner of Security Benefit. Dubitsky’s more recent work puts the exposure at roughly $14 billion.

Historically these loans received a 6.8% RBC charge. The problem is that the collateral can consist of things that might attract a 30% or 45% factor if the insurer simply owned them directly.

Dziedzic puts it nicely: “That gap is the point.” Iowa regulators were even less delicate, describing collateral loans as “the most easily exploited asset class for capital arbitrage.”

This is genuine regulatory arbitrage in the classic sense. And the legally appropriate sense.

Importantly for the story, they didn’t hide this fast, and the NAIC rules give the transaction favorable treatment.

Now, the NAIC has responded by changing its rules, effective for year-end 2027, to implement a look-through approach based on the underlying collateral, while giving some credit for overcollateralization.

Dubitsky estimates that the change could take Security Benefit’s required capital on these assets from roughly $1 billion to more than $4 billion. That’s a potentially enormous number, but I wouldn’t imagine the transaction survives in its current form once the rule is in place. CFOs are creative.

As a former regulator, they followed the rules. As a risk manager, I’d ask:

• How much economic risk is in the transaction?

• How independently have those collateral values been established?

Bespoke permitted practice; SARC/Vermont XOL asset.

Security Benefit’s Vermont captive, Sixth Avenue Reinsurance Company, or SARC, is a fundamentally different case.

SARC reinsures certain long-duration annuity risks from Security Benefit and has excess-of-loss reinsurance (XOL) protecting part of that exposure. Vermont specifically allows SARC, through a permitted practice, to recognize the value of that excess-of-loss protection as an admitted asset in a manner that ordinary NAIC rules would not.

Security Benefit’s own regulatory filings say that, without the permitted practice, its statutory value would become deeply negative and its RBC “would have triggered a regulatory event.”

Nemeth and Dubitsky are therefore quite right to focus attention here.

But even here I would formulate the question slightly differently from Nemeth and Dubitsky. The capital is not per se fake. There really is an excess-of-loss reinsurance contract. It presumably has some economic value. So the relevant questions are:

• Who provides the XOL protection?

• What is the attachment point?

• How remote are the scenarios under which it pays?

• How is the counterparty collateralized?

• And why does Vermont believe that future contingent recovery deserves current admitted-asset value when ordinary NAIC SAP does not?

Those answers might support Vermont’s treatment, undermine it, or land somewhere in between.

And one distinction is important: SARC failing an RBC test without the permitted practice is not the same thing as Security Benefit itself being insolvent. Nemeth’s underlying observation is real but we should be careful about the implications until we have more information. Especially about whether the losses would upstream to Security Benefit.

Definately worth digging deeper.

Prescribed captive financing; Sammons and the Iowa LPS companies

Nemeth uncovers another remarkable set of numbers at Sammons. Sammons Enterprises is a large, privately held conglomerate whose insurance arm owns Midland National and North American and has long been both a major shareholder of Guggenheim and one of its largest asset-management clients, making Sammons a key external node in the Walter/Guggenheim insurance network.

Three Iowa captive reinsurers — MNL Re, Solberg Re and Canal Re — assume insurance liabilities from the Sammons insurers. Like any reinsurer, they must support the reserves associated with those liabilities. But instead of funding all of that support today with cash, bonds or other conventional assets, the Iowa structure allows part of it to take the form of contingent financing commitments: if policy benefits on specified blocks exceed defined thresholds, outside guarantors are obligated to provide additional funds.

Iowa allows those contingent commitments to be recognized as admitted assets for statutory purposes, even though ordinary NAIC accounting would not.

That difference has a very large effect on how Midland values its investments in the three captive subsidiaries. With the Iowa treatment, Midland carries its interests in MNL Re, Solberg Re and Canal Re at a combined value of roughly $322 million. Without that treatment, the corresponding values would total roughly negative $1.53 billion, a difference of about $1.85 billion.

Again, those are numbers worth paying attention to.

But the mechanism appears to be somewhat different from Nemeth’s description. He characterizes the $1.85 billion of admitted notes as,

in substance, IOUs the group wrote to itself.

That prompts an important question: Are they?

I tried to confirm that characterization and could not get there. The current Midland filings say MNL Re’s contingent note guarantee comes from an unrelated third party. Solberg and Canal describe their providers more specifically as unrelated third-party insurance companies. The arrangements operate somewhat like standby letters of credit: the outside provider is required to supply funds only if actual policy benefits on the reinsured blocks exceed specified thresholds.

There is a second distinction that I think is even more interesting.

This is not really a Sammons-specific permitted practice. Iowa deliberately created a statutory regime for limited-purpose subsidiary life insurance companies. Iowa law expressly authorizes the commissioner to approve their financing arrangements and provides that assets approved under that regime are treated as admitted assets.

In other words, Iowa did not merely look the other way, it built the architecture into its regulatory framework.

That shifts the question from:

• Did Sammons find a clever accounting loophole?

to:

• Why did Iowa decide that contingent third-party financing commitments should receive current admitted-asset value, and how should a risk manager haircut that value compared with cash or conventional invested assets?

I find the second question substantially more interesting.

This is where Nemeth’s phrase “all capital should be tested for hardness” becomes particularly useful. A contingent promise from a financially strong independent reinsurer may have considerable economic value. But is a dollar of that value as available to absorb tomorrow morning’s loss as a dollar of Treasury securities? And it’s substantially worse if the guarantor party ends up being affiliated or related. That’s the house-of-cards risk.

Reinsurance sidecar capital; SkyRidge

Not to be confused with SkyBridge Capital.

Dubitsky’s newest and, to my mind, most interesting observation concerns SkyRidge Re.

Security Benefit cedes roughly $10 billion of reserves to SkyRidge entities but reports SkyRidge as unaffiliated. Dubitsky points to the close relationship and asks how that can possibly be so.

The answer may be that the legal classification is perfectly defensible.

But the underlying structure raises an excellent prudential question.

Security Benefit’s own disclosures say it owns approximately 20.2% of SkyRidge’s capital but only 9.9% of its voting power. Security Benefit manages the funds-withheld assets associated with the liabilities it cedes to SkyRidge, and another Security Benefit affiliate provides services to the vehicle.

Why is 9.9% interesting? Kansas insurance law, similar to banking law, presumes control when a person holds 10% or more of the voting securities. Importantly, the absence of that presumption does not prevent the commissioner from finding control based on the facts.

I would therefore pose Dubitsky’s question somewhat differently:

Is SkyRidge actually misclassified, or does the structure demonstrate the difference between legal control and economic dependence?

And then I would want to know more.

• What board rights does Security Benefit have? What vetoes?

• Who can replace the asset manager? Who sets investment guidelines?

• Can SkyRidge independently terminate its relationship with Security Benefit?

• How concentrated is SkyRidge’s business in Security Benefit?

• What happens if Security Benefit stops originating new business?

The 20.2%-economic/9.9%-voting structure is certainly worth understanding. But the proximity to a 10% legal presumption is a reason to ask the question, not evidence by itself that anybody circumvented the law. Things get structured this way all the time.

The larger issue is more important anyway:

• A counterparty can be legally independent and still be economically very closely tied to you. That matters to a risk manager even when it does not matter to the legal definition of affiliate.

In the global financial crisis, SIVs were legally independent, but closely tied to the banks for liquidity upport.

Affiliate/control classification

This brings us back to Nemeth’s broader The Guggenheim Universee thesis.

His reconstruction finds hundreds of overlapping private securities held across insurers with four nominally separate ownership groups. He argues that they behave like

one manager, one book, one set of marks.

The common holdings are important. But I would stop short of saying that eight insurers are therefore economically one insurer.

They have different liabilities, capital cushions, owners, liquidity positions, hedges and reinsurance arrangements. An identical asset impairment can be survivable for one and fatal for another.

What Nemeth has identified more persuasively is common-factor risk.

Legal independence does not necessarily create asset independence, valuation independence, origination independence or funding independence.

That distinction suggests a concept that may be more useful than continually debating whether a company belongs in an “A” or “U” column of a statutory schedule.

Call it economic affiliation.

Legal affiliation asks:

• Who controls the entity?

Economic affiliation asks:

• How independent is the risk?

An economic-affiliation analysis might consider voting control, economic ownership, common origination, investment management, servicing, guarantees, reinsurance, common directors, source of repayment and dependence on the same underlying businesses.

SkyRidge could therefore be legally unaffiliated but highly economically connected.

The same question can be asked of some of the opaque borrowing vehicles Nemeth has identified, and of the post-2024 relationship between Sammons and Guggenheim.

That does not mean the legal classifications are necessarily wrong.

It means legal affiliation may be an incomplete prudential risk measure.

Once upon a time in the banking world there was unit banking, in which a bank was confined to a single branch rather than having a network of branches. Bankers got around this by establishing chain banks and group banks with commorn contrrol, the predecessor to the modern bank holding company.

And what was observed was that duress was transmitted across the chain/group. Interconnection creates channels of contagion, but it does not mean contagion must occur. Capital, liquidity, asset quality and management determine whether the channel actually transmits distress.

In the GFC we found that legal seperation DID matter, particularly as it related to the US vs. UK. Banking history provides a useful distinction between correlation and contagion.

• If Delaware Life and EquiTrust both own the same private-credit security and it falls in value, that’s primarily a common-shock/correlation problem.

• If Delaware Life’s distress causes the security to be marked down at EquiTrust, that’s a valuation-contagion channel.

• If Clear Spring has supplied capital to another insurer that must then be withdrawn or written down, that’s a direct financial contagion channel.

• If policyholders or counterparties see trouble at one Guggenheim-network insurer and pull business from another, that’s information/reputational contagion/run-risk.

Not every “permitted practice” is a loophole

SFG Bermuda’s statutory statements contain BMA-granted permitted practices covering embedded derivatives, unrealized asset gains and losses and market-risk-benefit effects. At first glance that sounds like another source of regulatory capital. But the actual numbers tell a subtler story.

Those permitted practices increased SFG Bermuda’s capital and surplus by about $570 million in 2024, and reduced capital and surplus by about $252 million in 2025.

That sign reversal should make us cautious about treating every occurrence of the words permitted practice as evidence of forbearance. Some differences are principally attempts to deal with accounting mismatches between economically related assets and liabilities.

The question should always be:

What economic risk did the accounting treatment remove from the balance sheet, and did it also remove that risk from the insurer?

If the answer to the second question is no, the capital treatment deserves scrutiny. If the accounting merely removes artificial volatility while the economic capital framework still captures the risk, that is a different story.

The broader point

This leaves me somewhat more sympathetic to Nemeth and Dubitsky’s central concern than to some of their specific characterizations. There really does appear to be a recurring pattern in which the regulatory form of risk changes more than its underlying economics.

On the asset side:

• private equity/JV-like risk → collateral loan → lower RBC

On the liability side:

• insurance obligation → captive reinsurance → contingent financing or XOL protection → reserve credit/admitted asset

On the capital side:

• contingent future financial support → state-authorized admitted asset

And at the organizational boundary:

• deep economic relationship → sub-control voting structure → legally unaffiliated counterparty.

None of those transformations is inherently improper. Most of the examples above appear to operate under rules affirmatively established or approved by regulators.

That is precisely why the issue is interesting.

Regulatory arbitrage is not the same thing as regulatory evasion.

A sophisticated financial institution should be expected to understand the rules and optimize within them. Indeed, that is one of the things we pay sophisticated financial institutions to do.

The regulatory question is different:

• After all the structuring is finished, does the regulatory representation of the risk still bear a reasonable relationship to the economic loss that would arrive in stress?

And the risk-management question is:

• Do I have enough genuinely loss-absorbing capital to survive the economic tail loss, and enough unencumbered liquid assets to meet obligations while the loss is working its way through the balance sheet?”

That is where I think Nemeth and Dubitsky’s work leads.

The Walter investigation started with the more straightforward possibility that economically related exposures were simply classified incorrectly. The wider private-capital insurance story is harder. In many cases the classifications may be entirely correct.

The more uncomfortable possibility is that the rules themselves permit legal independence, asset classifications and capital resources that look considerably less independent or considerably less robust when viewed economically.

And that is a much bigger question than whether somebody filled out the affiliate column incorrectly.

• The Numbers Person

• The Cash Machine (Mispriced Assets)

• Exclusive: Walter and TWG Global Likely Can’t Use CVNA for Liquidity. Citi Already Secured It. (Hunterbrook Media)

• Guggenheim’s Quiet Owner: $130B Insurance Giant Dodges Walter Drama Tied to Lakers Sale (Hunterbrook Media)

• The WSJ and Bloomberg articles linked in the post.

Anyway, this is just a primer, in case things here go sideways.

The Walter–Guggenheim–Delaware Life Puzzle

Who is Mark Walter?

Mark Walter is the CEO of Guggenheim Partners, which is a mid-sized investment bank. Importantly for this evolving story, Mr. Walter has substantial interests outside of Guggenheim, quite a bit through a company called TWG Global. Among other things, TWG Global is the controlling shareholder of Delaware Life Holdings.

This story, in at least part, will be about the financial architecture he built over two decades encompassing Guggenheim Partners, insurance company balance sheets, private credit, and other businesses he controls.

Delaware Life is an insurer with $4.0 billion of capital and surplus. Note this for later.

Earlier this year, two Walter-controlled insurers, Delaware Life and Clear Spring Life and Annuity, received federal grand-jury subpoenas. An internal review determined that billions of dollars of investments were misclassified as “unaffiliated” when they were in fact “related-party” investments. Federal prosecutors and the SEC are now examining how the transactions were structured and disclosed. There is already enough evidence to say that something significant went wrong. There is not yet enough evidence to say exactly what or why.

At one end of the spectrum, this could prove to be an extraordinary failure of accounting, governance and related-party controls. At the other, prosecutors appear to be exploring whether apparently independent intermediaries and special-purpose vehicles were deliberately used to conceal the true economic relationship between Walter’s insurers and businesses elsewhere in his empire.

Background

We’ve discussed the private credit / life insurance nexus before,as recently as a week ago. And I’m on the record that I think this is a potentially superior way to fund these loans than the traditional banking system.

Life insurers and annuity companies collect large pools of relatively predictable, long-duration liabilities. Alternative asset managers realized that an insurer could become a powerful source of long-term funding for private credit, earning themselves tens of billions of dollars of assets under management along the way.

Apollo’s relationship with Athene is probably the best-known modern example. KKR, Brookfield and others have built variations on the same model.

Walter and Guggenheim were early practitioners.

After the financial crisis, Guggenheim built Clear Spring Life and helped executives acquire or recapitalize several other insurers, including Delaware Life, Security Benefit and EquiTrust.

Two Guggenheim executives may pop up as things develop: Dan Towriss was hired by Guggenheim in 2009 to help Walter acquire insurance companies, and Andrew Kenney, Delaware Life’s current chief investment officer, whopreviously ran Guggenheim’s Private Credit Group.

Today’s companies are legally distinct. Guggenheim Partners is the asset manager, still led by Walter. TWG Global is Walter’s broader investment holding company. Group 1001 contains Delaware Life, Clear Spring and other insurance businesses.

What Delaware Life has already told us

In February, Delaware Life and Clear Spring received grand-jury subpoenas from the U.S. Attorney’s Office for the Southern District of New York. Delaware Life says it understands the investigation to concern whether certain private-credit investments introduced to Delaware Life and Clear Spring by an affiliate should have been treated as affiliated or related-party transactions.

After receiving the subpoenas, Delaware Life conducted its own internal investigation. It found errors, large errors. At the end of 2024, Delaware Life had $9.53 billion of private-credit investments that, after restatement, were judged to be “predominantly contingent on the performance of affiliates.”

By the end of 2025, the number was $16.37 billion. At June 30, 2026, it was $16.82 billion. Delaware Life also disclosed an additional ~$821 million of funding-agreement and trust-note structures whose repayment similarly depends predominantly on affiliates.

$16.8 billion in investments “predominantly contingent on affiliates” compared with $4.0 billion of capital and surplus.

The $16.8 billion is not the direct investments in affiliates, that figure is only $1 billion, it’s something elset. These are investments with nominal counterparties that may not themselves be affiliates, but where Delaware Life has concluded that the investment’s return depends predominantly on related parties. That is the heart of the concern.

