The Loan Is Not the Pool

I — The Claim and Its Appeal

Private credit has become one of the largest asset classes in the world, and one of the least liquid. The Financial Stability Board puts the global market at between one-and-a-half and two trillion dollars; Moody's expects assets under management to pass two trillion in 2026 and approach four trillion by 2030 (FSB, 2026; Moody's, 2026). It is held by institutions and the wealthy, in positions that begin in the millions and lock up for years. There is no continuous price and no ready exit; a holder who wants out negotiates a sale or waits for maturity. The asset class has grown enormous precisely while retaining the character of a private arrangement between a lender and a borrower.

Tokenisation is offered as the dissolution of all of this. Represent the loan as a token on a ledger, the argument runs, and what was indivisible becomes fractional, what was locked becomes transferable, and what was exclusive becomes open. A holder could mark and move a position continuously rather than waiting out the term. An investor without an institutional allocation could own a slice of a direct-lending book. The secondary market that private credit has never had would come into being, because the instrument would finally be tradable.

The appeal is real and worth stating without irony. Much of what tokenisation promises private debt is genuinely useful, and some of it is genuinely deliverable. But the promise is, first and above all, a promise of liquidity — that is the headline, the thing that justifies the enthusiasm, the industry projections that run to sixteen trillion dollars of tokenised assets by 2030 (BCG & ADDX, 2022), and the valuations attached to the platforms that promise it. And the liquidity promise is the one part of the pitch that does not survive contact with the structure of the asset. To see why, it is necessary to separate what tokenisation changes from what it does not.

II — What Tokenisation Actually Changes

A token is a representation of a claim on a ledger. Tokenising an asset changes the mechanism by which ownership of that claim is recorded, subdivided, transferred, and settled. It does not change the claim. The cash flows are what the loan agreement says they are; the seniority is what the documents establish; the risk is whatever the borrower's circumstances make it; and the information available about all of this is whatever it was before the token existed. The wrapper changes. The asset inside it does not.

Within that boundary, the merits are real, and a fair account should grant them in full. Settlement becomes atomic: the transfer of the claim and the payment for it occur in a single step, removing the settlement cycle and the counterparty risk that lives inside it. Servicing becomes programmable: interest and principal waterfalls can be executed by code, distributions automated, transfer restrictions and eligibility rules written directly into the token. Ownership becomes divisible to an arbitrary degree, lowering minimums and widening the base of potential holders. Pool composition and performance can be made visible on a ledger rather than disclosed in a quarterly statement. And the administrative layer — fund accounting, transfer agency, primary distribution — becomes faster and cheaper.

These are worth having. But they belong, every one of them, to a single category: operational efficiency and access. None of them is liquidity. Liquidity is the standing presence of willing buyers at a narrow spread, available on demand (Amihud & Mendelson, 1986). It is a property of a market, not of a settlement system. A claim can settle instantly, divide infinitely, and report transparently, and still find no bid worth hitting. The merits tokenisation delivers and the liquidity it is sold on are different things, and the distance between them is where the argument quietly goes wrong. Everything tokenisation genuinely does sits on the wrapper. Liquidity is determined by the asset and the information environment around it.

III — Why the Single Loan Resists Liquidity

On a single private loan, three barriers stand between the asset and a deep market. All three are informational, and none is addressed by putting the loan on a chain.

The first is heterogeneity — stated carefully, because the naive version of this argument proves too much. Every private loan is bespoke: principal, maturity, and amortisation differ; the coupon may be fixed or floating and set at an idiosyncratic margin; seniority, security, and collateral differ; the covenant package is negotiated case by case; the governing law and the borrower are specific to the deal. It is tempting to stop there and declare heterogeneity fatal to trade. But the corporate bond market falsifies that strong claim: every bond is its own instrument too, with its own covenants, maturity, seniority, and identifier, and bonds nevertheless trade. What the bond has is not fungibility. It is infrastructure — a mandated disclosure regime that makes the issuer legible to strangers, an agency rating that compresses the credit into a comparable grade, and post-trade price transparency that lets every executed trade inform the next quote. Heterogeneity is not itself the barrier; heterogeneity without a shared informational reference is. The private loan has the bond's heterogeneity and none of its scaffolding: no mandated disclosure, no rating, no tape. Pricing it therefore means analysing it from scratch, and a market where every unit demands its own underwriting is not a deep market in one thing but a thin market in thousands of distinct things. This, it should be said, is the one barrier on which a tokenisation ecosystem could in principle make progress — standardised data schemas, common documentation, on-chain servicing records could begin to build for loans what disclosure regimes built for bonds. But that is institution-building, slow and collective, a property of a market's rules rather than of any ledger. The token records whatever terms exist. It does not standardise them, and it does not make anyone disclose.

