Bonds are called the easy case for tokenisation — the coupon was never the hard part; paying it to every holder, on a calendar the ledger does not keep, is
Plain-vanilla bonds are routinely described as the simplest asset to tokenise: a coupon, a redemption, a calendar, and nothing to vote on. Strip the label and the convention rests on two pieces of infrastructure that nobody prices separately — a legal calendar that defines what a day is, and a fan-out that turns one issuer payment into thousands of holder payments.
Tokenisation removes neither. A 24/7 ledger runs on a clock the bond's terms do not recognise, and a stablecoin coupon still has to reach every holder. The intermediated chain bundles the per-holder cost into custody fees where the holder cannot see it; the token can make the network fee a separately visible charge, and the design chosen decides who pays it — in the unsponsored claim model, the holder. The celebrated native issues have not solved the fan-out. They have sidestepped it.
Ask anyone selling tokenisation infrastructure where to start and the answer is fixed income. The June 2026 BCG and Anchorage Digital playbook for banks puts it plainly: bonds are a strong candidate because of standardised structures, high issuance volumes, and manual workflows such as distributions and reporting that would benefit from programmability and automation (BCG and Anchorage Digital, 2026). One of the two authors sells the stack it recommends, but the claim is not theirs alone. When the ASX convened its fixed-income working group in September 2026, asset servicing came out as one of the strongest value propositions on the table: automate coupons and redemptions, cut manual processing and reconciliation, reduce error rates (ASX Tokenisation Industry Working Group, 10 September 2026). A workshop's value proposition is a hypothesis, not a measured saving; it is cited here as evidence of what the industry believes, which is the thing under examination.
The same workshop attached a condition that deserves more attention than it got. The benefit, participants said, depends on resolving which register of ownership is authoritative and whether the legal agreement or the smart-contract logic prevails when the two disagree. That condition is this essay.
What is striking about the servicing claim is how little has been measured. Two prominent empirical studies of tokenised bonds — the BIS in July 2025 on government issues and the ECB in April 2026 on a matched sample of mostly European corporate and financial bonds — look at issuance and trading: underwriting fees, yield spread at issue, bid-ask spread afterwards (Aldasoro et al., 2025; Born et al., 2026). Both find lower bid-ask spreads. Neither finds a saving at issuance; the ECB's underwriting fees come out four hundredths of a point higher, without significance. Neither measures what it costs to pay a coupon to a beneficial holder. That is not proof that nobody has solved it operationally. It is proof that the saving most often claimed has not been measured by the studies most often cited for it.
Part of the reason is in the exhibits. The European Investment Bank's 2021 digital bond, the most frequently cited early native issue, was a zero-coupon note. There was no coupon to pay. Siemens' 2023 bond on a public chain, the first under Germany's electronic-securities law to skip central clearing, was placed directly with three institutional investors (Siemens, 14 February 2023). There were three payments to make. The fan-out — the thing that makes coupon servicing a business rather than a bank transfer — was not tested by the headline native issues. It was sidestepped by them.
This is not a sequel to The Dividend Is the Hard Part, but it does narrow one of that essay's concessions. That essay covered the servicing nobody calls easy — elections, entitlements, the corporate action with options — and granted the coupon as the one event class that is genuinely code, for the wholesale holder base it had in view. This essay holds the concession to that base and withdraws it for the fragmented one. It covers the case everyone calls solved.
A coupon looks like a number. It is a schedule indexed to a legal calendar. The terms specify a day-count fraction, a business-day convention, a holiday calendar — TARGET for euro, SIX for Swiss francs — and two dates that are not the same date. The record date determines who is entitled. The payment date determines when the entitled are paid. Under the European market standards for corporate actions processing, the convention for debt distributions is a record date one business day before the payment date and no ex-date: entitlement follows the settled position at close on the record date, read together with the market rules on claims and failed settlement. These are market standards, not statute, and instrument terms can depart from them (CAJWG market standards).
Entitlement and payment are therefore separate events with separate failure modes. A trade executed before the record date but settled after it leaves the coupon with the seller's settled position, and the infrastructure repairs the mismatch with a market claim: the CSD's claims process reallocates the coupon between participants, from the seller who received it to the buyer who was owed it, and the custodians below pass it on. Accrued interest itself runs to the settlement date, not the trade date, under the applicable day-count and settlement conventions — which is why every trade settling on the same day carries identical accrued interest regardless of when it was executed, a fact this essay will need again.
None of this is in the token. It is in the terms, and the terms were written for a calendar.