The issue is not principally:

• Delaware Life lent $17 billion directly to companies called “Mark Walter Holdings.”

It is closer to:

• Delaware Life lent to Company A, but Company A’s ability to repay ultimately depended upon Company B, and Company B was economically connected to Walter.

This is why the identities and functions of the entities sitting between the insurer and the ultimate economic exposure now matter so much.

OK, But who are the intermediary companies?

According to Prosecutors Focus on Four Businesses Tied to Dodgers Owner Mark Walter (WSJ), investigators have focused on four firms: ABS Capital, Amistad Financial, Bradford Allen and Hudson Trading.

The Journal reports that proceeds from loans made by Walter’s insurers passed through entities purportedly controlled by these firms before reaching other Walter-linked businesses.

• ABS was founded by former Guggenheim executives. As far back as 2016, a Guggenheim compliance lawyer reportedly became concerned that some Walter investments appeared to be financed through ABS and identified a maze of LLCs that appeared to function as Walter vehicles.

• Amistad owns EquiTrust, itself descended from the old Guggenheim insurance network. Delaware Life and Clear Spring filings contain loans to Amistad-related financing vehicles.

Bradford Allen has had longstanding business relationships with Walter and Guggenheim, but nothing per se wrong with that. Hudson remains less transparent publicly.

None of that establishes wrongdoing. An “unaffiliated” counterparty is not required to be a stranger with no prior commercial relationship.

But it raises the central question:

Were these firms genuinely independent counterparties assuming economic risk, or did some of their entities principally function as conduits between insurance capital and Walter-related businesses?

The Guggenheim angle

According to Federal Probe Into Dodgers Owner’s Financial Empire Was Sparked by Whistleblower (WSJ), the current inquiry began not with a whistleblower complaint inside Guggenheim Investments concerning the accounting for advisory contracts at Guggenheim Private Investments, or GPI. The original issue reportedly involved a Middle Eastern financing transaction and questions about how Guggenheim recognized revenue.

Payments associated with those advisory contracts reportedly led investigators to ABS, Amistad, Bradford Allen and Hudson. Investigators then discovered connections between entities associated with those same firms and loans from Walter-controlled insurers into Walter-related businesses.

This week, according to Guggenheim loan trades in distressed territory after investor call (FT) a $1.2 billion loan to a vehicle affiliated with Guggenheim Investments fell from roughly 96 cents to about 78 cents on the dollar following a lender call that failed to reassure creditors.

According to the article, Guggenheim Private Investments reportedly has only four clients, two of which are affiliated with Walter’s wider businesses. Delaware Life and Clear Spring are not among the four.

At June 30, GPIM formally managed only about $2.52 billion of Delaware Life investments, and just $3.4 million of the company’s reported related-party investments. That sounds inconsistent with a $16.8 billion problem.

But “assets managed by GPIM” does not capture all the roles Guggenheim entities can play.

One transaction, reconstructed by Nick Nemeth (The Cash Machine) from statutory filings, is PDIF GCF CLO 2022-1, a roughly $438 million CLO managed by Guggenheim Corporate Funding. Delaware Life reportedly owns the entire capital structure, from senior securities through the first-loss residual. This is a bit strange, but we don’t yet know exactly why this occurred.

That is economically quite different from Delaware Life buying a normal syndicated CLO in which independent investors establish market prices across the capital structure.

The rating-agency angle

A while ago we discussed the role of smaller rating agencies in rating private credit, particularly as these ratings helped determine capital requirements. Egan-Jones was the main focus of these discussions. Now Egan-Jones is popping up in this emerging story.

To Probe Mark Walter’s Business, DOJ Sought Egan-Jones’ Records (Bloomberg) reported this week:

The Justice Department has gathered records of Egan-Jones Ratings Co.’s work with at least two insurers Walter controls, Delaware Life Insurance Co. and Clear Spring Life and Annuity Co., according to people with knowledge of the matter, who asked not to be identified discussing the sensitive inquiries.

Further reporting:

Delaware Life and Clear Spring are among the only four US life insurers that had more than 15% of their bond portfolio rated by Egan-Jones, the industry data show.

At the end of last year, the firm was the lone provider of ratings on about 16% of the $32 billion of bonds in Delaware Life’s portfolio … [and at] the much smaller Clear Spring insurer, at least 50% of its $6.3 billion bond book was exclusively rated by Egan-Jones, the data show.

The insurers’ reliance on Egan-Jones was unusually heavy.

Again, why does that matter? Because the rating can be part of the regulatory economics: investment-grade bonds can require substantially less capital than equity or other high-risk private investments.

Three possible explanations

I don’t want to speculate too much at this time. Perhaps this has a rational explaination, or perhaps this is the tip of an iceberg, we just don’t yet know. At this stage I think there are three broad hypotheses.

The first is a control failure. Delaware Life and Clear Spring may simply have applied too narrow or legalistic a definition of affiliation. Complex private-credit structures proliferated faster than related-party controls evolved, and management, auditors and regulators failed to look through the legal counterparties to the ultimate source of repayment.

The second is regulatory engineering. The structures may have been intentionally designed to remain outside formal affiliate classifications and to achieve attractive capital treatment, with participants believing that they complied with the letter of the rules even while creating economic exposures the rules were intended to capture.

The third is deliberate concealment. Intermediaries and SPVs may have been used specifically to disguise transactions everyone understood were economically related-party financing, with disclosures designed to prevent regulators from seeing that connection. That is much more serious and appears to be one of the questions federal investigators are examining.

We do not currently know which explanation best fits the facts. Different transactions may ultimately have different answers.

What we should not conclude

There are several temptations worth resisting.

First, $16.8 billion of related-party exposure does not mean $16.8 billion of losses. The accounting correction principally changed the classification of the investments, not their carrying values. We still know remarkably little about the actual credit quality of many underlying borrowers.

Second, we should not assume that every opaque LLC is a sham. Purpose-built SPVs are routine in private credit and structured finance. The relevant question is what economic function each one performed.

Third, the evidence does not establish that Guggenheim originated or controlled the entire Delaware Life portfolio. The emerging evidence suggests Guggenheim’s role may be materially broader than the formal GPIM management numbers indicate, but its precise scope remains unknown.

Fourth, Egan-Jones being subpoenaed does not make Egan-Jones a participant in wrongdoing. The company explicitly says it is not a subject or target of the investigation.

Finally, no criminal charges have been brought against Walter or the companies at the center of the story. Investigations often end more quietly than they begin.

Why I think this matters

The reason to follow the story is not Mark Walter, although ultimately it could be.

The larger issue is whether the modern marriage of insurance and private credit contains conflicts that become difficult for conventional insurance regulation to see. One of the most interesting papers from the subprime crisis was, at least for me, Ashcraft and Schuermann’s Understanding the securitization of subprime mortgage credit (FRBNY) that detailed the market microstructure of subprime credit. We need the same for private credit: Adam, Til?

Separate legal entities can obscure economic commonality. A portfolio can look diversified on paper while ultimately depending on a much smaller number of economic actors. My key question is:

• Was legal separateness repeatedly mistaken for economic independence, or was legal separateness deliberately used to manufacture the appearance of economic independence?

The investigation may eventually answer that.

For now, the disclosure failure is real, the financial architecture is increasingly visible. Is this the tip of an iceberg? Or will the producers throw us a plot twist along the way?

Perhaps we need to move the Private Credit warning level up to Defcon 2.

Heat puts Parmigiano Reggiano at risk (translated from dissapore)

Extreme heat threatens Italy’s Parmesan heartland (Reuters)

Rising bills are also hitting managers of the warehouses where cheese wheels are stored during the ageing process for at least 12 ​months, sometimes three years or even longer.

More than 500,000 wheels of Parmigiano Reggiano, together worth more than €300 million, are stored in the two warehouses operated ​by Credito Emiliano (EMBI.MI), opens new tab unit Magazzini Generali delle Tagliate (MGT) in the provinces of Reggio Emilia and Modena.

A Bank That Takes Parmesan as Collateral: The Cheese Stands a Loan (HBS)

Since 1953, the regional bank Credito Emiliano has accepted curious collateral for small-business loans: giant wheels of Parmigiano-Reggiano cheese.

Known locally as Credem, the bank is the subject of a new Harvard Business School case study, "Credem: Banking on Cheese."

Due to the slow time to market, it’s essential that farmers have access to credit lines. However, it’s understandable why lenders might hesitate to grant loans to the cheese producers.

A Credem subsidiary, Magazziini Generali delle Tagliate, keeps the pungent collateral in two bank-owned warehouses that offer storage capacity for 440,000 80-pound wheels of cheese. MGT’s warehouses sport state-of-the-art climate controls and a staff of trained inspectors.

Lest students think the collateral model is totally risk-free, the authors of the case included an extreme cautionary tale—the so-called Salad Oil Swindle of 1963, in which crooked trader Tino de Angeles used his soybean oil inventory as collateral for huge loans from several Wall Street banks. Inspectors regularly conducted inventory checks, but the tanks contained mostly water, with just enough oil floating on top to fool them.

More price risk than spoilage.

Disclosure: long ~11 oz. Parmigiano-Reggiano

Three documents were published in the same ten days of June: Beijing’s white paper on global governance, the G7’s reply, and a Council on Foreign Relations essay on why rules-based orders fail. Read together, they confirm the thing my bloc data has been saying for five posts:the China pole is built on institutions and arms, not money. And an order nobody will pay for doesn’t get a successor. It gets erosion.

While the world was watching the wars in the Middle East and Ukraine, the State Council Information Office published a 37-page document with the title More Just and Equitable Global Governance: China’s Principles, Proposals and Actions*. It is Beijing’s bid to be the architect, not just a participant, of the next international order. The white paper claims the project already has a constituency:

Upon its introduction, it swiftly gained support from nearly 160 countries and international organizations, with over 60 countries joining the Group of Friends of Global Governance. [link added]

The same week, the G7 published its own statement of how the world should run, and Benn Steil at the Council on Foreign Relations published an essay, Why Rules-Based Orders Fail, arguing that the whole idea of a rules-based order resting on its own rules is a category error. Put the three side by side and you get something better than any one of them: a primary-source confirmation of the central finding of this whole series, written by the party it should embarrass.

I have spent five posts arguing that the world has sorted into two blocs that are built very differently. The American bloc binds on security and finance. The Chinese bloc binds on trade and institutions and arms sales, and conspicuously not on money. In the bloc paper, when I ran the alignment data through a principal-components decomposition, arms imports loaded at 0.69 on China’s axis and US Treasury holdings loaded at −0.07. Translation: tell me a country buys Chinese weapons and joins Chinese institutions and I can place it. Tell me where it keeps its reserves and I learn nothing about its China tilt, because the answer is almost always “the dollar.” I called this the Eichengreen null, after his work showing that financial blocs in the 1930s followed the flag rather than leading it.

The white paper is Beijing writing down the Eichengreen null hypothesis in its own hand.

A 37-page bid to run the world that asks for a bigger seat at someone else’s table

A document about reshaping global governance, from the country that supposedly wants to displace the United States, does not propose a single major new financial institution, currency, or transfer. What it proposes instead is more votes inside the institutions that already exist:

the World Bank should conduct shareholding reviews and the International Monetary Fund should carry out quota share realignment in accordance with the agreed timeframes and roadmaps to address the democratic deficit in global financial governance.

Read that as what it is: China is not building a rival to the IMF, it is asking for a larger share of the IMF. On trade it goes further and volunteers to give something up:

… has announced that it will not seek new special and differential treatment in current and future negotiations at the WTO.

And it locates the problem with the present order not in the rules but in their enforcement:

Confrontation and injustice in today’s world do not arise because the UN Charter is outdated, but because it is not effectively implemented.

This is a reformer’s document, not a revolutionary’s. It wants a bigger seat at the dollar table, not a new table. Yu Jie at Chatham House, in the sharpest of the week’s commentaries, put her finger on what is missing:

It speaks extensively about principles, cooperation and institutional reform. But there are no major new financial commitments to help realize these ambitions.

She is right, and you can put a number on it.

Normative power, quantified: about four hundredths of one percent of GDP

Every architect of an international order has paid for the privilege. The Marshall Plan ran about $13.3 billion from 1948 to 1952, roughly 90% of it grants — close to 5% of a single year’s American GDP, better than 1% of GDP a year for four years. That is what underwriting a bloc looks like.

So I did the arithmetic on what China actually spends across borders on public goods, using the white paper’s own figures where it gives them and verified assessments where it doesn’t. The recurring, grant-equivalent total comes to something like $6 to $10 billion a year:

Against a Chinese economy of about $19 trillion, that is 0.03 to 0.05% of GDP. Even if you throw in the gross flow of New Development Bank lending, $42.9 billion cumulative since 2015, which is loans at roughly market rates in dollars, not transfers, you stay under a tenth of a percent. For comparison, even the United States, in the middle of gutting its own aid agency, still runs foreign assistance at something like 0.22% of GDP. The Marshall Plan was twenty-five to forty times China’s current intensity. Even diminished American aid is five to seven times it.

Yu Jie’s framing is the correct one, and it is worth quoting at length because it is the whole argument in three sentences:

Viewed through this lens, the white paper is less about financial hegemony and more about projecting normative power... Beijing is not proposing a Marshall Plan 2.0, nor is it offering to underwrite a global order through vast financial transfers or open-ended security guarantees... What remains uncertain is whether Beijing is willing — or able — to match its growing normative ambitions with the economic statecraft needed to realize them.

This is exactly the picture the financial-bifurcation posts in this series have been building. In Post 5, Dollar Debts, Yuan Revenues, the leverage China holds over the developing world runs through dollar loans it made through the Belt and Road: eighty-seven percent of that lending was denominated in dollars, not yuan. In Post 6, the asset side of the system would not move even after the largest reserve freeze in sixty years, because there is no liquid alternative that escapes the same jurisdiction. China spends 0.04% of GDP abroad and lends in dollars. That is not the balance sheet of a country building a financial bloc. It is the balance sheet of a country that has decided not to.

The G7 answered in the language of security, because that is the language it speaks

If the white paper tells you what the China pole is made of, the G7 statement tells you the same thing about the Western one by what it chooses to talk about. The document runs roughly 85% security and 15% economics. Ukraine, the Trump-brokered Iran deal, the Strait of Hormuz, Lebanon, Gaza, North Korea … and then one paragraph, near the end, on the world economy.

The binding clause is about China without naming it as the adversary:

We reaffirm our opposition to any unilateral attempts to change the status quo, in particular by force or coercion, in the East and South China Seas and across the Taiwan Strait, which should only be resolved peacefully through dialogue.

That is the Western bloc drawing its boundary on the security dimension, which is precisely the dimension my data says the Western bloc binds on. The one economic paragraph is almost plaintive by comparison:

We reaffirm our common interest in converging with other large economies on the causes of large and persistent global imbalances and on the need to address them.

“Large and persistent global imbalances” is diplomatic for China’s surplus, and the G7 punts it to the G20 under the American host year. This is the cooperative-rebalancing path I flagged in the trade posts as the alternative to bifurcation: the G7 would rather negotiate the imbalance down than sort away from it. Two poles, each binding on a different dimension: the West on security, China on institutions, and both of them only contesting economics rather than committing to break it. The domain divergence I built Post 2, How Countries Pick Sides, around is right there in the two documents’ tables of contents.

Why neither side can anchor the order it claims to want

This is where Steil’s essay does work the other two can’t. His piece, Why Rules-Based Orders Fail, is not about China at all. It is about a deeper problem, and it happens to explain why the white paper’s whole project is built on a foundation of sand.

Steil’s argument leans on Gödel’s Incompleteness Theorems and Arrow’s Impossibility Theorem, which sounds like a stretch until you see where he takes it. A rules-based order, like any system elaborate enough to govern its own operation, eventually meets questions its rules cannot answer: who decides the exception, who interprets the rule about interpreting the rules. So in practice the order never ran on its rules. It ran on a power willing to stand both inside and outside them:

In reality, the system depended heavily on the one country capable of operating simultaneously within and beyond the rules. The erosion of American predominance, and the rise of China as a near-peer competitor, has shattered that sensitive equilibrium.