The second is adverse selection. Private debt is, by definition, private information. The originator has met the borrower, conducted the diligence, and accumulated the soft knowledge a lending relationship produces (Diamond, 1984; Petersen & Rajan, 1994); the secondary buyer has none of this. So when an originator offers a particular loan for sale, the buyer must ask why this one. The seller knows more, and the offer is therefore a signal — the buyer rationally infers that the loans most likely to be put up for sale are the ones the originator most wants off its book, which are the deteriorating ones. This is the market for lemons (Akerlof, 1970; Stiglitz & Weiss, 1981), and its consequence is a bid wide enough, or absent enough, to prevent the trade. In a continuously traded fungible asset the signal is muted, because participants buy and sell constantly for reasons unrelated to the specific instrument, and any single sale carries no information. In single private loans there is no such cover. A sale is a statement about the loan.

Here, and only here, the token appears to reach the mechanism rather than the wrapper, and the objection deserves to be stated at full strength. Adverse selection bites hardest when the seller is the originator — and today the seller almost always is, because institutional minimums keep the holder base to a handful of informed parties. Fractionalise the loan across many small holders, the objection runs, and most sales become liquidity-motivated noise: someone needs cash, not someone knows something. That is precisely the cover that continuous markets provide, and the token is what makes the wide holder base possible. The objection is fair, and it fails one step later. The cover exists only after uninformed holders are in — and to get in, they must first buy an information-sensitive claim from the informed party, at issuance and again at every deterioration the servicer observes before they do. The discount they rationally demand for holding a claim whose bad news always reaches someone else first is permanent, and it does not shrink because there are more of them. The literature followed this logic to its conclusion decades ago: an issuer facing that discount responds by redesigning the security — retaining the information-sensitive piece and selling a claim engineered to be insensitive to what the issuer knows (DeMarzo & Duffie, 1999). That engineered claim is the tranche of a pool. Fractionalisation, pursued to its economic end, does not rescue the single loan. It reinvents securitisation.

The third is the credit-risk question. To trade a claim you must price its risk, and for a single private loan there is no cheap, continuous, comparable way to do so. A public bond carries an agency rating, a set of observable comparable spreads, and a live market price. A single private loan carries none of these. Rating it is a bespoke underwriting exercise — manual, expensive, and stale as soon as the borrower's circumstances move. There is no market price to defer to, because by construction there is no market. Every prospective buyer must therefore underwrite the credit independently and from the beginning, and the cost of doing that for one loan, measured against the size of the position, is prohibitive for all but a few specialists.

The three compound: heterogeneity without infrastructure means a buyer cannot price the loan by comparison to others; adverse selection means a buyer cannot trust the price implied by the seller's willingness to sell; the rating gap means a buyer cannot cheaply establish a price alone. But it is worth being exact about what the compounding does. It does not forbid trade; it prices it. The proof is standing in plain sight: the syndicated loan market, where hundreds of billions of dollars of equally bespoke, privately negotiated, single-name loans change hands over dealer desks every year. That market is real — and it is thin per name, quoted at wide spreads, restricted to specialists, and settled over weeks. It is what the three barriers produce when they operate on single-name credit: trading that exists and is expensive, negotiated, and slow. It also happens to be a controlled experiment for the token's limits. The one conspicuously broken thing in loan trading that a ledger genuinely could fix is the settlement; fix it, and the market would settle in seconds and trade exactly as thinly as before, because the width of the bid was never in the back office.

IV — Why the Pool Can

What is impossible for the loan becomes possible for the pool, and for a reason that maps exactly onto the three barriers: the behaviour of large numbers.

A single loan either performs or it does not, and which it does turns on facts specific to one borrower that are difficult to know from outside. A pool of thousands of loans is a statistical object. No one can say whether a particular loan in it will default, but the fraction of a large and diversified pool that will default can be estimated with usable confidence, because idiosyncratic outcomes offset one another and what remains is aggregate behaviour that can be modelled. The pool is describable even though its constituents are not — and that single fact addresses, in turn, each of the obstacles that defeated the loan.