Consider three ways to pay a coupon to N holders. They are operating models, not legal categories.
In the first, the issuer administers the register itself. It knows its holders by name, holds a KYC file on each, executes N bank transfers on the payment date, chases the returned payments, and issues N tax statements. This is manageable at twenty holders. It is absurd at twenty thousand. Private placements and Schuldscheine live here; retail bonds do not.
In the second, the issuer pays one amount to one paying agent and, where the terms so provide, is discharged. The paying agent passes it to the central securities depository, which credits each participant according to its settled position on the record date. Each custodian then allocates its participant credit down its own books to the beneficial owners it knows and the CSD does not. Omnibus accounts mean the CSD sees a few hundred counterparties; the chain below it sees millions. The market claim mechanism sits here too, repairing the trades that straddled the record date. The issuer's obligation ends at the paying agent. Everything below is someone else's liability, governed by a rulebook that has been tested.
The third way is the one some platform-led implementations under Germany's electronic-securities law and Switzerland's ledger-based securities have adopted. A register operator — a licensed crypto-securities registrar, a platform — keeps the register and runs the distribution as a service, two functions that are legally distinct and commercially bundled. The issuer still pays once. The platform fans it out and invoices the issuer a fee for doing so. The fan-out was not deleted by tokenisation. It was re-intermediated by a thinner intermediary, a theme Who Keeps the Register? treats as the contest tokenisation is actually about.
These are archetypes, not a census of the market, and the comparison between them has to use the same perimeter: not just the transfers, but register maintenance, entitlement computation, reconciliation, failed payments, tax documentation and legal responsibility for getting it wrong. On that perimeter the question is not whether the per-holder cost exists. It plainly does in all three. The question is where it sits and whether anyone can see it.
In the second model, the answer is that it sits inside a bundled custody fee and nobody can see the coupon's share of it. Retail custody pricing is a mixture: a flat fee per securities line, a per-event charge for corporate actions, and an ad valorem safekeeping fee on top. The coupon's processing cost is in there somewhere — per line, per event, cross-subsidised or priced into the relationship — and the holder pays it as part of a number that does not say what it is for. In the third model, the cost sits on an invoice to the issuer, who recovers it from the coupon, the issue price or the platform's own pricing. Visible to the issuer, at best indirectly to the holder.
Now replace the paying agent with a smart contract and the cash with a stablecoin. There are three ways to do this, and they differ in exactly the dimension the previous section identified: who pays the per-holder cost, and whether they can see it. On a permissioned ledger that cost may be an internal accounting entry rather than a market-priced fee; on a public network it is gas, and gas is the clearest case, so it is the one worked through here.
The first is the self-administered register, automated. The issuer's distribution address sends one stablecoin transfer to each holder. Cost scales with the number of holders. The issuer bears it, as gas, one transaction at a time.
The second is a batch distribution contract. The issuer funds the contract once and calls a function that loops over the holder list. This saves the per-transaction overhead but not the per-holder balance update in the stablecoin contract, which on most implementations is where most of the gas goes, and block gas limits mean a large holder list is several batches rather than one. Still roughly linear in N, still the issuer's cost, cheaper per head.
The third is the claim model, which the crypto-native world calls a Merkle distributor. The issuer computes the entitlement file off-chain, publishes a cryptographic commitment to it, and funds the contract with the total. In the unsponsored version the issuer's on-chain cost — root publication and funding — is now largely independent of N. Each holder then submits a claim and receives their coupon, and each claim costs gas. Who pays it is a design choice: the issuer or platform can sponsor it, a custodian can claim on behalf of its clients, or the holder can be left to pay. Sponsored, the cost scales with claims again and sits with whoever sponsors. Unsponsored, the per-holder cost has not disappeared. It has been moved from the issuer to the holder, and it has become a separately visible, wallet-level charge rather than a share of a custody fee. The issuer's own cost is not zero either — proof generation, support, lost-key recovery, address changes, sanctions screening, failed and duplicate claims, reconciliation — but it no longer scales with N.
At this point the arithmetic matters. A claim is rational only if the coupon exceeds the cost of claiming it. For a holding H at an annual coupon c paid in periods of fraction τ, the gross coupon-versus-gas threshold is H* = G / (c × τ), where G is the gas cost of one claim. At a five per cent semi-annual coupon, a claim costing two euros is worth making on holdings above eighty euros; a claim costing five cents — an illustrative layer-two figure, not a market constant — is worth making above two euros. The threshold ignores tax, conversion and the holder's time; where the distributor lets unclaimed periods roll into a later root, one claim collects several coupons and the effective threshold falls further.