Big words, simple point. The rules never enforced themselves. A hegemon enforced them, and paid for them, and broke them when it judged it had to, and the order held as long as everyone believed the exceptions were temporary. That is the thing the white paper does not grasp. Beijing thinks the problem with the order is that the rules are applied unfairly and the cure is more inclusive rules. Steil’s second move, borrowed from Arrow’s impossibility theorem, says the cure is the disease:

as a rules-based regime becomes more inclusive and diversity of preferences increases, coherent and broadly legitimate outcomes become harder to sustain.

His example is the one that should give the white paper’s drafters pause. The General Agreement on Tariffs and Trade worked, it cut industrial tariffs from about 35% in 1947 to 4% by the early 1990s, partly because it excluded the Soviet Union and so held together a club of broadly compatible economies. Then the United States insisted on a universal World Trade Organization and “assented to China’s admission before it had demonstrated adherence to core market principles,” and the WTO has been paralyzed ever since by the collision of incompatible economic models. More inclusive, less coherent. The white paper’s central demand — universalize the order, give everyone a vote, let the Global South in — is, in Steil’s frame, a recipe for an order that cannot decide anything.

The order doesn’t get a successor. It gets erosion.

Put the three together and the conclusion is not the one either side is selling. China is not the rising hegemon about to take the keys; it spends 0.04% of GDP abroad and asks for a bigger share of the dollar institutions it supposedly wants to replace. The G7 is not a confident bloc; it is assertive on security and almost apologetic on the economic imbalance that is actually driving the fragmentation. And Steil tells you why there is no clean handoff coming: the order was never the rules, it was the hegemon behind them, and the hegemon is now invoking the rules selectively itself.

The institutions remain, but they have been drained of authority. The rival major powers invoke rules selectively, interpret them opportunistically, or ignore them outright. The exceptions are no longer hidden in the background... They have moved to the foreground. And once seen, they can never be unseen.

This is the deep reason the contested middle of my alignment maps is metastable rather than sorting. I have made the point from the flows side, that the dollar architecture holds because, as the companion Three Clocks piece argues, the slow clock of reserve incumbency barely ticks while the fast clock of capital flight screams. The white paper is the same point from the other side. The architecture holds because there is no successor, and the white paper is Beijing telling you, in 37 pages, that it is not auditioning to be one. The renminbi sits near 2.3% of global reserves. China lends in dollars. It wants IMF votes, not an anti-IMF. A weak dollar with no replacement is not the changing of the guard. It is the guard staying put for lack of a relief.

But “no successor” is not the same as “no risk,” and this is the part that should worry you. The interwar parallel that opens the bloc paper, — the one where the open system of the 1920s fragmented into discriminatory blocs in the 1930s, did not feature a clean succession from sterling to the dollar either. It featured both incumbents and challengers invoking the rules when it suited them and abandoning them when it didn’t, until the rules meant nothing and the blocs hardened by default. That is the Steil scenario, and it is the one the three documents jointly describe: not China winning, but everyone defecting from the rules at once while the institutions stand hollow. An order erodes faster than it is replaced. You do not need a new hegemon to lose the old order. You only need the old one to stop underwriting it and the challenger to decline the bill — which is precisely what these three documents, read together, show both of them doing.

Three things, each a test of whether the erosion reading or the succession reading is right:

1. Whether China ever writes a check that contradicts the 0.04%. A genuine bid for financial leadership would show up as a real number — a development institution at Marshall-Plan scale, yuan lending that displaces the dollar loans, a reserve facility anyone actually draws on. So far, every Chinese financial move is denominated in the system it claims to be reforming.

2. The “global imbalances” paragraph. If the G20 under the American host year actually convenes a rebalancing negotiation with Chinese participation, the cooperative path is alive and the blocs are contesting economics, not breaking it. If it dies quietly, that is a tell the other way.

3. Which way the ~160 endorsers lean when it costs something. Endorsing a white paper is free. The Group of Friends of Global Governance can sign Beijing’s document and keep their reserves in Treasuries and their security under American guarantee, and most of them do. The number to watch is not how many endorse the vision but how many move a material variable — reserves, arms, a base — to match it.

TL;DR

China just published its plan to lead the world and forgot to fund it, because funding it was never the plan. The bid is for normative power on the cheap: more votes, more institutions, more “Groups of Friends,” and about four hundredths of a percent of GDP. That confirms what the data in this series has said all along: the China pole is institutions and arms, not money, and the dollar architecture holds because nobody is paying to replace it. The danger is not that Beijing wins. It is that Washington stops underwriting the order, Beijing declines to pick up the check, and the rules quietly stop meaning anything while the buildings still have the old names on them. That is not a handoff. That is the 1930s.

Bill Gross had a piece in the Financial Times, America’s hegemonic glory is under threat, and the line that stood out to me is:

Inflation cannot be the upwards force here — more likely hegemonic decay and worries over future government liabilities are a major cause.

He is talking about a 30-year real yield that has climbed three percentage points since January 2022, a trade-weighted dollar down 10% over eighteen months, and a long bond he thinks is still expensive at 5%. The story he tells is the classic one: a hegemon loses free trade, loses a strong currency, and the bond market figures it out. Britain to America, America to whoever’s next.

I want to use Gross as a foil, because reading him alongside the four globalization pieces that crossed my desk forced me to confront something.

I have written two things about how financial globalization comes apart, and a careful reader could be forgiven for thinking I can’t keep my story straight.

In the Jan. 28 Carney post I argued that finance fragments first; that capital retreats before trade does. In the bloc-formation work I’ve been building out in the Globalization series, I argued the opposite: that finance fragments last; that the dollar system is the slowest, stickiest layer of the whole arrangement, the one that bifurcates long after trade and technology have split.

Both claims are in my own files. They cannot both be the headline, right? So which is it?

It turns out they are both right, because they are about two different objects. One is about flows. The other is about architecture. Once you separate the two, not only does the contradiction dissolve, but the gap between them tells you exactly where the risk in the dollar system resides, and it lets you say something sharper to Gross than “you’re wrong.”

The flows are the fastest thing in the system

Start with the claim that finance leaves first. The cleanest evidence is a Bank of England working paper, Geopolitical risk and cross-border bank lending, which finds that a one-standard-deviation rise in firm-level geopolitical risk cuts cross-border bank lending by roughly 4% within a year. That is fast. Trade contracts respond on the timescale of supply-chain requalification and shipping; a bank’s decision to roll a cross-border credit line responds on the timescale of a risk committee meeting.

This is the sense in which I wrote, in January, that capital retreats before trade. Gross-border positions — bank lending, portfolio holdings, the hedge ratios sitting on top of foreign equity books — are the most mobile claims in the world. They can be unwound with a keystroke. When a country’s risk premium moves, the marginal lender pulls back long before the marginal exporter finds a new buyer.

The Federal Reserve’s own fragmentation work tells the parallel trade-side story and confirms the ordering.1 A one-unit increase in geopolitical distance is associated with about an 8% decline in bilateral trade: real, but it shows up over years, not quarters.

Capital is twitchy. Trade is the slow one.

So on the flow of capital, finance is the leading indicator of de-globalization, not the lagging one. That January claim stands.

The architecture is the slowest thing in the system

Now the other claim. In the bloc-formation paper I wrote, flatly:

Financial bifurcation is proceeding more slowly than trade or technology bifurcation, but the trend is discernible.

And in the conclusion, ranking the layers by speed: sorting is “fastest in technology and slowest in finance.” That is the exact opposite of “finance leaves first” — unless the word “finance” is pointing at something different, which it is.

Here it points at the architecture: the dollar’s job as the world’s invoicing currency, funding currency, and reserve asset; the payment rails; the safe-asset complex that sits underneath all of it. That structure does not move on a risk committee’s timescale. It moves on the timescale of network effects and switching costs, which is to say, barely.

The single best piece of evidence landed in January from the Bank for International Settlements. BIS Paper No. 165, Dollarization Waves: New Evidence From A Comprehensive Bond Database, built the most complete database of international debt securities to date and found that the dollar’s share ran near 60% in the early 2000s, fell to about 43% by 2008, and climbed back to roughly 60% by 2024. The authors’ summary of the whole sweep is one line:

Plus ça change, plus c’est la même chose!

What looks like a secular trend inside any given window — the euro’s rise into 2008, the post-2008 retreat, the softening after Liberation Day — turns out, in the long sample, to be one segment of a wave. The architecture oscillates around a remarkably stable center.

History rhymes with this. Britain’s share of world trade peaked in the 1870s and declined for six decades, but sterling’s share of global reserves did not drop below 50% until the late 1940s and did not reach its modern floor until the 1970s. The lag between trade decline and financial decline ran 70 to 100 years. The plumbing is the last thing to go.

The mechanism that makes the architecture this sticky is the one I keep coming back to, and it’s the thing Gross’s piece misses entirely. From the bloc paper:

the dollar’s role as universal intermediary breaks the bilateral link between trade and capital: surpluses earned in China trade are automatically recycled into dollar assets through the global payments infrastructure.2

A Vietnamese exporter sells to China, gets paid in dollars or converts to them, and parks the surplus in the US Treasury market. The trade relationship is with China; the financial relationship is with the United States. This is why the alignment maps show something with no historical precedent: a dozen-plus countries that trade more with China than with America yet keep their savings in the dollar system. Trade gravity pulls east; capital gravity pulls west; the dollar’s intermediation role lets both happen at once. I called the result a metastable equilibrium; “stable so long as the US exercises restraint in weaponizing the financial chokepoints that sustain it.”

So the architecture claim also stands. Finance-as-plumbing leaves last.

The reconciliation is stock versus flow

The two claims were never in tension. They are about two different layers moving on two different clocks.

Flight capital and reserve status are not the same variable, and they do not even point the same direction at the same moment. The fast clock can be screaming — bank lending pulling back, hedge ratios flipping, the dollar selling off — while the slow clock barely ticks. That is precisely the configuration we are in.

I originally wrote that table with two rows. The 2025 episode added a third.

The third clock the rare-earth episode revealed

The two-clock reconciliation was tidy. Then April 2, 2025 happened.

Liberation Day tariffs launched a sequence that played out on a timeline that fits neither the fast clock nor the slow one. China’s response — restricting rare-earth exports, the inputs to defense electronics, EV batteries, and advanced semiconductors — did not work in the quarter-to-quarter register of flow signals. It worked in months. Seven months after Liberation Day, at the Korea summit in November 2025, the US accepted a framing of “constructive strategic stability” that Beijing had chosen. Stalemate on China’s terms.

That is a third clock: the supply-chain cascade. The mechanism is distinct from either of the first two. It doesn’t operate through risk premia or through the network effects of reserve status. It operates through physical chokepoints, the intermediate goods that the US defense and technology industrial base cannot readily substitute, and it resolves on a political timeline once enough sectors are under pressure simultaneously.

I can calibrate this against the bloc-formation data. Using the BRS cascade framework I applied in Dollar Debts, Yuan Revenues: Testing the BRS Paper Against Our Demographic Framework over in my other substack, the supply-chain cascade fires when a network penetration rate (Λ) exceeds twice the exposure heterogeneity parameter (2h). For the yuan-debt cascade, the one about sovereign creditors diversifying into renminbi. Λ_CN ≤ 0.043 per year and the threshold is 0.18–0.30. That cascade is 4–7 times away from being met. For the supply-chain cascade, calibrated from the seven-month episode: Λ_US ≈ 1.03 per year. The asymmetry ratio is roughly 20 to 24 times.

Put that differently: China’s yuan-debt leverage runs on a years-long fuse and hasn’t cleared the threshold. China’s supply-chain leverage runs on a months-long fuse and cleared it in 2025. The same model, two very different clocks.

The cascade also has internal structure worth understanding. Cross-country trade exposure within the US-aligned bloc is right-skewed, not uniform. Australia, New Zealand, Japan, and South Korea carry China trade shares of 20–30%. The European members of the same alliance cluster at 5–13%. A supply-chain shock activates the Pacific tier first, then propagates to European allies through indirect dependencies — semiconductor components, defense electronics, battery inputs — over a lag of weeks to months. This is precisely the sequence the rare-earth episode traced: Pacific supply chains stressed immediately, European exposure arriving on a lag, diplomatic accommodation at month 7. China can calibrate this. It can activate the Pacific tier, observe the US response, and hold the European propagation leg in reserve.

Gross finds a weak dollar and a high real yield as the bond market pricing hegemonic decay. But a falling exchange rate is a ‘flow’ signal, and reserve-currency incumbency is an ‘architecture’ fact, and the two are running on different clocks. Reserve share has slid from about 71% in 2000 to roughly 58% today, but most of that went into other Western currencies, not the renminbi, which has stalled near 2.3% of reserves. There is no successor. A weak dollar with no replacement is the fast clock moving while the slow clock holds. That is not “hegemony under threat.” That is the architecture doing exactly what the BIS says it has always done.

The slow clock has a demographic engine

Here is where the globalization story has to borrow from the demographics story, because the reason the slow clock is so slow is not only network effects. It is the lifecycle, which is the slowest clock of all.

The Demographics & Capital Flows series keeps surfacing the same fact from different angles: aging populations are a structural, multi-decade bid for exactly the kind of asset the dollar architecture supplies. The demographic signal on external balances loads almost entirely on the income balance, returns on accumulated foreign assets, not on trade:

Income balance: Z₁ = 41.4*** … Trade balance: null.

In plain terms, aging economies don’t move capital by shifting what they import. They move it by accumulating foreign financial claims and living off the returns. And the claims they accumulate are specific: the demographic effect runs through portfolio debt, with foreign direct investment showing no demographic signal at all across every specification. Aging investors want liquid, safe, arm’s-length instruments. That is a description of the US Treasury market.

The demand is also self-reinforcing in a way that flatters the incumbent. Aging simultaneously raises the demand for safe assets and erodes their supply, because the same fiscal pressure that comes with old-age dependency threatens the credit standing of the issuers who manufacture safety. My safe-asset work projects the number of genuinely safe sovereign issuers falling from around two dozen today toward roughly fourteen by mid-century. When the safe-asset menu shrinks, the deepest, most liquid issuer left standing captures a rising share of a rising demand. The bloc paper’s gravity projections show this directly: capital flows toward the US widen their advantage through 2044 even as the US share of global GDP falls. Exorbitant privilege survives the demographic transition.

This is the deep reason the architecture outlasts the flows. Geopolitical flight is a one-year pulse. The demographic bid for safe dollar assets is a thirty-year tailwind. As long as the slow clock is powered by the lifecycle, the fast clock can panic without the structure giving way.

But the supports under the architecture have gone contingent

If I stopped there, the post would be a tidy “don’t panic, the dollar is fine” piece, and I don’t believe that either. The update is that the architecture is still standing, but what is holding it up has quietly changed from structural to contingent. The slow clock now rests on a set of fast-clock props.

For a generation, the marginal Treasury buyer was foreign officials: central banks and sovereign funds recycling trade surpluses, price-insensitive, sticky. That bid has thinned. What has grown to fill the gap is a different animal.

Start with who actually owns Treasuries now. As Gillian Tett summarized recent New York Fed work in the FT:

The Cayman Islands is in fact the largest foreign holder of US Treasury securities — holding significantly more than China, Japan and the United Kingdom. Indeed between 2022 and 2024, hedge funds absorbed 37 per cent of net issuance of notes and bonds — nearly the same amount as all other foreign investors combined.

The marginal holder of the world’s safe asset is now a leveraged basis-trade book in an offshore jurisdiction, sized to specific carry conditions. That book has flushed twice in six years, March 2020 and again after Liberation Day in April 2025. A price-insensitive central bank does not flush. A levered hedge fund does.

The same shift shows up at the high-quality end of the credit market. When the Iran war broke out, the textbook reaction would have been a flight into Treasuries. Instead, as the FT reported, supranational dollar bonds — the European Investment Bank, the World Bank, KfW — rallied to within a hundredth of a point of Treasuries. A Mizuho strategist called them “the golden child,” because they “let investors stay in the US currency while minimising exposure to idiosyncratic US risk.”

Read that again. The marginal safe-asset buyer is no longer fleeing into Treasuries during a crisis; it is fleeing out of Treasuries while staying in dollars. That is the architecture holding and one of its core supports cracking, at the same time.