Against heterogeneity, the pool is tranched. The tradable units are not the underlying loans but claims on the pool's cash flows defined by their position in the loss waterfall — senior, mezzanine, equity (DeMarzo, 2005). A tranche is defined by where it sits in the structure, not by the terms of any loan beneath it, and tranches of the same class are fungible in a way the loans never were. The unit of trade is no longer the loan; it is the tranche.

Against adverse selection, the structure does what the fractionalised loan could not: it manufactures a claim that does not require the buyer to know what the seller knows. The senior tranche is engineered to be information-insensitive — protected by enough subordination that its value is indifferent to the news any single borrower generates, which is exactly the design that lets uninformed holders own debt without analysing it (Gorton & Pennacchi, 1990). The buyer of a senior tranche over several thousand loans is not exposed to the originator's choice of any single credit; the exposure is to the aggregate, and origination quality can be assessed at the level of the sponsor — through retention requirements and a demonstrated record — rather than loan by loan.

Against the rating gap, the diversified pool can be rated, because its loss behaviour is statistically characterisable in exactly the way a single loan's is not. This is what rating agencies do for asset-backed securities and collateralised loan obligations: they model the pool, not each loan, and assign the tranches ratings, comparable spreads, and reference prices. The tranche acquires the very things the single loan lacks — a rating, a comparison set, a price.

The account so far is, however, too kind to the pool, and the argument is stronger for saying so. Diversification cancels idiosyncratic risk only. What it leaves in the senior tranche is systematic risk, concentrated: a claim that pays in every state of the world except the correlated ones, which is why structured seniors behave like economic catastrophe bonds and why every pool of the same vintage goes bad in the same downturn at once (Coval, Jurek & Stafford, 2009). Nor is adverse selection dissolved; it migrates from the loan to the pool's composition, where the sponsor chooses what goes in — the crisis produced pools composed to fail — and it is held in check by retention rules, disclosure, and reputation rather than removed by the mathematics. And the tranche's liquidity is itself conditional. Securitised debt trades cheaply precisely because holders do not need to analyse it; when events cast doubt on the pools, the debt turns information-sensitive, everyone must suddenly analyse what no one can, and the liquidity vanishes at once — which is what happened in 2007–08 (Gorton & Metrick, 2012). These caveats do not weaken the thesis. They complete it. The pool's liquidity is not a free gift of aggregation but a manufactured, regulated, regime-dependent achievement of information engineering — which is precisely the point. Liquidity in credit is produced by structure and institutions, and by nothing else.

So the pool is fungible where the loan was heterogeneous, insensitive where the loan was information-laden, and ratable where the loan was not. It possesses — in good states, under regulatory scaffolding — the three properties whose absence keeps the market in single loans thin. This is not a novel construction. It is securitisation, several decades old, and its products — asset-backed securities, mortgage pools, collateralised loan obligations — already trade in functioning secondary markets. A tokenised structured pool can be liquid for the simple reason that the thing being traded, the tranche, is already the kind of thing that trades.

V — The Liquidity Was Never in the Token

Set the two halves beside each other and the conclusion is forced. Take any instance of tokenised private debt that is genuinely liquid, trace the liquidity to its origin, and the origin is never the token. It is the pooling, the tranching, and the rating — the engineering that converts thousands of bespoke and unratable claims into a handful of standardised, information-insensitive, ratable units. That engineering long predates tokenisation and works without it. The token, added on top, contributes its real merits: atomic settlement, programmable waterfalls, divisible units, on-chain transparency. But those are efficiencies applied to an instrument that had already been made liquefiable by its structure. The token makes an already-tradable thing cheaper and faster to trade. It does not make an information-sensitive thing insensitive — and it is the sensitivity, not the settlement, that keeps the single loan's market thin.

The converse confirms it. Apply the same token to the single loan — the claim that has not been pooled, tranched, or rated — and the loan trades exactly as it did before: rarely, by negotiation, at a discount priced for everything the buyer does not know. The token has done nothing about the missing disclosure infrastructure, nothing about the asymmetry, and nothing about the absence of a cheap way to price the risk. A tokenised single loan settles flawlessly and finds the same thin, wary bid it always found. The settlement was never the problem.