So the honest version of the cost argument is narrower than the one usually made against tokenisation. On a layer-two network or a permissioned ledger, the per-head execution cost of paying a coupon is cents or nothing, which is plausibly below a custodian's per-event charge where one is itemised at all. The token does not lose on cost, and automation may well lower the all-in cost even though the fan-out survives. What it changes is visibility and allocation. A charge nobody saw becomes a charge somebody pays, and under the unsponsored claim model that somebody is the holder, who is also the only party in the structure with no say in which model was chosen.
Two consequences follow that no prospectus for a tokenised bond yet states. The first is that where the holder bears the claim gas, the net coupon a small holder receives is not fixed at issue: gas is a cost term, however small, and it is volatile. The second is what happens to coupons that are never claimed. Under the unsponsored claim model, amounts below the threshold sit in the distribution contract. Statute sets an outer limit — the claim to periodic interest is time-barred after five years under Swiss law, and a German bearer coupon carries a four-year presentation period after which the right lapses, two rules of different kinds (Art. 128 CO; §801(2) BGB) — so the terms need not say when. What the terms rarely say is who controls the funded contract meanwhile, whether the float earns anything, who receives it, and where the balance goes when the claim is barred. A funded contract is not automatically the issuer's money; it is not automatically the holder's either. It is whatever the terms, and the mandatory law behind them, make it, and most terms are silent.
Under an unsponsored claim model, the smallest holders leave their coupons in a contract they will never find it worth opening — and the terms rarely say who earns the float.
Ask who benefits from the gap and the answer is in whichever clause nobody drafted.
The fan-out is one of the two pieces of hidden infrastructure. The calendar is the other, and it is the one tokenisation disturbs most directly, because a ledger that never closes has no native concept of a business day.
Start with the record date. The terms say entitlement is fixed at close of business on a date. The ledger has no close of business. This is a governance question, not an impossibility: the terms can reference a conventional calendar and specify how the snapshot is taken. But someone has to map the legal date to a point on the chain, and the terms, not the chain, have to say which time zone, which chain and consensus rules count, which finality threshold applies — finality may arrive after the cut-off, and a block timestamp is not a trusted clock — and what happens if the ledger is unavailable or reorganised. Freezing transfers across the cut-off would defeat the purpose of continuous trading, so the practical mechanism is a checkpointed balance or an off-chain replay of transfer events to the chosen block, with the resulting entitlement file committed on-chain. This works. But it is a convention written into the terms, and if the terms are silent, the chain's answer and the law's answer can differ.
Then weekends. Calendar-day conventions already accrue interest on Saturdays; an atomic trade settled on a Saturday simply settles on Saturday, and the buyer earns the day. The difficulty is the cash leg, if the cash leg is central bank money. A stablecoin or commercial-bank-token leg exists on Saturday. The wholesale central-bank-money settlement service does not: T2 and T2S run on their business days, and the Eurosystem's Pontes link for DLT settlement runs on theirs. Atomicity — the simultaneous and inseparable exchange of security and cash — is advertised as the technology's gift. On a Saturday it breaks on the calendar before it has a chance to break on the code.
Finally, accrual granularity. If a bond trades and settles continuously, should interest accrue by the minute rather than by the day? The arithmetic says it does not matter economically: at a five per cent coupon, a minute of accrual is on the order of a hundred-thousandth of a price point, thousands of times smaller than intraday price noise and never larger than the trading spread. What does matter is a discontinuity in the quote, and it is worth being exact about where it comes from. The settlement consideration of a bond is clean price plus accrued interest, and accrued interest under every standard convention is counted in whole days to the settlement date. In a market where every trade settles on the same future date, that is invisible: everyone's accrued is the same number. Under instant settlement the accrued component steps up by one day's interest at each day boundary while the bond's economic value does not, so the clean quote — consideration less accrued — must step down by the same amount to keep the consideration continuous. Nobody earns a free day. But the clean price acquires a daily sawtooth of one day's coupon, about one and a half basis points of face at five per cent, which is the order of a government bond's spread and a shape no quoting convention has had to carry. The fix is not minute accrual, which would require redefining "day" in every day-count convention the market uses. One workable design is a single contractual day boundary in a stated time zone, tied to the register's own timestamp, so that at least the step is deterministic and the same for everyone — in price display, in entitlement and in the tax record. Daily accrual survives. The ambiguity does not.