And then the hedge ratios, which is the most mechanical of the three. The old practice was to hold US equities unhedged, because the dollar reliably rose when stocks fell, a free hedge bundled into the currency. That covariance has broken. As Katie Martin put it in the FT:

Since dollar hedging involves dollar selling, this can very easily become self-perpetuating.

Foreign managers are not selling their S&P 500 exposure. They are buying currency hedges on top of it, and since hedging the dollar means selling the dollar, a passive two-decade-long implicit long-dollar position is now unwinding mechanically. That is a chunk of the dollar weakness Gross reads as decline, and it has nothing to do with reserve status and everything to do with a hedge ratio adjusting.

None of these three is a structural buyer. Levered basis books, supranational substitutes capped at maybe $80 billion a year against trillions of Treasury issuance, and a hedge-ratio unwind are all flow phenomena. The architecture is intact, but it is increasingly propped up by things that behave like the fast clock, not the slow one.

A ‘metastable core’ on contingent supports is how phase transitions happen

Put the two halves together and you get the actual risk, which is neither “the dollar is fine” nor “hegemonic decay.”

The dollar’s incumbency is metastable: stable against small shocks, vulnerable to a large coordinated one. The demographic bid is real and keeps the slow clock slow. But the supports that bridge the gap between today’s thinned official demand and tomorrow’s structural demand are contingent, fast-moving, and correlated; they tend to fail together, in a crisis, which is exactly when the architecture is supposed to be the thing that holds.

And the drift is already happening, behind the cover of diplomatic management. I tracked 21 contested-middle countries, those with alignment scores close enough to the center to still plausibly go either way, across the stalemate window of 2022–2024, against the prior period 2018–2021. Mean annual drift shifted from +0.0015 (marginally toward the US) to −0.0102 (toward China). The reversal is statistically significant. Fourteen of 21 contested countries are drifting China-ward during the stalemate. Three that were actively moving toward the US in the earlier period — Egypt, Thailand, South Africa — reversed direction. The only contested heavyweight clearly accelerating toward the US is South Korea. India, the largest contested economy, has stalled.

That is the setup the models warn about: a system that sits near a bifurcation, drifts for years, and then tips fast once a shock pushes the return to defection above the switching cost. The renminbi does not need to win for this to bite. China still fails the test that matters, what I’ve called, after Miao Yanliang, the credibility chokepoint, and what Karthik Sankaran sharpened into the right diagnosis:

the real obstacle to the internationalisation of renminbi debt markets may be less the absence of the rule of law and more the absence of the rule of accounting.

You cannot fabricate twenty years of audited issuer disclosure in a crisis. So the danger is not displacement by a rival. It is that a sorting event — a Taiwan crisis, a broad financial embargo, one more weaponization of the clearing system in the Farrell-Newman sense — flushes the contingent supports simultaneously, and the slow clock is suddenly asked to do work that its thirty-year demographic engine cannot deliver in thirty days.

The gravity models project drift; history suggests crisis. The flows tell you when the system is under stress. The architecture tells you whether it breaks. Right now the flows are stressed and the architecture is holding, on supports I, and it seems many market participants, would not want to ean on in the next genuine panic.

What I’m watching

Four observables, each keyed to one clock:

1. The basis trade, for the fast clock. A third hedge-fund Treasury flush in six years, under sustained stress, would tell us the marginal Treasury buyer is genuinely contingent, not structural.

2. Pacific trade concentration, for the middle clock. Australia, New Zealand, Japan, and South Korea carry China trade shares of 20–30%, 3–6× the European members of the same alliance. Watch whether that gap is narrowing (decoupling working) or the European end is rising to meet the Pacific level (entrapment spreading). Right now, neither: the gap is structurally persistent. A compression of that spread would mean the supply-chain cascade’s sequential structure is collapsing into a simultaneous one, which is both more dangerous and harder to calibrate against.

3. SSA spreads going through Treasuries. If even one European Investment Bank or KfW bond prints inside the Treasury curve in primary, the “stay in dollars, escape Treasury risk” trade has stopped being subtle.

4. Foreign hedge ratios, for the slow leak. The conversion of unhedged to hedged foreign holdings of US equities runs on a multi-year clock and is, mechanically, dollar selling that has nothing to do with anyone’s view of American decline.

The Bottom Line

I wrote that finance fragments first and I wrote that finance fragments last, and both are true, because flows and plumbing run on different clocks. The 2025 rare-earth episode added a third clock between them, the supply-chain cascade, which runs in months and proved in real time that it can force political accommodation before either of the other two mechanisms has time to register.

The flows have already left. The plumbing hasn’t, and the lifecycle is paying it to stay. But the middle clock is now in motion, the contested-middle is drifting China-ward behind diplomatic cover at a statistically significant pace, and the sequential cascade structure means China can graduate its pressure — Pacific allies first, European propagation on a lag — without triggering the full-architecture confrontation the slow clock would require.

The honest worry is not that Gross is right about hegemonic decay. There’s still no successor, and the demographic bid for safe dollar assets is the strongest structural support the dollar has. The worry is that the props now bridging the gap are contingent enough to fail all at once, the middle clock is running faster than either the fast or slow clock narratives capture, and the next activation of the rare-earth lever doesn’t need to be a surprise to work.

Watch the fast clock for the warning. Watch the slow clock for the verdict.

Brad Setser and I co-authored a post China’s Customs-BOP Gap Is Not a Customs Problem on the CFR’s Follow The Money blog. Geeky trade statistics.

Global Trends 2045

Long time collegues and followers of this substack know I tend to be enamored of the Global Trends series of reports. I wrote about these reports in Perspective on Risk - Nov. 29, 2024 (Global Trends 2040). These reports were published every four to five years, taking a highly informed view of the world 20 years in the future.

The National Intelligence Council brought together experts from within the government (Central Intelligence Agency, Department of State, Department of Energy, US Agency for International Development, Department of Treasury, Department of Defense, National Security Agency, and Office of the Director of National Intelligence) and a bevy of outside experts (inluding Brookings Institution, Center for a New American Security, Center for Strategic and International Studies, Council on Foreign Relations, The Economist Intelligence Unit, Institute for the Future, International Institute for Strategic Studies, National Science Foundation, Oxford Economics, Pardee Center for International Futures, RAND Corporation, Royal Institute of International Affairs, Strategic Business Insights, and The Conference Board). The NIC produced seven reports 1997; the last in 2021.

But of course the Trump administration under Tulsi Gabbard eliminated the NIC.1

So since we were do for a 2026 version, I decided to have Claude write one.2 And it did an excellent job.3

Global Trends 2045 - A More Constrained World

Here is a link to the full report and a brief side deck that contains an overview (also produced by Claude). You can tell Claude wrote it if you are familiar with its language.

It opens by refusing the thing most forecasts are built to sell:

The world of 2045 cannot be forecast, and this assessment does not try to [and instead] widens the aperture on the plausible futures rather than naming one.

How the report was built.

Before projecting twenty years forward, the Claude “audit[ed] the instrument,” scoring seven prior NIC editions (1997–2021) against what actually happened, on the logic that

the error modes that recur across editions are the ones most likely to recur in this one.

The recurring failure was not picking the wrong category but mispricing it: the series

repeatedly named the right shock category but quarantined it in a low-probability annex.

As a result, it independently made two explicit decisions:

• First, every judgment carries a fixed estimative lexicon (very likely = 80–95%, likely = 55–80%, and so on) with likelihood held separate from confidence - because “Confidence (high / moderate / low) is a statement about evidence quality,” not probability, and conflating the two “is among the most common and most misleading analytic errors.”

• Second, discontinuity is treated as ordinary: over a twenty-year horizon, at least one system-reorienting shock - “financial, pandemic, technological step-change, or major war” = is rated “very likely (80–95%; confidence moderate)” and built into the base case rather than annexed. That single choice, pricing the tail into the center, is the report’s methodological signature.

Six forces are already reshaping the terrain

The report identifies six structural forces.

1. Demographics. Roughly 90% of the people who will make up the world’s working-age population in 2045 have already been born. China, Korea, Japan and much of Europe are aging and shrinking. Sub-Saharan Africa and parts of South Asia remain young and growing.

2. Climate. The report expects the 2045 global temperature to be roughly 1.7–2.0°C above the preindustrial level under a wide range of policy outcomes. The near-term question is increasingly not whether warming can be avoided, but how adaptation, insurance and asset repricing will occur.

3. Machine intelligence and biotechnology. The capability floor is rising, but the economic outcome remains extremely wide. AI could produce a broad productivity acceleration, a narrow concentration of rents, an investment bust or something that overwhelms the report’s conventional scenario structure.

4. The energy-compute-critical-minerals nexus. AI is becoming a power story. Electrification, data centers, grids, storage, advanced chips and critical minerals are no longer separate sectors. They are one connected industrial system.

5. The sovereign balance-sheet squeeze. Public debt is near record peacetime levels just as interest rates, defense costs and aging-related expenditures have risen.

6. Weaponized interdependence. Chips, minerals, payment systems, shipping routes, financial sanctions and reserve assets are now instruments of state power. Globalization is not disappearing, but security increasingly determines where trade and capital can flow.

These forces jointly produce what the report calls the Great Fiscal Vise.

Five rising, largely non-discretionary claims — aging entitlements, debt service after the rate reset, the defense supercycle, climate adaptation, and the energy/grid build-out — bid for one constrained savings pool.4

This is the report’s central thesis.

The post-1990s world rested on abundant savings, cheap capital, falling defense burdens, expanding globalization and limited recognition of climate costs. Governments could borrow, postpone trade-offs and assume that growth would eventually make the arithmetic work.

That world is ending,or at least becoming much harder to rely on.

The same capital cannot simultaneously finance retirement systems, sovereign debt, defense production, climate adaptation, electrical grids and an enormous AI-compute build-out without affecting prices, returns and political choices. The future therefore looks less like a simple contest between the United States and China and more like a global argument over who gets access to capital, energy, compute and fiscal capacity.

AI is the escape valve—and the largest uncertainty

Nearly every scenario in the report turns on whether AI produces a measurable, widely diffused productivity dividend.

Not another impressive demonstration. Not chatbot adoption. Not hyperscaler capital expenditure.

Measured productivity.

The cardinal economic rule is not to bank fiscal or growth plans on the AI dividend — adoption is not the tell, a measured TFP break is.

If AI raises economy-wide productivity enough to keep economic growth above governments’ financing costs, the fiscal vise loosens. Aging becomes easier to finance. Defense and adaptation spending become manageable. Debt ratios can stabilize without inflation or austerity.

If the dividend remains concentrated among a few firms—or if the capital expenditure boom ends in an investment bust—the denominator does not grow fast enough. Governments then have to choose among taxation, spending cuts, inflation and financial repression.

That is why the report treats the technological-economic dividend as the “load-bearing hinge” of the entire scenario set.5 The optimistic outcome is plausible. It is not safe to budget against.

Four worlds - and three ways the framework can be overwhelmed

The report’s central scenario grid has two axes:

• managed interdependence versus hard fragmentation;

• broad technological dynamism versus a stalled or concentrated dividend.

That creates four worlds.

1. Rival Renaissances: roughly 5–12%—combines technological dynamism with fragmentation. AI works, but it works behind walls. The United States and China lead separate technology, finance and industrial systems. Growth continues, but the developing world has difficulty entering either capital-intensive, automated bloc.

2. Managed Dynamism: roughly 8–15%—is the best conventional outcome. AI raises productivity, rivalry remains bounded and growth creates enough fiscal capacity to fund aging, defense, adaptation and the grid. It is internally coherent, but it requires several favorable things to happen at once.

3. Fortress Worlds: roughly 8–14%—combines fragmentation with technological disappointment. Blocs duplicate supply chains, weaponize chokepoints and devote scarce capital to defense. Governments manage debt through captive buyers, controlled interest rates and negative real returns.

4. Hollow Peace: roughly 10–15%—combines managed rivalry with stagnation. There is no major war, but there is also no productivity miracle. Governments preserve stability through mildly elevated inflation, financial repression and gradual institutional decay.

This last scenario may be the least dramatic scenario, but it is arguably the easiest to imagine: no collapse, no renaissance, just a long period of managed underperformance.

The report places another 30–40% in an interior or hybrid zone sharing characteristics of several worlds. That is sensible. Real economies rarely remain at the clean corner of a two-by-two matrix for twenty years.

It then identifies three “off-grid” events that could overwhelm whichever conventional world was developing:

1. Pale Horse: an AI-enabled engineered pandemic;

2. Broken Threshold: a kinetic conflict over Taiwan;

3. The AGI over-top: transformative AI that changes the distribution of power too quickly for the existing framework to remain useful.

The probabilities assigned to these off-grid cases overlap with the conventional scenarios. They should not be added together. They are possible overwrites of the underlying world, not additional cells in the same probability table.

Globalization is being resegmented, not reversed

The capital-flows analysis is the most distinctive part of the report.

Aging surplus economies - including Japan, China, Korea, Northern Europe and the Gulf - continue exporting savings. Young economies in Africa and South Asia need capital but receive relatively little of it. Instead, much of the world’s savings is intermediated through the dollar system and recycled into U.S. assets.

Capital continues to run “uphill.”

AI may make this worse. Automated production reduces the value of cheap labor, weakening the manufacturing ladder that helped East Asia industrialize. Compute, finance and intellectual property concentrate capital in economies that already possess capital.

The young countries remain young. The old countries receive the machines.

At the same time, capital increasingly carries a geopolitical price. Friend-shoring, sanctions exposure, payment-system access and political alignment become part of the required return.

This is not deglobalization. Trade and investment continue, but they are rerouted through trusted jurisdictions, connector economies and bloc-compatible infrastructure.

The report’s description of the monetary order is useful:

Multipolar at the margin, unipolar at the core.

Marginal de-dollarization is likely. Central banks diversify into gold. China expands its payment infrastructure. Bilateral settlement outside the dollar grows.

But wholesale replacement of the dollar by 2045 remains unlikely. The renminbi is constrained by China’s closed capital account and weak rule-of-law protections. No other market matches the liquidity and network depth of the dollar system.

The larger risk is not a gradual reserve transition. It is what the report calls the master switch: a geopolitical or fiscal event that causes foreign investors simultaneously to reduce exposure to both U.S. equities and Treasuries.

A Taiwan conflict would be the obvious trigger. A serious deterioration in U.S. institutional credibility is another. The problem would not simply be Treasury selling or an equity correction. It would be the breakdown of the assumption that the United States can simultaneously provide the world’s primary safe asset and absorb its surplus capital.

That is a genuine tail risk.

On February 28, 2026, the United States and Israel began military operations against Iran. The energy shock that followed sent the Turkish lira to a fresh record low. Turkey’s central bank had already spent $26 billion in foreign exchange reserves against prior exchange-rate pressure. In the weeks after the strike, it deployed 127 tonnes of gold, part of the 220 tonnes it had accumulated since 2022, its largest reserve drawdown on record.

The gold was not sold. Most of the operations were gold-for-currency swap futures, structured so the gold would return to Turkey’s reserves at maturity, while in the interim it backed dollar liquidity in the London market and with local commercial banks. Turkey liquefied its gold without permanently disposing of it.

The scale of that buffer has been repriced. Gold hit a high of more than $5,500 per troy ounce in January 2026, before the Iran war shock and subsequent ceasefire diplomacy introduced two-way price volatility. The ECB’s June 2026 annual report documents the system-level consequence: gold has overtaken U.S. Treasuries as the world’s largest official reserve asset class for the first time, accounting for 27 percent of global central bank reserves against Treasuries’ 22 percent at current market prices. The countries that have been accumulating gold since 2022 did so at prices well below this level. Their reserve buffers are worth substantially more than the purchase cost.

This was the first large-scale operational test of gold as an alternative reserve buffer in the post-Russia-freeze world. A new NBER working paper by Joshua Aizenman, Jamel Saadaoui, Gazi Salah Uddin, and Naoki Yago, US Monetary Spillovers, Foreign Exchange, and Gold Reserves at Times of Geopolitical Fragmentation documents the episode and provides the cleanest empirical evidence available on what gold does in a reserve crisis, and under what conditions it works.

The answer is more precise than most commentary on gold and de-dollarization has appreciated.