And the experiment has, by now, been run. The tracked universe of tokenised credit — private, corporate, structured, and on-chain lending together — stood near forty-three billion dollars in July 2026, and the tracker's own taxonomy states this essay's argument in an official vocabulary. Nearly thirty-six billion of that value is classed as represented: tokens that, in the tracker's definition, cannot be moved outside the issuing platform or transferred between wallets, whether by design or by regulatory constraint — the blockchain serving as a recordkeeping and reconciliation layer. Only seven billion is distributed: tokens that can leave the platform and move peer-to-peer (rwa.xyz, 2025, 2026). And transferable is not the same as traded: the empirical studies of that remaining sixth find low turnover, thin active-address counts, long passive holding periods, and minimal secondary trading — outstanding token value does not predict trading activity, and tokenisation and liquidity emerge from the data as distinct outcomes (Mafrur, 2025, 2026). So the ladder has three rungs, and the population thins at each: five-sixths of tokenised credit cannot change hands at all, a sixth technically can, and of that sixth almost none does. It is telling which products sit on which rung — among the assets the tracker classified as recordkeeping-only were the platforms tokenising individual private-credit loans, the loan-level products landing on precisely the side of the line this essay predicts. Meanwhile the entire universe, recordkeeping included, amounts to between two and three percent of a one-and-a-half-to-two-trillion-dollar asset class. Issuers have adopted the wrapper for exactly what Section II said it delivers — reconciliation, transparency, operational efficiency — and the market has declined to supply the rest. The wrapper, given several years to prove otherwise, has behaved like a wrapper.

So the claim that tokenisation will make private credit liquid is true and false in a precise pattern. It is true wherever the credit has already been engineered into a liquid form, and there the token is an efficiency layered on liquidity the structure produced. It is false wherever the credit has not been so engineered, and there the token changes the wrapper while the asset trades as thinly as before. The liquidity, where it exists, lives in the structure and the institutions around it. It was never in the token. The move that makes the pitch persuasive — accurate in its parts and misleading in its whole — is to credit the token with a liquidity that the securitisation beneath it created, and that the securitisation could have delivered alone.

— What Will Actually Happen

Tokenisation of private debt will grow. That much is not in question. But it will grow along the channels where it adds value, not the one on which it is marketed.

The first is operational efficiency in the primary market and the fund layer. Tokenised issuance, automated servicing and distributions, cheaper administration, and fractional access for a wider base of investors are real advantages, and they will be adopted because they lower cost and friction. This is tokenisation doing what it is genuinely good at: being a better wrapper around the asset and the fund that holds it.

The second is the settlement layer of such loan trading as already exists. Syndicated loan settlement is a genuine dysfunction — weeks of manual reconciliation for a market that trades daily — and a ledger is a genuine remedy. This is worth doing and will be done. It will make an existing thin market faster and cheaper to operate. It will not make it deep, for the reasons Section III gives.

The third is secondary trading of tokenised structured product at the tranche level. As securitised private credit is tokenised, its tranches can trade with the settlement and programmability advantages the token provides. This will produce a genuine secondary market, though a measured one — and it is worth being exact about what trades in it. It is a market in tranches, the same instruments that already trade today, now settling on a ledger. The novelty is operational, not categorical.

What will not arrive is a deep, continuous secondary market in single private loans. The barrier was never the wrapper. It is the missing informational infrastructure — the disclosure, the rating, the shared price that the bond has and the loan does not — compounded by an asymmetry no ledger neutralises. Single-loan trading will look, at best, like syndicated loan trading with modern settlement: real, thin, negotiated, and priced for what the buyer cannot know. If a deeper market in loan-level risk ever emerges, it will come from the slow, collective construction of that infrastructure — standardised data, common documentation, credible loan-level reporting — and the token will have been one useful piece of the plumbing, not the cause.

The question, in the end, was never whether private debt can be tokenised. Anything that can be owned can be represented on a ledger. The question is whether tokenisation produces liquidity, and the answer falls cleanly along the line dividing the loan from the pool. The pool can be made liquid — conditionally, by engineering — and for the most part already has been; the token makes it somewhat cheaper to trade. The loan can be made to trade, but only thinly and at a price for ignorance, and the token does not change that; what would change it is infrastructure the token does not supply. A loan is not a smaller or rougher version of a pool that better technology will smooth into tradability. It is a different kind of object — singular, opaque, and bespoke where the pool is aggregate, characterisable, and standardised — and that difference is exactly the thing liquidity requires. Tokenisation is sold on the one property it does not confer, and the properties it does confer are real, and are not the one it is sold on.