Every mechanism above assumes the issuer knows who to pay. On-chain, the issuer knows an address.
Where that matters first is withholding, and the two neighbouring regimes answer it in opposite places. In Switzerland the withholding tax on domestic bond interest is a duty of the debtor: the issuer deducts it before the coupon leaves, so the token changes nothing, because the deduction happens before the fan-out. In Germany the tax on investment income is withheld by the paying institution — ordinarily the custodian that credits the holder's account (§44(1) EStG). A self-custody holder of a tokenised bond has no custodian in that sense, and the obvious inference is that the withholding point has vanished with it. The statute forecloses the inference. For crypto-securities under the electronic-securities law, §44(1) sentence 4 no. 6 EStG makes the register-keeping entity the paying office where no custodian or other listed office exists. The German legislator did not let the withholding agent disappear; it rebuilt the agent inside the registrar — the thinner intermediary of the third model — and handed it the identification, deduction and reporting that the custodian used to do. That is this essay's argument written into a tax statute: the operational point where identity met payment was not removed by the ledger, it was relocated to whoever keeps the register. Which structures fall under the provision, and what a registrar must do to discharge it, are questions for the instrument and its counsel.
Where it matters second is omnibus. The intermediated chain's problem was that the CSD saw participants, not beneficial owners. The token was supposed to fix this by making every holder a register entry. It does, until the holder deposits the token in a lending pool, parks it in a bridge, or leaves it in an exchange's hot wallet. The register then shows the contract as holder, and whatever record maps the beneficiaries behind it sits outside the token's own transfer history and, for the instrument's purposes, records nothing. Intermediation was not removed. It moved into code, and the code does not keep the books that matter.
This is also where reconciliation lives, and it is worth one paragraph rather than a specification. A plain stablecoin transfer carries no payment reference; it is money arriving with no label. One interoperable design for a tokenised coupon that an incumbent back office can reconcile is a distribution contract that emits a holder-level event for each allocation, whether the allocations are executed in a batch or separately — instrument identifier, token identifier, corporate-action reference, record and payment dates, rate, gross, withholding, net — with the entitlement file committed on-chain and the data mapped to the ISO 20022 corporate-action notification and movement-confirmation messages that custodians already consume, under their usage guidelines (ISO 20022 seev.031, seev.036). None of this is hard. All of it is the paying agent's job, rewritten.
The identifiers deserve a sentence of their own, because they reveal an assumption. The ISIN still identifies the instrument, allocated by the issuer's numbering agency exactly as before; the Digital Token Identifier, a separate ISO standard with a separate registration authority, identifies the token's implementation on a specific ledger. One ISIN can carry several DTIs grouped as functionally fungible, for the case where the same bond is tokenised on more than one chain (ANNA and DTIF, 2024). The standard, in other words, already assumes that one instrument may be represented on several ledgers. Whether those representations are legally interchangeable is a question for the terms and the governing law; what is certain is that the entitlement snapshot then has more than one register to read. The fan-out fragments across ledgers before it reaches a single holder.
And discharge. A coupon paid in a stablecoin discharges the issuer only if the governing terms and applicable law recognise that transfer as payment. The regulatory status of the token is a separate question — for a euro stablecoin in the EU, whether it is an e-money token under MiCA — and answering it does not answer the first. The terms have to say what counts as valid tender, whether payment to a designated address discharges, and what follows if the stablecoin issuer has blacklisted the holder's address — a reattempt, an alternative asset, escrow, deposit with a court. Where they say nothing, the obligation stays open to a holder the issuer cannot pay.
Two books now run side by side. The settlement consideration carries accrued interest to the settlement moment; the tax book counts whatever its own rules count. Where the two rules differ, instant settlement opens a window, and Switzerland is the regime where the window has always been widest. For a private investor holding a bond as private wealth, the coupon is taxable income, but the accrued interest received on a sale between coupon dates is, under long-standing practice, part of a tax-free capital gain — the exception being bonds with predominantly one-off interest, which are taxed on the difference (ESTV, Kreisschreiben Nr. 15). A bond sold the day before the coupon and bought back the day after converts a taxable coupon into an untaxed gain for the seller and a deductible cost for the buyer. The amounts on one trade are trivial. The classification is not, and instant settlement makes the trade free of the settlement risk that used to accompany it.
The bond market's own name for this is older than tokenisation. Bond washing sells before the record date and buys back after, converting a coupon into a capital gain for whoever is better placed to receive one. The equity cousins — dividend stripping, cum-cum, and the cum-ex schemes that turned one withholding into several refund certificates — belong to The Dividend Is the Hard Part and are not re-argued here; they are different phenomena, and only the last is fraud as such. What is worth asking is which of them a single register with atomic settlement actually prevents.