Gold Works When Swap Lines Don’t

The paper’s design exploits high-frequency US monetary policy surprises — changes in federal funds futures in a 30-minute window around FOMC announcements, orthogonal to other news — to identify how exchange rates respond to dollar tightening, and then asks how that response varies with a country’s reserve composition. The baseline finding: a 10 basis-point surprise increase in the fed funds rate produces about 0.4% depreciation in the exchange rate of the affected currency. Both FX reserves and gold reserves attenuate this. Per unit of GDP, gold’s buffering efficiency is substantially higher than FX, it just looks comparable in the baseline because central banks hold much less of it.

The authors split their 18-country panel by whether a country has a bilateral dollar swap or repo facility with the Federal Reserve.

For countries with US swap or repo lines — Australia, Canada, the Euro Area, New Zealand, Norway, Sweden, Switzerland — FX and gold reserves have limited effect on exchange-rate stabilization following a US monetary shock. The Fed backstop handles the dollar funding stress. Reserve self-insurance becomes largely redundant.

However for countries without US dollar swap lines, the estimates flip. Dollar reserves offset approximately 100% of the depreciation shock. Gold reserves offset approximately 25%.

Our results suggest that not only the aggregate volume but also the composition of foreign exchange and gold reserves and access to dollar liquidity facilities are empirically relevant for exchange-rate responses to US monetary shocks.

Gold’s value as a reserve asset is not a constant. It is conditional on whether you have a dollar backstop. Inside the network, gold is mostly redundant. Outside the network, gold is the relevant self-insurance instrument. The distinction is binary, and it matters enormously for how to read the past three years of central bank gold purchases.

The Russia Freeze Created a Binary Outcome, Not a Trend

In a prior post in this series, I documented that aging creditor central banks — Japan, Korea, Germany — did not rotate their portfolios after the Russia reserve freeze of February 2022. The prediction that a $300 billion expropriation would trigger broad de-dollarization was wrong. The wealthiest central banks extended duration into higher-yielding Treasuries.

The Aizenman paper explains why. Japan, Korea, and Germany are inside the dollar liquidity network. They have standing Federal Reserve swap lines. For them, the jurisdiction problem, the risk that dollar assets could be politically inaccessible, is substantially mitigated by backstop access. If their reserves came under pressure during a crisis, the Fed would provide dollar liquidity. There is no marginal value from holding an alternative. The null is not complacency; it is rational behavior for a country inside the network.

The freeze did not produce uniform de-dollarization. It produced a bifurcation. Countries inside the network had no reason to change. Countries outside — China, Russia, Turkey, India, South Africa, Poland, Kazakhstan — drew the correct inference from the same event: being deeply embedded in the dollar system without a dollar backstop is a visible and growing vulnerability. Gold is the response.

The post-2022 active gold accumulators — Poland, Kazakhstan, Brazil, China, Turkey — overlap heavily with the outside-the-network cohort. This is not a coincidence. It is the empirical signature of the same rational portfolio decision made independently across multiple jurisdictions, in response to the same demonstration that dollar reserves are political objects, not just financial ones.

The Network Itself Is Becoming Conditional

The Aizenman finding assumes that swap line network membership is stable. Recent reporting suggests that assumption deserves scrutiny.

Kevin Warsh, the new Federal Reserve chair, testified before the Senate that the Fed’s international role “doesn’t come with the same independence as its interest-rate-setting function.” Translation: geopolitics and national security priorities can affect who receives a dollar backstop in a crisis. Separately, Scott Bessent’s Treasury has been deploying its own swap facilities as foreign policy instruments; extending liquidity to the Milei government in Argentina and reportedly offering backstop capacity to Gulf allies during the Iran war. A June 2026 FT piece, The Fed is going to have to rethink its global role, by Atlantic Council vice-president Josh Lipsky reported that European officials have been “gaming out” scenarios where dollar lifelines are cut off during a political dispute with Washington.

This is not a fringe concern. It is being modeled in the finance ministries of countries that currently hold standing Federal Reserve swap lines: countries that, on the Aizenman framework, should have little reason to accumulate gold. Poland has a Fed swap line. It has nonetheless purchased more gold over the past two years than any other European peer. The deviation makes sense once the binary is understood as probabilistic rather than fixed: a country that faces a small but non-negligible probability of finding itself outside the network will rationally self-insure, even from an inside-network starting position.

The PBOC has at least 30 active bilateral swap lines of its own, and counting. They are yuan-denominated, providing yuan liquidity for trade settlement but no dollar backstop capacity. As the Federal Reserve’s swap line network becomes more explicitly geopolitical in its conditioning, the PBOC’s alternative architecture becomes more relevant — not as a dollar substitute, but as the infrastructure available to countries that find dollar network access uncertain. The gold and the yuan swap networks are being built by the same cohort, in response to the same underlying uncertainty.

China Is at the Extreme of Both Dimensions

China sits at the intersection of maximum dollar exposure and maximum exclusion from the dollar backstop.

As I described in the prior post, China’s consolidated dollar exposure runs across formal PBOC reserves, state bank dollar books, policy bank lending, and the China Investment Corporation, a total that Setser estimates may be flat or rising even as the formal PBOC reserve line has drifted down. Dollar-denominated external liabilities, dollar-invoiced trade, and dollar-recycled surplus earnings create an exposure profile that rivals any economy in the world.

China is simultaneously not in the Federal Reserve’s swap line architecture. The PBOC has bilateral swap lines with over 60 countries, but these are yuan-denominated, providing yuan liquidity, not dollar backstop capacity.

The paper finds that for countries at the 90th percentile of dollar external liability exposure, gold reserves reduce depreciation by 0.2–0.3% per standard deviation, five to seven times the all-country baseline estimate. China’s dollar-liability profile sits near the extreme of any comparable cross-country measure. And the gold stock it has been building has a current value that earlier estimates in this newsletter understated: gold peaked above $5,500 per troy ounce in January 2026, making the PBOC’s disclosed 2,345 tonnes worth substantially more than the $241 billion figure derived from late-2025 price data.

At any point in the current price range, China’s gold holdings have crossed all three of the tipping-point thresholds I identified in prior posts — M0 monetary base backing, bilateral swap line coverage, and the threshold at which gold functions as a credible first-response buffer for a balance-of-payments stress episode. The paper’s framework predicts, with credible causal identification, that China’s gold accumulation has considerably more reserve-buffer value than conventional commentary on Chinese portfolio diversification implies.

This is not portfolio optimization in the conventional sense. It is self-insurance against the specific political risk that Russia demonstrated in real time: dollar reserves that are simultaneously essential to an export model and reachable by the same sanctions architecture that moved in 72 hours.

Gold and Dollars Are Complements, Not Substitutes

The ECB’s 2026 annual report puts the system-level evidence in direct terms. Central banks globally now hold 36,000 tonnes of gold, approaching the 38,000-tonne peak of the Bretton Woods era, after 850 tonnes of net purchases in 2025. The single largest buyer in 2025 was Tether, the dollar-denominated stablecoin company, at over 100 tonnes. An entity whose entire business model depends on dollar clearing infrastructure was simultaneously the world’s largest non-central-bank gold accumulator. The complementarity between dollar architecture and gold hedging is not a theoretical observation. It is being built by the dollar system’s most commercially sophisticated participants.

The paper’s complementarity finding cuts against the simple de-dollarization narrative directly. Gold and dollar reserves do not trade off. Countries holding more of both benefit more from both. There is no evidence in the Aizenman panel that building gold reserves reduces the stabilizing value of dollar reserves or vice versa.

This means China’s dual-track reserve strategy, running the sinodollar recycling loop while accumulating gold at the margin, is not internally contradictory. It is the rational portfolio response to being inside the dollar system (you need the dollars to operate the surplus model) and outside the dollar liquidity network (you need the gold because the backstop is unavailable to you specifically).

The gold “leak” in the sinodollar, described in the prior post, is small in flow terms — low single digits of the annual surplus on official figures, running higher if shadow accumulation is included. It is not undermining dollar demand. The Aizenman complementarity result confirms it structurally: the gold purchase is not a vote against the dollar. It is insurance against the political risk of holding the dollar without a political backstop.

What PBOC Swap Lines Cannot Do

The paper contains a quieter finding that bears directly on the yuan internationalization debate. Non-dollar FX reserves, which would include yuan-denominated assets held through bilateral PBOC swap lines, are statistically indistinguishable from zero in exchange-rate buffering against US monetary shocks. The point estimate runs slightly in the wrong direction.

China has approximately $580 billion in outstanding bilateral yuan swap lines with partner countries. Those lines provide yuan liquidity for trade financing, bilateral settlement, and the closed-loop shadow fleet architecture I described in the January 2026 post. That function is real and growing.

What PBOC swap lines do not provide is dollar exchange-rate stability for recipient countries. Countries in the yuan swap network face the same FX stabilization problem as countries with no monetary network access at all. Yuan liquidity covers the trade architecture. It does not cover the dollar-currency-mismatch exposure that makes US monetary tightening dangerous for emerging-market balance sheets.

This is a clarification of where Shin’s ladder currently stands, not a verdict on where it ends up. Invoicing, trade financing, and liability-side debt conversion, Steps 1 through 3, are functioning at scale. Step 4, currency hedging, providing exchange-rate insurance against external shocks, requires the kind of deep, liquid, politically accessible reserve backstop that only dollar assets currently provide. Gold is the one instrument that partially substitutes for that function without requiring the full reserve-currency infrastructure.

Concluding

The Russia freeze created two reserve management worlds. In the first, countries inside the dollar liquidity network continue to hold Treasuries without meaningful change, because the network handles their dollar funding risk and gold adds little on top. In the second, countries outside the network have been systematically building the one asset that buffers exchange-rate risk without requiring network access.

Turkey’s deployment of 127 tonnes in 2026 proved the mechanism is operational, not theoretical. The gold-for-currency swap structure — collateralized dollar liquidity without permanent disposition of the gold — is the playbook. The NBER paper provides the causal identification: gold works precisely for the countries that have been building the most of it.

China is not accumulating gold because it expects the dollar system to collapse. It is accumulating gold because it is the dollar system’s dominant structural prop and simultaneously excluded from the dollar system’s liquidity backstop — and the only rational response to that position is to build the reserve buffer that functions outside the network. As the current account surplus compresses over the 2030s under demographic pressure, the sinodollar flow declines. The gold stock appreciates. The reserve portfolio rebalances passively toward the non-seizable asset. China does not have to decide to exit the dollar system. The arithmetic does it incrementally.

The active gold accumulators and the countries building alternative settlement architecture — yuan swap networks, CIPS, shadow fleet closed loops — are largely the same group, responding to the same underlying reality through two complementary instruments: gold hedges the political risk of dollar-denominated reserve assets; yuan settlement hedges the transactional dependence on dollar clearing. Post-2022, both strategies are accelerating in the same cohort. That is the reserve architecture of fragmentation being built in real time.

A short piece by Robin Harding, Why sinodollars outweigh the petroyuan (FT), does something useful: it cuts through the petroyuan noise with a single analytical distinction that I think is exactly right, connects directly to four years of prior work in this substack (warning: potential confirmation bias here), and illuminates something important about where the dollar’s structural position actually rests.

The concept is worth examining carefully, because it both validates part of what I argued in 2023 and complicates what I argued in late 2025.

The Petroyuan Lacks the Essential Property of the Petrodollar

Harding’s central move is a distinction most commentary on this topic blurs: settlement is not savings*

What made the petrodollar systemically important was not that oil was invoiced in dollars. It was that Saudi Arabia and the Gulf states held the dollars they earned, reinvesting them into dollar assets and creating the offshore dollar liquidity pool that finances global trade, sovereign debt, and aircraft leasing to this day. The petrodollar is best understood not as a payment, but as a dollar of oil profits in search of a home.

The petroyuan, by this standard, doesn’t exist in any meaningful sense. Today’s yuan earners don’t accumulate yuan savings, they spend them. Russia converts its yuan oil revenues into drone and ATV imports. Iran, facing reconstruction after the war, will spend every yuan it earns. Gulf producers no longer run the kind of structural surpluses that created investable pools in the first place. As Harding puts it:

This is a good backdrop to increase the use of the yuan in trade settlements, which is happening fast, but not to create a pool of offshore petroyuan.

Yuan settlement of China’s goods trade has reached 33.5 percent of total in March–April 2026, a record, per PBoC data. That is a real development. But 33.5 percent settlement share is a measure of invoicing and clearing. It says nothing about where the proceeds are parked.

When Saudi Arabia began recycling petrodollar earnings in 1975, the mechanics were constrained by market depth, not by political choice. David Mulford, who ran the Saudi Monetary Agency’s reserve portfolio in those years, later explained why alternatives to dollar assets were effectively unavailable at the required scale: purchases of German bonds, Japanese yen bonds, or Swiss franc notes “were just not possible in the sizes common in the U.S. market.” Settlement created savings in dollars not by design but because no other market could absorb the volumes involved. Today’s yuan earners face the same arithmetic in reverse: they settle in yuan and immediately spend it, because no market at sufficient depth exists to hold yuan savings at scale.

The more important monetary force, Harding argues, is what he calls the sinodollar: China’s relentless accumulation of dollar assets, driven by its current account surplus. China’s economic model requires running a surplus and recycling it into dollar assets. In this reading, the sinodollar is not a rival to the petrodollar, it is its structural successor as the dominant offshore dollar-creation mechanism. The world’s largest goods exporter is simultaneously the world’s largest dollar accumulator.

This Is the Pettis Argument, From the Other Side

For readers who have been following this series since 2023, Harding’s sinodollar concept should ring a bell. In the April 18, 2023 Perspective, I introduced a framework from Hyun Song Shin, articulated in his Odd Lots appearance, to evaluate the threat to dollar hegemony. Shin argues that for a currency to displace the incumbent, it must climb a ladder of reinforcing roles:

1. Invoicing transactions

2. Trade financing

3. Investing and borrowing

4. Currency hedging

Each step creates demand for the currency that makes the next step easier. Most challengers fail because they stall at Step 1: increased invoicing just leads to swapping the new currency back into dollars, because no one wants to hold the alternative. In 2023, I cited Michael Pettis making the same point with characteristic directness:

For the world meaningfully to switch from dollars to RMB, exporters will have to want to hold their accumulated surpluses in RMB and, much more importantly, China will have to give up control of its monetary policy and abandon its surpluses for permanent deficits. It is extremely unlikely that Brazilians will accumulate RMB assets in exchange for its surpluses.

Harding is making the same argument, from the opposite direction. Pettis said: foreigners won’t hold yuan because they prefer dollar assets. Harding says: China itself won’t stop accumulating dollar assets, because its growth model depends on the surplus. The two constraints are two sides of the same argument; the yuan can’t become a savings currency while both the issuer and the recipient prefer dollars for savings.

That argument still holds. But the picture has moved since 2023, and it has moved in a specific direction that Harding’s framing doesn’t fully capture.

The Ladder Has Been Moving — Just Not Where Harding Looks

By November 2025, the analysis in this newsletter had shifted. The November 26, 2025 Perspective documented a wave of data suggesting the RMB was advancing from Step 1 to Step 3, not through the conventional route (exporters choosing to save in yuan) but through a different mechanism: sovereign debt conversion.

Ethiopia was negotiating to convert at least part of its $5.38 billion in Chinese loans into yuan-denominated debt. Kenya had already completed a similar swap, saving $215 million a year in interest costs. Sovereign borrowers like Indonesia and Slovenia announced plans to issue renminbi bonds. Chinese trade credit, which ran 17 percent RMB-denominated as recently as 2021, had flipped to 72 percent RMB as of 2024.

The key insight from that analysis:

If Ethiopia owes debts in RMB, it must earn RMB to service them. This creates a structural demand for the currency that mere trade invoicing does not.

This is a fundamentally different pathway than what Harding’s framework addresses. The sinodollar story is about whether yuan earners will hold yuan savings. The liability-side story is about whether yuan debtors must earn yuan to repay. These two mechanisms are structurally independent. The sinodollar can remain intact — China accumulates dollars, China-bloc borrowers service yuan debt — and both can be simultaneously true.

That is, in fact, what appears to be happening. The sinodollar is real. The liability-side yuan bloc is also real. They are not in contradiction; they are operating on different parts of the monetary system’s balance sheet.

A parallel channel is visible in Chinese commercial bank lending. Dollar loans from Chinese institutions overseas fell from $587 billion in 2022 to $375 billion by late 2025, while yuan-denominated loans rose to $357 billion: the two now nearly equal, a crossing that was unthinkable five years ago. As Karthik Sankaran of the Quincy Institute wrote in FT Alphaville in February 20261, the real obstacle to renminbi bond market internationalisation “may be less the absence of the rule of law and more the absence of the rule of accounting.” That constraint narrows as Chinese lenders standardize disclosure and as borrowers convert dollar obligations into yuan. The BRI’s dollarization is unwinding from within, one refinancing at a time, through a mechanism Shin’s ladder framework did not anticipate.

The Shadow Fleet Complicates This Further

In the January 30, 2026 Perspective, I argued that the shadow oil trade had revealed a third pathway, one that bypasses Shin’s ladder entirely. The shadow fleet’s closed loop pairs yuan invoicing, yuan settlement through CIPS, yuan-denominated trade financing, and insurance through Sinosure, all within a single self-contained system. The argument was that this system “no longer needs to climb Shin’s ladder” because it operates as a closed loop outside dollar clearing.

Harding’s piece implicitly addresses this, and I think he is right to push back. A closed loop can function without climbing the ladder. But it cannot create the offshore savings pools that give a currency reserve status. Russia doesn’t hold yuan; it spends yuan. The closed loop is a transaction architecture, not a savings architecture. The shadow fleet proves the renminbi can work as a payment medium in bilateral sanctioned trade. It does not prove the renminbi can become a store of value at systemic scale.

Both observations are correct at different levels of analysis. The closed loop is real and growing:CIPS hit a record 921 billion yuan daily in March and briefly exceeded 1.22 trillion yuan per day in April following the Iran war shock, two consecutive monthly records. As Bert Hofman, former World Bank country director for China, observed of the shadow fleet’s structural yuan dependency: “Russia can’t use dollars. There’s nothing much else out there that would not have to hit the dollar system at some point.”2 The 193 institutions now bypassing SWIFT entirely, up 40 percent since 2024, are not choosing yuan as a preference. They are held in it by the same sanctions architecture that removed the dollar option. But Harding is right that this activity, at its current scale, does not build the reserve pool that the petrodollar built. It is a functional bypass; it is not yet a structural replacement.

The Sinodollar Is the Last Structural Prop

In the April 30, 2026 Perspective I argued that the dollar’s incumbency in 2026 is being held up by supports that are structurally different, and more contingent, than what held it in place a decade ago. The foreign official sector held 46 percent of Treasuries in 2008; it holds less than 30 percent today. Hedge funds in the Cayman Islands absorbed 37 percent of net Treasury issuance in recent years. Dollar SSA spreads have compressed to near-zero as a Treasury substitute. The Swiss franc hit a 14-year high against the dollar driven by foreign asset managers adjusting hedge ratios: not de-dollarization, but a mechanical unwind of passive dollar longs implicit in unhedged S&P holdings.

Each of those is a real market mechanism. None is structural in the way that foreign official demand for Treasuries was structural for the previous generation.

What Harding’s piece adds to that analysis is the identification of the one remaining structural prop: the sinodollar. China’s CA surplus recycled into dollar assets is the support that the contingent props are sitting on top of. It doesn’t show up in COFER because Brad Setser has documented that China moved its dollars — SAFE’s formal reserves fell, but state commercial banks absorbed over a trillion in dollar claims, the policy banks hold another trillion in dollar lending, and the China Investment Corporation holds $450 billion. As Setser put it: “Nice little trick. It seems to have fooled most of the internet.” The dollars are there. They just changed nameplates.

A May 2026 FT piece, America needs to put the renminbi back on the international agenda, by Mark Sobel, Brad Setser, and Robin Brooks put the system’s scale in direct terms: China’s manufacturing surplus is “close to 1 per cent of world GDP — a much bigger surplus than any single country has run in the last 70-plus years.” On their assessment, the reported current account surplus of 3.7 percent of GDP in 2025 understates the true figure, with state commercial bank purchases of dollar assets serving as the de facto sterilization mechanism. Setser’s February 2026 FT piece documented the operationalization: state banks purchased an estimated $100 billion in foreign exchange in December 2025 alone, a record monthly figure, with another $70 billion in January 2026. These are not passive accumulations. They are deliberate interventions designed to prevent RMB appreciation from compressing the surplus model that generates the sinodollar in the first place.

The sinodollar is why the BIS wave thesis holds. It is why, in the April analysis, “the movement remains dollar” even as the bezel gets more crowded.

One Thing Harding Does Not Address

Harding frames the sinodollar as a durable feature of China’s model. I think this is the one place where the analysis needs to be more careful.

China has been a consistent and large buyer of gold for its central bank reserves. This matters because the chain Harding describes — export surplus, earn dollars, recycle into dollar assets — is being partially interrupted at the savings step. A portion of China’s dollar earnings is not being recycled into Treasuries or dollar-denominated claims. It is being converted into gold: a neutral reserve asset, outside any currency system, immune to the kind of freeze that immobilized Russia’s dollar reserves in 2022. On official PBOC disclosures, the conversion rate runs in the low single digits of the annual surplus, with 2023 as a peak year where even the official figure touched the mid-single digits, and evidence that purchases routed through state commercial banks and the Shanghai Gold Exchange run the true rate higher. The direction is deliberate and the pace has accelerated since 2022.

This is not, by itself, a reversal of the sinodollar mechanism. China is still the world’s dominant dollar recycler. But the sinodollar prop has a leak in it, and the leak is the PBOC’s gold desk. The accumulation is not speculative. China holds no US dollar swap line; no Federal Reserve backstop for dollar funding stress. In a reserve framework where buffer value depends critically on network access, gold is the rational self-insurance instrument for a country that is simultaneously the dollar system’s largest structural prop and excluded from the dollar system’s liquidity architecture.

More significantly, the sinodollar has a demographic expiration date. China’s working-age population is contracting. The current account surplus that generates the sinodollar flow is likely near its structural peak. The same demographic forces that our bloc formation work identifies as reshaping geopolitical alignment over the 2030–2044 window will, by compressing China’s surplus, also compress sinodollar creation. The structural prop that is currently holding up the dollar’s incumbency is time-limited in a way that Harding’s framing does not acknowledge.

Summing Up

Harding’s sinodollar concept is the clearest short-form articulation of why dollar dominance is more durable than the petroyuan narrative implies. Settlement is not savings. China’s own model has been, and for now remains, the largest source of structural demand for dollar assets. The 33.5 percent yuan settlement share is real; the reserve accumulation it implies is not.

But the concept also does exactly what good analytical framing does: it clarifies where the real vulnerabilities lie. The dollar’s structural prop is not the Fed’s liquidity backstop, or the Treasury market’s depth, or the SWIFT network’s inertia. It is China’s CA surplus, and China is using that surplus in ways that Kissinger’s Saudi recycling model did not anticipate. Some of it goes into gold. All of it rests on a surplus that will compress.

The analysis in this substack has been that the ladder is being climbed, step by step, through mechanisms Shin’s original framework did not fully anticipate: closed-loop shadow settlement, liability-side debt conversion, and now the gold drain on the sinodollar mechanism itself. Harding’s piece doesn’t change that conclusion. It sharpens the picture of what is holding the current equilibrium in place, and how contingent that equilibrium has become.

Benguria, Rojas and Saffie (BRS) have recently published Geopolitical Fragmentation, Sovereign Debt, and Dollar Dominance.

• Side note: I’ve posted Dollar Debts, Yuan Revenues: Testing the BRS Paper Against Our Demographic Framework over on the Demographics & Capital Flows substack that analyzes the paper through the demographic, rather than geopolitical, lense.

The first four posts in this series posited that the world has sorted into two blocs over twenty years, that Chinese development finance is the most powerful time-varying predictor of which way countries go, that the countries most economically dependent on China still route their financial surpluses through the dollar system, and that this trade-capital divergence is a structural feature, not a transitional lag that demographic gravity alone will not resolve.

What none of those posts addressed is the question underneath the question: if the sorting continues and the mismatch between where goods trade and where money flows keeps widening, what is the specific mechanism by which the financial side eventually gives way? Post 3 ended on the word “metastable” and noted that the transition, when it comes, is triggered by crisis, not drift.

The BRS paper is the first formal model I have seen that identifies a concrete liability-side mechanism for that crisis: not a sudden flight from Treasury securities by central banks, not a collapse in dollar FX volume, but something quieter and more structural; sovereign borrowers converting the debt that China’s own lending program created into the currency that China’s own trade relationships are making more attractive. The BRS paper is worth taking seriously because it closes the loop that Post 3 left open.

Dollar Debts, Yuan Revenues

In March 2025, Kenya converted $4.9 billion in Chinese railway loans from dollars into yuan, saving roughly $215 million per year in interest costs — 4.3 cents on the dollar of face value. The conversion was quiet enough that it barely registered in the financial press. It should not have been quiet.

The Kenyan railway deal is the first concrete data point in what a team of economists at the University of Kentucky, University of Florida, and University of Virginia have formalized as a sovereign debt restructuring cascade. BRS describe a mechanism by which geopolitical fragmentation could erode dollar dominance not from the asset side of sovereign balance sheets, what central banks hold in reserves, but from the liability side: what governments owe, and in what currency.

The asset side receives most of the empirical attention, which is why the BRS paper is worth taking seriously. The liability side is where the map is actually moving.

I spent the last several weeks running their framework against our 46-country geopolitical alignment data from this series. This post is what I found.

China Lent in Dollars

The counterintuitive fact at the center of the BRS argument: the loans China extended through its Belt and Road Initiative and state policy banks are overwhelmingly denominated in dollars. Not yuan. Dollars. Eighty-seven percent by commitment value, according to the AidData dataset tracking Chinese overseas lending from 2000 to 2021.

Think about what this means structurally. China’s principal instrument for building economic dependence across Africa, Asia, and Latin America created dollar liabilities for the recipient countries, not yuan liabilities. Kenya did not borrow yuan to build a railway. Ethiopia did not borrow yuan to build industrial parks. Angola did not borrow yuan to develop oil infrastructure. The net effect of a decade and a half of Chinese development finance was to deepen the developing world’s exposure to the dollar system, not to replace it.

This is the setup for the mismatch. Fragmentation has redirected trade toward China and yuan-linked markets. Kenya’s exports to China grew from 9% of its total trade in 2010 to 16% in 2024. Ethiopia’s from 19% to 31%. Angola’s China trade share sits at 34%. As trade reorients toward China, the revenues that service those dollar loans become increasingly yuan-denominated at origin, even if they clear in dollars. The friction rises. The mismatch is the structural result.

The Mismatch Zone

The BRS framework identifies the mismatch zone as countries where the dollar share of sovereign debt exceeds their effective dollar export-revenue exposure. These are the countries below the 45-degree line in their diagram: dollar debt greater than dollar invoicing. They face a rising real cost of debt service as fragmentation continues, and they are the candidates for following Kenya.

I do not have the Boz, Casas, Diez, Gopinath, and Gourinchas (Patterns of Invoicing Currency in Global Trade, IMF WP 2025/178) data locally, and that gap matters; I will come back to it. Using the China trade share as a proxy for yuan revenue exposure, and merging World Bank International Debt Statistics on the dollar share of public external debt for the 24 developing-economy borrowers in our 46-country sample, seven countries fall clearly into the BRS mismatch zone with meaningful yuan exposure alongside high dollar liabilities:

These are the countries structurally positioned to do what Kenya did: convert existing dollar debt to yuan if yuan refinancing becomes cheaper.

Notice what is not on that list. Kenya itself does not fall below the 45-degree line in the data. Its dollar debt share is 73% and its China trade proxy is 16%, dollar invoicing exceeds dollar debt, by this crude measure. That is not a failure of the framework. It is a finding. Kenya converted not because it was the most structurally exposed country in the sample but because it found a specific deal at a specific moment with a specific creditor. The structurally exposed countries are in the table above, not in Kenya.

The Problem With Angola

Angola belongs at the top of that table in multiple ways. It has the highest cumulative Chinese development finance in our sample, 0.63 times its average annual GDP, accumulated over 2000–2021. Its dollar debt share is 87%. Its China trade proxy is 34%.

But Angola is also the clearest illustration of why invoicing data matters more than trade shares.

Angola exports crude oil. Oil accounts for roughly 60% of its exports. Crude oil is priced in dollars globally and invoiced in dollars regardless of whether the tankers go to Rotterdam or Qingdao. A Luanda exporter selling crude to Sinopec receives dollars, not yuan. The effective yuan revenue exposure is not Angola’s 34% China trade share. It is closer to 34% times 40%, adjusting for the oil tranche — roughly 14%.

At 14% yuan exposure and 87% dollar debt, Angola is still currency-mismatched. But the mismatch is meaningfully smaller than the headline number suggests, and the incentive to restructure is weaker than the table implies. The countries with the strongest restructuring logic are the non-commodity cases: Pakistan (textiles, remittances), Kazakhstan (more diversified than Angola despite mining dependence), the Philippines (electronics, remittances). Their yuan invoicing tracks closer to their trade share.

This matters for the cascade mechanism because that mechanism is hardest to trigger when countries are heterogeneous in their effective yuan exposure. Angola and Nigeria, the two largest African recipients of Chinese development finance, are both commodity exporters whose revenues remain more dollar-linked than their trade shares imply. The cascade is structurally weakest precisely where the lending went deepest. This is the main reason I would not read too much into the headline mismatch numbers without the Boz invoicing data.

A Cascade Condition That Is Not Met

Here is the formal structure of the Benguria et al. model. Countries vary in their yuan revenue exposure. Each country restructures its dollar debt into yuan when that exposure crosses a threshold, and the threshold falls as more countries restructure, because each restructuring deepens the yuan sovereign debt market, lowering refinancing costs for the next country. The math produces a tipping condition. Let h be the half-range of the cross-country distribution in yuan revenue exposure, a measure of how similar or different countries are. Let Λ be the liquidity feedback parameter, how much each restructuring deepens the yuan market. The cascade triggers when:

Λ > 2ℎ

When countries are similar enough in their yuan exposure, early restructurings can pull the rest across. When they are heterogeneous, the cascade fails.

From our 46-country data, h is large. The China-aligned cluster shows a half-range of roughly 0.13 in the trade-share proxy for yuan exposure, running from Turkey (8%) to Russia (35%). The full non-China sample extends further, to a half-range of 0.15. Under any reasonable h specification, the cascade threshold is 2h ≈ 0.18 to 0.30.

Kenya’s restructuring implies Λ at most 0.043. The 4.3% interest saving on that specific deal is an illustrative ceiling on Λ, not a market-depth estimate — Λ measures how much each restructuring deepens the yuan market for the next borrower, which is a different and structurally harder thing to measure from a single deal. Taken at face value: at Λ ≈ 0.043 and 2h ≈ 0.18 to 0.30, the cascade condition falls short by a factor of four to seven. Closing the gap from near-zero to the cascade threshold requires roughly four to seven additional Kenya-scale restructurings happening close enough together to materially deepen the yuan sovereign debt market. That is approximately the scale of a coordinated conversion wave across several HIPC-eligible African and Asian Chinese borrowers simultaneously.

Not impossible. Not imminent. Somewhere in the 5–10 year range on current trajectories.

Central Banks Are Not Adjusting on the Other Side

The liability side of the sovereign balance sheet is, slowly and case by case, starting to move. The asset side is not.

In the Chenard, Eichengreen, Monnet, and Morvillier paper (CEPR DP21488, May 2026) and in my own follow-up work on reserve composition, the result is consistent: the dollar’s reserve role is orthogonal to geopolitical alignment in central bank reserve management behavior. Countries that have drifted toward China — in their political alignment, their institutional memberships, their trade patterns — do not show lower dollar securities shares in their official reserves. Countries that have drifted away from China do not show higher ones.