It can prevent the ones that needed a settlement window. Cum-ex worked because a short sale straddling the record date created a gap between legal ownership and settled position, and a custodian issued a tax certificate on each side of the gap, so that the certificates outnumbered the tax. Collapse the window; make the ledger record the one the tax authority treats as authoritative; and issue the certificate only against that record. Under those three conditions — and they are conditions, not consequences of the technology — each unit of the bond has exactly one settled holder at the cut-off and one certificate. This is a structural gain, not merely a visibility gain, and the people who say tokenisation makes markets more transparent are, for this narrow class of abuse, right for a reason better than the one they give.
It does not prevent the ones that needed a legal form. Cum-cum and bond washing were never hidden. The trades were real, correctly booked, visible to anyone who looked. The arbitrage lay in who counted as the beneficial owner on the record date, and in what the tax law made of accrued interest, which are questions of law, not of ledger. A perfect register records the arbitrage perfectly and prevents none of it.
And visibility is not attribution. The tax authority that can read a public chain sees flows between addresses. It sees taxpayers only where a permissioned token's identity registry is open to it. On a public chain with self-custody, the trail is complete and the person is missing.
Tokenisation closes the frauds that depended on the books disagreeing, not the ones that depended on the law agreeing.
It is tempting to read all this as a defence of the incumbent, or as a verdict on tokenisation. It is neither. Tokenisation's case rests on more than servicing — settlement speed, collateral mobility, programmability — and this essay tests one claimed benefit, not the whole. What it corrects is the account of what the incumbent was selling.
The CSD's coupon product was never messaging, and it was never the fan-out as such. It was discharge plus a rulebook: one payment, one discharge where the terms so provide, and a governed process under which every dispute about who was entitled — the straddling trade, the failed settlement, the wrong account — is resolved by procedure rather than litigation. The issuer's primary obligation stays where the terms and the law put it; the infrastructure makes everything below it predictable. The platform in the third model of Section III sells the same product with a narrower or less-tested rulebook. The token, in any of the three designs of Section IV, supplies the mechanics. Unless the terms, a regulated register operator or an incumbent custody chain supply the rulebook around it, the ledger alone does not.
Where the token wins is clear enough. Cash-leg speed. Fewer market claims, because atomic settlement removes the settlement-lag cases that generate most of them, though not the need for an entitlement rule and an exceptions process. A transparent entitlement file that any holder can verify against the committed root. And, for a concentrated institutional holder base on a cheap network, a marginal execution cost per payment near zero. For a wholesale placement with forty holders, the easy case is actually easy.
Where it is untested is the fragmented retail holder base — which is, inconveniently, the distribution case tokenisation is marketed for. There is one signal in the data and it should be read as a signal rather than evidence: in the ECB's matched sample, the liquidity gain reverses for the tokenised bonds accessible to retail investors, which show wider spreads than their conventional matches. The authors are explicit that three bonds drive the result and that it should not be generalised (Born et al., 2026). It is consistent with everything above, and it proves nothing.
The falsifier is therefore a disclosure, not a number. A natively tokenised bond with several thousand beneficial holders that publishes its all-in coupon-servicing cost by function — distribution, sponsored gas, reconciliation, exceptions, tax documentation, support — and by who bore it, issuer, platform, custodian or holder, broken down by holder size, together with its unclaimed balances, would let the comparison be made like for like. If that disclosure showed the smallest cohort paying no more per euro of coupon than the largest, the incidence argument of Section IV would be wrong for that bond. Nobody has published such a disclosure. Until someone does, the easy case is an assertion.
Bonds are the easy case only if the calendar and the fan-out are free. They never were. The intermediated chain hid their cost inside custody fees nobody itemises. The platform re-invoices it to the issuer. The token can itemise it and, in one widely admired design, hands the itemised bill to the smallest holder in the structure — the one party that did not choose the design.
The question was never whether a coupon could be automated. A coupon is a schedule and a schedule can always be automated. The question was who would be sent the bill for paying it, on which clock the bill would come due, and whether anyone would measure the whole of it.
The coupon was always the easy part. Paying it was not.
Written in a personal capacity. Analytical views only — not legal or investment advice.
Julian Gretzinger — Investor and writer on monetary history, real wealth mechanics, and financial markets. substack.com/@juliangretzinger