I tested this directly using our composite alignment score as the predictor variable in a panel regression on securities shares, following the Chenard et al. specification. The 7-dimension alignment index (combining trade, capital flows, diplomacy, arms, development finance, and institutional membership) has no predictive power for reserve composition within countries over 2010–2024 after controlling for country fixed effects (coefficient +51 percentage points per unit, p = 0.27). It has no cross-sectional predictive power either (p = 0.46). The richer index fails where Chenard’s simpler UNGA-based proxy fails, and that is the result. Central banks compartmentalize. Their reserve management is driven by financial-system considerations — adequacy, liquidity, market depth — and not by the political and economic alignment decisions being made simultaneously across the rest of the government.

This is consistent with the larger H4 null from the reserve composition work: aging creditor countries did not rotate toward deposits after the 2022 Russian reserve freeze (β = −0.203, p = 0.570). The reason, I argued there, is institutional logic, deposits at the Bank for International Settlements face the same legal jurisdiction as US Treasuries. Within the dollar system, instrument switching does not reduce seizure risk. Genuine de-risking requires gold or non-dollar currency reallocation, not instrument switching within the system. Central banks know this. They are not, on average, doing it.

The Map Is Making the Mismatch Worse

The mismatch would be concerning if it were static. It is not static.

For each of the 24 developing-country borrowers in our sample, I computed the trend in China trade share over 2010–2024. Every single one of the seven BRS-mismatch countries shows a positive trend. None is drifting toward a smaller mismatch. All are drifting deeper into one.

Indonesia is the most striking. Its China trade share has risen at 1.2 percentage points per year. At that rate it crosses 39% by 2034, against a dollar debt share of 90%. Angola projects to 41% China trade by 2034. Brazil ran from 15% (2010) to 26% (2024) and projects to 36%. These are large middle-income economies, not isolated HIPC cases.

A second wave is approaching the threshold. Thailand, Bangladesh, Kenya, and Colombia are currently below the 20% China trade share that anchors the BRS-faithful mismatch condition. On current trends, all four cross it by 2034 while carrying dollar debt shares above 70%. Thailand at 93% dollar debt and 24% projected China trade share is an illustration of how quickly the second wave could materialize.

What This Means for Post 3’s Conclusion

In The Trade-Capital Divergence, I characterized the current international monetary arrangement as a metastable equilibrium: stable against small perturbations, vulnerable to a sufficiently large shock. The dollar system’s network effects and institutional depth make it self-reinforcing. Countries trade with China and finance through the US not because they chose this but because the plumbing routes surpluses into dollar assets automatically.

The Benguria et al. mechanism is the liability-side version of the same logic. Developing countries did not choose to borrow in dollars. They defaulted into it — because China, their principal bilateral creditor, offered dollars. The dollar system is self-reinforcing on both sides of the balance sheet simultaneously. Dollar assets accumulate because the plumbing intermediates everything. Dollar liabilities accumulated because the largest alternative creditor priced in dollars.

The tension is that both sides are now beginning to move, but at different speeds. The liability side moves through sovereign debt restructuring: case by case, one railway at a time, at the pace of distressed debt negotiations. The asset side is not moving at all in the central bank data. Individual country defections face prohibitive switching costs on both sides, but the switching costs are structured differently. Restructuring a specific loan is a bilateral negotiation with one creditor. Shifting reserve composition away from Treasuries requires replacing a market with no peer; $27 trillion in outstanding securities, 88% of global FX transaction clearing, 54% of global trade invoicing. The threshold for unilateral reserve reallocation is much higher than the threshold for restructuring a single loan.

This asymmetric stickiness is the specific risk. In a world where liabilities and assets adjusted together, where each yuan restructuring was accompanied by a corresponding shift in yuan reserve accumulation, the balance sheet would rotate gradually. But central banks are not doing that. The liability side shifts. The asset side does not. Each dollar loan that gets restructured into yuan deepens the structural logic for yuan reserve accumulation without triggering it. The balance-sheet divergence compounds quietly.

When that divergence eventually closes, it closes through the asset side catching up to the liability side. That process, when it comes, will not be gradual.

The Bottom Line

The cascade is distant. Four to seven more Kenya-scale restructurings, and then only if the yuan sovereign debt market deepens proportionally, which requires PBOC commitment to market infrastructure that does not yet exist at scale. The mismatch is real for seven countries but smaller than the headline numbers suggest, for the same reason that Angola’s trade with China does not make Angola a yuan-revenue economy: oil is oil.

What is not distant is the accumulation. The BRS-mismatch countries are getting deeper into the zone every year. Indonesia is adding more than a percentage point of China trade share annually. The reserve portfolio is not adjusting. The structural gap between dollar liabilities and potential yuan reserves is widening each year the sorting continues. Kenya converted one deal. Indonesia is building toward a much larger one.

The question I ended Post 3 with, whether the financial plumbing adapts gradually or all at once, now has a partial answer on the liability side: gradually, so far. The trend is not toward gradually.

If You Want Beef Then Bring The Ruckus
New York Knicks Ain’t Nothin’ To F#ck With - RZA

After so much digital ink spilled on globalization, let’s get back to bank sup and reg.

Three Conditions Not Met: Bail-In, Supervision, and Capital

If you’ve followed the Perspective for a while you will understand that despite my background I am not a blind advocate for the status quo. The deregulatory case for banking has a logical structure, and The Bank of England’s Financial Policy Committee articulated it precisely: capital relief is warranted when three conditions are met; credible resolution, effective supervision, and an active countercyclical buffer. The FPC meant this as a framework for calibrating its own approach. It works equally well as a diagnostic.

What follows tests each condition against what is actually happening in the United States in 2026. The three conditions are being revised simultaneously, in the same direction. None of them passes.

Bail-In: The First Test Failed

The post-crisis architecture rested on a specific promise: that large financial institutions could be resolved through the bankruptcy code without taxpayer support. The mechanism was single-point-of-entry: losses absorbed at the holding company, capital downstreamed to operating subsidiaries, continuity of critical functions. The credibility of that promise has always depended on two things: that bail-in instruments actually absorb losses in a crisis, and that the legal infrastructure to execute resolution exists and works.

Credit Suisse tested both.

FINMA’s decision to write down SFr16 billion in Alternative Tier 1 (AT1) instruments to zero while equity holders received consideration was executed under emergency ordinance powers, bypassing the normal creditor hierarchy. For markets, the episode was not a Swiss idiosyncrasy. It introduced jurisdictional uncertainty into a global instrument class explicitly designed to remove ambiguity from the loss-absorption chain.

The credibility of bail-in (BIS 1356), which conducted the most systematic post-CS review of AT1 markets, found that bail-in credibility diverged across jurisdictions following the episode and that markets repriced implied government support accordingly, reducing that implied support by between one-third and one-half for banks where bail-in was judged more credible. The disciplining effect is real and measurable. Lower-rated banks captured the largest benefit, which is precisely the theoretically correct result: market discipline flows where it is most needed. But that benefit is conditional on the credibility being genuine.

The cross-border architecture supporting bail-in had a structural vulnerability that took a decade to partially address.

Bank bail-in bonds convert to equity upon triggering conditions. In April 2026, the SEC confirmed what practitioners had long suspected: that conversion constitutes an “offer” and “sale” of securities requiring registration under the Securities Act of 1933. Rodge Cohen of Sullivan & Cromwell, Meg Tahyar of Davis Polk, and former SEC Chair Jay Clayton had all flagged the concern years earlier. The gap matters specifically at speed — resolution of a failing G-SIB can happen over a single weekend; a registration process cannot.

The Bank of England resolved it for UK banks through structural innovation. It designed a new intermediate instrument, non-transferable contingent beneficial interests, that qualify for the Section 3(a)(9) exemption as an exchange between an issuer and its existing security holders. The SEC issued a no-action letter accepting that structure on April 10. The fact that the BoE had to redesign its bail-in execution architecture to accommodate US securities law is itself the signal: the gap was serious enough to require years of engagement, a new instrument class, and a formal no-action letter from the Commission.

The relief is narrow. It covers UK banks using the BoE’s specific non-transferable mechanism. Other jurisdictions whose bail-in procedures create transferable interim certificates remain outside the exemption. SEC Chairman Atkins has directed staff to prepare broader rulemaking; it is not yet issued. US bank holding companies in SPOE resolution under Chapter 11 have a separate statutory exemption that was never in question; the gap was always specifically about foreign banks with US shareholders. More than fifteen years after Dodd-Frank, the international resolution architecture is still being constructed instrument by instrument.

The FDIC’s response has moved in the opposite direction from what BIS analysis would recommend. FDIC Chair Travis Hill, in remarks on June 9, signaled a preference for sale-of-assets resolution and proposed eliminating the bridge bank option that has been the cornerstone of SPOE execution. The bridge bank is the mechanism that allows critical operations to continue when a buyer is not immediately available; it preserves franchise value while the resolution process unfolds. Removing it assumes that resolution will always produce an acquirer on the market’s timeline.

I keep coming back to Chesterton’s Fence: the bridge bank was added to the resolution toolkit because planners understood that assumption would sometimes be wrong. Removing it does not simplify resolution. It removes the tool that handles the cases where simplification fails.

The Fed’s own May 2026 feedback to Citigroup closes the loop on the first condition. The Agencies noted that Citi’s 2021 resolution plan shortcoming on data integrity “would not be remediated until the Covered Company addresses the weaknesses outlined in the Consent Order.” Supervision and resolution planning are not parallel tracks. They are the same track, which matters when the second track is being revised.

The FPC’s first condition is credible resolution. It is not met.

Supervision: Decommissioning the Early Warning System

The thing that distinguished useful examinations from compliance theater was qualitative judgment: the ability to identify a management culture, a risk governance structure, or an incentive system that was accumulating risk that had not yet materialized in the financial data. That judgment mostly lives in the Management component of CAMELS.

On May 19, the Federal Financial Institutions Examination Council proposed revisions to the rating system that would raise the bar for management-related downgrades to require evidence of “material financial risk to the institution.” The Management component would lose its current “special consideration,” the provision that gives it primacy over the other five factors, and the composite rating could no longer be held below satisfactory by M concerns alone if capital, asset quality, earnings, liquidity, and sensitivity were strong.

Greg Baer, President and CEO of the Bank Policy Institute, called the direction correct:

The Management component has had undue weight in determining bank ratings. Improving supervision requires reforming the ‘M.’

This is the industry’s preferred framing, and it is wrong in a specific way. The M rating’s value is precisely its forward-looking character: its ability to flag governance failures before they produce the financial deterioration that would be visible in the other five components. Setting “material financial risk” as the threshold for intervention makes the instrument retrospective by design.

The case against this was made, empirically, in 2008. Cliff Rossi served as credit risk officer at Washington Mutual before its failure. At the time, WaMu’s financial metrics were clean. The bank’s incentive structure was driving loan officers to prioritize mortgage volume over borrower quality — a management failure that would not appear in credit risk metrics until it was too late. Rossi told American Banker in Ready or not, bank examination is changing for good:

The material risks that we saw — in terms of, let’s say, credit risk metrics — would not have risen to a level that would have warranted downgrading them to [below satisfactory] at that time. That’s a really big miss on the part of the FFIEC. Should the proposed changes go forward, it could undercut all the advances in risk management, organizational stature and effectiveness that have been in place since the ‘08 crisis.

The 2023 failures repeat the pattern. The FDIC’s post-mortem on Signature Bank identified poor management, specifically, pursuit of “rapid, unrestrained growth” without sufficient risk controls, as the root cause of the failure. It simultaneously disclosed that FDIC examiners had never downgraded Signature’s management rating, scoring it satisfactory every year from 2017 through 2021 while the failure developed. It should have. The current framework didn’t catch it. The proposed framework would have had even less basis to act.

The GSIB resolution plan feedback letters released by the Fed and FDIC in May 2026 are instructive from a different angle. Four of the eight largest US banks - Bank of America, Citigroup, Goldman Sachs, and JPMorgan Chase - could not model their derivatives portfolio unwind at the counterparty level in their 2023 plans. This is not an edge case: counterparty-level derivatives modeling is a core requirement for executing their stated resolution strategy. Goldman Sachs allowed a UCC-1 financing statement, the security interest underpinning its resolution financing structure, to lapse; the Agencies noted the lapse “could have undermined the execution of the Covered Company’s resolution strategy.” Morgan Stanley’s secured support agreement, the document governing intragroup capital flows in resolution, did not reflect which entities were actually parties to it. None of these failures would have registered in capital ratios, earnings, or any other quantitative screen. Each was a management failure.

Phillip Basil, a former supervisory policy specialist at the Federal Reserve Board, put the function plainly:

Management is the last early warning system. If management is failing to do their job appropriately, it’s creating financial risks.

Basil also identified the reform’s internal contradiction. Remove examiner discretion and replace it with a “material financial risk” standard, and regulators will eventually be forced to define precisely what that standard means, producing universal thresholds that put risk management entirely in regulators’ hands.

It becomes the government telling banks how to manage their risks, which has never been the role of supervision. The regulators haven’t thought this all the way through — they’re just giving the banks what they want.

David Zaring of Wharton offered the wider frame:

There’s always been a deal between the government and the banks: banks get access to cheap funding from depositors and they get a lot of intrusive and discretionary supervision that they can’t really complain about. What we’re seeing now is a different approach.

The deal has two sides. Access to the federal safety net - deposit insurance, Fed liquidity facilities - has not been revised. The intrusive supervision that was the quid pro quo is being softened.

The FPC’s second condition is effective supervision. It is not met.

Capital: Wrong Direction, Explained By Its Own Logic

Solvency, Not Runs

Before engaging the conditional arithmetic, the evidentiary base matters. Bank Failures: The Roles of Solvency and Liquidity (Liberty Street) examined 160 years of US bank failures across more than 5,000 institutions. The finding is direct: bank failures are primarily solvency events, not liquidity events. Across the roughly 2,000 cases with sufficient data, runs were identified as a contributing factor in fewer than twenty. When a run does occur, it typically follows insolvency, a rational withdrawal by depositors reading what the balance sheet already tells them.

The policy implication matters here. The reform hierarchy being pursued, relaxing capital requirements while strengthening liquidity infrastructure and resolution planning, is inverted relative to the evidence. Capital absorbs solvency losses. Liquidity bridges solvent institutions through short-term funding gaps. They address different failure modes, and the data are clear about which mode actually drives failures. A framework that leads with liquidity and treats capital as the adjustment variable has the causation backwards.

The Conditional Arithmetic

The BoE FPC’s three conditions are worth restating as arithmetic. Credible resolution earns a credit of up to 5 percentage points of required capital, but only if resolution is genuinely credible. Effective supervision anchors the early warning system that makes pre-crisis intervention possible. An active CCyB provides the buffer to cut during downturns. All three are complements, not substitutes: weakening one raises the required contribution from the others.

The United States in 2026 is weakening all three simultaneously.

BIS WP 1356 found that bail-in credibility was damaged in some jurisdictions by the CS episode. The cross-border securities law gap required a formal SEC no-action letter and a new instrument structure to partially address — and remains unresolved for most foreign jurisdictions. The CAMELS reform is explicitly reducing supervisory discretion. The US countercyclical buffer has been structurally underdeployed; the Basel Committee noted in 2019 that the US CCyB framework would remain untested until a credit cycle turned. It has not been meaningfully deployed since.

Remove the 5-point resolution credit because the first condition fails. Remove whatever supervisory option value is embedded in the capital adequacy judgment because the second condition is being weakened. Note that the buffer which was supposed to absorb tail risk has never been road-tested. What remains is a capital reduction being pursued as though all three credits were available.

The Level Playing Field That Isn’t

Governor Bowman has argued in Capital Rules for the Real Economy that US capital rules are more stringent than those of peer jurisdictions and that the excess imposes competitive costs without commensurate safety benefits. The ECB published a counterfactual analysis supporting the first part of this claim: for the largest European banks, current US capital requirements are in many respects more stringent than EU equivalents. The GDP-indexed framework Bowman has proposed for measuring implementation excess is technically defensible.

I find the argument partially persuasive … for mid-tier institutions. But for the eight GSIBs whose failure would threaten financial stability, the direction is wrong when none of the three BoE conditions is satisfied. The rule change is not calibrated to the condition of the safety net. Goldman Sachs’s CET1 ratio has moved from 15.1 percent to 13.3 percent under current trajectory. That reduction is occurring as bail-in credibility is impaired, supervisory quality is being reduced by regulation, and the CCyB remains undeployed, and there are rumors that the bigest banks may be allowed to further expand, becoming even more difficult and systemic.

The Substitution That Doesn’t Work

One remaining argument for capital reduction is that strengthened liquidity requirements compensate. They do not.

Capital and liquidity address different failure modes. Capital absorbs losses. Liquidity ensures that a solvent institution can meet obligations during stress. An institution that is insolvent cannot be stabilized by liquidity, which is precisely what Correia, Luck, and Verner demonstrated across 160 years of failure data.

The 2020-2023 episode made the point empirically. As excess deposits accumulated during the pandemic, Treasury securities holdings across the system grew from roughly $400 billion to $550 billion, a portfolio decision driven by deposit inflows rather than active duration management. The resulting exposure was not a capital question at the time of accumulation. It was a risk management question, the kind that a functioning M rating is supposed to surface before it produces losses. The supervisory framework did not catch it. The proposed $117 billion reduction from a $2 trillion Basel III package is occurring in an environment where this specific failure mode, visible, manageable risk not incorporated into management decision-making, is documented at scale.

Capital reduction while supervision is softened does not improve the risk-return profile of the banking system. It removes the buffer that absorbs the consequences of the management failures the M rating was designed to catch in advance.

Summarizing

The Bank of England’s framework is useful precisely because it is not hostile to capital relief. It says: if three conditions are met, lower requirements are justified. The conditions are specific. The reasoning is coherent. Check the conditions.

None of them are met in the United States in 2026. Bail-in credibility is impaired by an unresolved securities law gap and a regulatory direction that is simplifying resolution at the moment BIS research says the infrastructure needs building. Supervisory quality is being reduced by design; the CAMELS M reform explicitly removes the only forward-looking, qualitative early warning in the examination toolkit. The countercyclical buffer has never been deployed in a credit downturn.

The FDIC’s own root-cause finding for Signature Bank, poor management, coexists with its acknowledgment that examiners never downgraded Signature’s management rating while the failure developed. Goldman’s UCC-1 lapsed undetected. Morgan Stanley’s support agreement was wrong. Four of eight GSIBs couldn’t model their derivatives books by counterparty. Citi’s resolution shortcoming traces directly to a supervisory enforcement action from 2020. These are not failures of capital ratios. They are failures of the management oversight that the proposed CAMELS reform would make even harder to act on.

Bank Supervision Officer At The Fed Once Carried Firearms

I guess examiner’s will need more firepower.

Russian banks to arm themselves against Ukrainian drones (FT)

Under new legislation adopted on Tuesday, the central bank, top lender Sberbank and the Russian Cash Collection Association will be able to equip themselves with jamming devices and other air defences. Their staff will also be permitted to carry weapons, according to a lawmaker, who said the measures would not be state-funded.

The GFC Is Finally Over

Lone Star Weighs Sale of Lender IKB After Nearly 20 Years (Bloomberg)

Covid Support is Finally Over Too

The Exit Ramp for the Main Street Lending Program: FAQ M.6 and the Boston Fed’s Sale of Participation Interests

The Trade-Capital Divergence

Thirteen countries in our sample trade more with China than with the United States yet park their financial surpluses in the dollar system. Australia sells 30% of its exports to China and sends 57% of its portfolio investment to the US. South Korea trades 20% with China and directs 64% of its portfolio to the US. Japan, Germany, the Netherlands, New Zealand, Poland: same pattern. The economic gravity of these countries is split.

Their goods flow east. Their money flows west.

Standard macroeconomics says this shouldn’t persist. The balance-of-payments identity requires that trade surpluses flow somewhere as capital. A country running persistent surpluses with China should, in principle, accumulate Chinese financial assets. Yet the data show the opposite: surpluses earned in China trade are recycled into dollar assets.

The question is whether this disconnect is a transitional lag or a durable structural feature of the international system.

The answer has implications well beyond the alignment maps. It connects to everything I’ve written about in prior Perspective posts on the dollar, on liquidity, on the financial system’s architecture. The trade-capital divergence is the structural reason the geopolitics story and the financial-system story are the same story.

The Domain Divergence

The broader finding first: economic and political alignment are partially decoupled, and the degree of decoupling is large enough to be the defining structural feature of the current geopolitical environment.

The evidence is in the principal component analysis. When we run PCA on all fourteen sub-components of the alignment index, the first principal component explains only 43.6% of variance on the US side and 52.7% on the China side. That sounds high until you compare it to what happens when you restrict the analysis to political sub-components alone (diplomacy, arms, institutions): the first PC jumps to 68-71% explained variance. The drop occurs because the economic sub-components load weakly or orthogonally to the political ones.

The loadings tell the story. On the China side: arms procurement loads at 0.69, institutional membership at 0.65, diplomacy at 0.28. All four economic sub-components (trade, portfolio, FDI, development finance) are near zero: 0.04, 0.13, -0.02, 0.03. A country’s choice of arms supplier and institutional memberships tell you almost everything about its political alignment with China. Its trade and investment patterns tell you almost nothing.

Country by country, the divergence is vivid:

Mexico: economic net alignment +0.55, political net alignment +0.12. Trades 61% with the US and sends 99% of FDI there, but votes with China at the UNGA. If forced to align politics with economics, Mexico would be firmly in the US bloc. If forced to align economics with politics, Mexico would need the largest trade reorientation of any country in the sample.

India: economic net +0.19, political net -0.48. Portfolio investment flows to the US (0.66 share). Diplomacy, arms, and institutions pull toward China. India’s economic gravity is Western; its political gravity is non-aligned-to-Eastern.

Australia: economic net +0.09, political net +0.59. Politically deep in the US network (Five Eyes, AUKUS, Quad, 96% Western arms), but 30% trade with China. If a sorting event forced economic alignment to match political alignment, Australia would face the largest adjustment cost of any US-allied economy.

Taiwan: economic net -0.18, political net +0.50. Sends 26% of outward FDI to China (the highest in the sample) while being a de facto US ally. Taiwan’s economic dependence on China is its single greatest geopolitical vulnerability.

In every prior episode of bloc formation, this divergence eventually resolved. The interwar sterling and dollar blocs aligned trade with finance. Cold War blocs aligned trade with security. But the convergence wasn’t automatic. It required a sorting event: the Marshall Plan, the Korean War, Britain’s departure from the gold standard. Before those events, countries maintained economic ties across the emerging political divide. Exactly as they do today.

The current environment has produced several candidate sorting events (the Huawei ban in 2019, the October 2022 chip controls, the Russian reserve seizures in 2022). But none has been comprehensive enough to force economic alignment across the board. The chip controls affected South Korea and Taiwan. The Russian sanctions affected Russia and its immediate partners. Neither forced Brazil or India or Saudi Arabia to choose. The domain divergence persists because no single event has yet forced the full sample to reveal preferences simultaneously.

The Deeper Puzzle: Trade and Capital Pull in Opposite Directions

The domain divergence is the known known. The deeper puzzle is that it’s getting worse, not better, under structural projections.

Using bilateral gravity models calibrated to demographic trends and GDP forecasts, I projected trade shares and capital flow shares for each country through 2044. The results move in opposite directions.

Trade gravitates toward China. China’s share of the 46-country GDP universe rises from 19.2% (2024) to 22.8% (2044). Under the baseline projection (demographic-driven GDP growth, no policy shocks), trade shares with China rise for most countries. By 2044, 13 countries trade more with China than with the US yet remain net US-aligned, up from 10 in 2024. The list includes Australia, Japan, South Korea, Germany, the Netherlands, Poland, New Zealand, Ukraine, Hungary, Turkey, Chile, and Argentina.

Capital gravitates toward the United States. Using the demographic gravity model from my earlier work on capital flows and demographics, we projected portfolio investment and FDI shares. The results diverge sharply from the trade projections. As Japan, Korea, and Europe age, their outward capital gravitates toward the US, whose deeper markets, open capital account, and relatively younger demographics (the US working-age population is still growing, unlike every other advanced economy) make it an increasingly attractive destination.

Portfolio shares directed to the US rise for virtually every country in the sample:

Capital flows toward China also increase, but from a much smaller base and at a slower rate: Singapore’s portfolio share to China rises from 0.11 to 0.14; Korea’s FDI to China rises from 0.16 to 0.22. The US’s relative advantage in attracting capital widens through 2044, even as its advantage in attracting trade narrows.

This is the finding with no historical parallel. In the 1930s, trade blocs and currency blocs were mutually reinforcing. Eichengreen and Irwin (1995) document that the sterling bloc traded internally and invested internally. The projection suggests the opposite: the structural forces driving trade (GDP mass, geographic proximity) and the structural forces driving capital (financial depth, institutional quality, demographic complementarity) push in different directions.

The implication: the domain divergence is not a transitional artifact that gravity will resolve. It is a structural feature of the US-China competition that bilateral gravity models expect to persist and deepen.

A comprehensive sorting event that forces convergence would need to overcome the fundamental divergence between comparative advantage in goods (which favors China’s manufacturing scale) and comparative advantage in finance (which favors US capital market depth). This structural tension makes clean bipolar sorting costlier than any prior historical episode.

The Dollar as Universal Intermediary

So how does it work? How can countries trade with China and finance through the US without the balance of payments identity breaking?

The resolution lies in the dollar’s role as universal intermediary. A Vietnamese exporter sells goods to China, receives payment in dollars (or converts renminbi to dollars almost immediately), deposits those dollars in a correspondent bank in New York, and the surplus flows into US Treasury securities or dollar-denominated money market instruments. The bilateral trade relationship is with China. The bilateral financial relationship is with the United States. The dollar system intermediates even flows that do not involve the US as a counterparty.

This is the operational mechanism behind Barry Eichengreen’s “exorbitant privilege.” Not a privilege of seigniorage, but a privilege of intermediation. As long as the dollar clears 88% of foreign exchange transactions (BIS Triennial Survey, 2022) and denominates 54% of global trade invoicing Boz, Casas, Diez, Gopinath, and Gourinchas (2022), the financial system routes surpluses earned anywhere into dollar assets regardless of where the goods trade occurs.

Countries are not choosing to invest in the US despite trading with China. The plumbing of the international monetary system automatically converts trade surpluses into dollar claims. The choice is not bilateral. It is systemic.

The persistence of the divergence also reflects a fundamental asymmetry between goods markets and financial markets. Reorienting trade is costly but bounded. The evidence on US import rerouting from China to Vietnam and Mexico Global Supply Chains: The Looming “Great Reallocation” (Alfaro et al., Jackson Hole, 2023) shows trade can shift 6 percentage points in five years when policy pressure is applied. Financial infrastructure operates on an entirely different timescale. Replicating the depth and liquidity of the $27 trillion US Treasury market, the legal infrastructure of English-law contracts and New York arbitration, the SWIFT/CHIPS messaging and clearing network (11,000+ institutions across 200+ countries), and the correspondent banking relationships that undergird dollar clearing: this is a generational project with no historical precedent for rapid substitution.

The sterling precedent puts a number on the lag. Britain’s share of world trade peaked in the 1870s and declined steadily for six decades. Sterling’s share of global reserves did not fall below 50% until the late 1940s, and did not reach its current level (roughly 5%) until the 1970s. The lag between trade decline and financial decline was 70 to 100 years. If the dollar follows a similar trajectory, the trade-capital divergence could persist well beyond 2044.

Three Things That Could Break It

The divergence is durable but not permanent. Three mechanisms could force convergence.

Weaponization of Financial Chokepoints

Every exercise of US financial coercion raises the cost of intermediation through the dollar system. SWIFT exclusions, secondary sanctions, asset freezes, threatened restrictions on dollar clearing. Farrell and Newman’s framework predicts a self-undermining dynamic: the more aggressively the US exploits the chokepoint, the faster countries build alternatives.

The freezing of $300 billion in Russian central bank reserves in 2022 was the most consequential demonstration that dollar-denominated assets can be weaponized against their holders. For the 27 countries not firmly in the US political orbit (21 China-aligned plus 6 on the contested diagonal), this created an incentive to diversify that did not previously exist at scale.

Eichengreen’s most recent assessment (”When the Facts Change,” CIGI, June 2025) identifies the Russian reserve seizure as the event most likely to have shifted the structural equilibrium. But his conclusion is cautious: the alternatives are not yet credible enough to trigger rapid reallocation. The incentive exists. The infrastructure does not.

China Opens Its Capital Account

The binding constraint on financial bifurcation is supply, not demand. Countries that might prefer to hold renminbi-denominated assets face a practical problem: China does not fully allow it. The capital account is managed. The onshore bond market lacks foreign-friendly infrastructure for custody, settlement, and legal recourse. Property rights under Chinese law are uncertain for foreign holders.

Until China liberalizes, the dollar remains the default recycling destination almost by elimination. The renminbi’s share of global reserves has stalled at approximately 2.3% since 2022, far below what China’s 15% share of global goods trade would predict. The gap between trade weight and financial weight is itself the single best measure of how far China is from offering a credible alternative.

China’s Cross-Border Interbank Payment System (CIPS) is growing: 193 direct participants, 1,573 indirect, spanning 124 countries, with RMB 180.2 trillion (roughly $26.4 trillion) in transactions in 2025, up 43% from 2024. But CIPS still relies on SWIFT messaging for over 80% of transactions. The yuan accounts for 3% of global SWIFT payments versus 48% for the dollar.

Project mBridge (the BIS initiative with central banks of China, Thailand, Hong Kong, and the UAE) reached MVP stage in June 2024 and processed $55.49 billion through November 2025. The architecture for SWIFT bypass exists. The scale is trivial.

But as I’ve documented in the Demographics and Capital Flows series, China’s window for openning is very tight.

A Phase Transition

The Morrison, Kutz, and Gabbay model of international relations as a signed network (Network Science, 2023) provides the formal framework for how the divergence could close suddenly rather than gradually. Their model shows that the system exhibits a sharp bifurcation from a mixed equilibrium to a polarized one as structural balance pressures exceed a critical threshold, and the transition is hysteretic: once the system tips, returning to the prior state requires reducing pressures well below the original tipping point.

Applied to the trade-capital divergence: if a Taiwan crisis, a comprehensive financial embargo, or an equivalent sorting event forced countries to choose between the dollar system and a China-centered alternative simultaneously, the financial divergence could close in years rather than decades. Individual country defections from the dollar system face prohibitive switching costs. Coordinated defection (a critical mass of countries moving together, as mBridge is designed to facilitate) could generate the network effects needed to make an alternative viable.

The key word is simultaneously. The gold standard was a metastable equilibrium until September 1931, when Britain’s departure triggered coordinated exits by 25 countries within 18 months. The question for the dollar system is whether its depth advantage is large enough to prevent a similar cascade.

Metastable

There’s that word again.

The most accurate characterization of the trade-capital divergence is that it represents a metastable equilibrium. Stable against small perturbations. Vulnerable to a sufficiently large shock.

The dollar system’s network effects, institutional depth, and legal infrastructure make it self-reinforcing under normal conditions. Countries trade with China and finance through the US not because they have chosen this arrangement but because the global financial plumbing routes surpluses into dollar assets automatically. No individual country can profitably defect. The switching costs exceed the benefits of any unilateral move.

But metastability is not stability. Each exercise of financial coercion incrementally raises the return to coordinated defection. The system’s vulnerability lies precisely in its dependence on continued US restraint in weaponizing the chokepoints that sustain it.

If the divergence persists, the current equilibrium costs nothing and the alignment map remains structurally ambiguous. If the divergence must close, the direction of closure determines the architecture of the international system for the next century. Trade converging with finance means countries reduce their China trade to match their dollar-system commitment: the “friend-shoring” outcome. Finance converging with trade means countries reduce their dollar-system dependence to match their China trade orientation: the de-dollarization outcome. The second is far more consequential and far less studied, because it implies not just a reallocation of portfolios but a restructuring of the global monetary system itself.

History suggests that when trade and financial patterns diverge long enough, it is the financial arrangement that eventually gives way. But the lag is measured in decades. And the transition is triggered by crisis, not drift.

The world’s financial plumbing was built for a unipolar moment that no longer exists. The question isn’t whether it adapts. It’s whether it adapts gradually or all at once.