Settlement Speed Is the Wrong Margin
Faster finality — CBDCs, instant rails, mBridge — can remove at most a few percent of the working-capital cost in large-value trade. The binding constraint is the receivables cycle, and the firm-level
Faster finality — CBDCs, instant rails, mBridge — can remove at most a few percent of the working-capital cost in large-value trade. The binding constraint is the receivables cycle, and the firm-level evidence says so plainly.
KEYWORDS / TAGS: cross-border payments, trade finance, working capital, days sales outstanding, payment finality, CBDC, BRICS, receivables finance, settlement, monetary plumbing
There is a particular kind of error that is hard to see because everyone is making it at once. The cross-border payments debate is in the grip of one. We have decided that the problem with international trade is that money moves too slowly, and we have built an enormous reform agenda on top of that premise: central bank digital currencies, multi-CBDC settlement platforms, instant domestic rails bolted to cross-border corridors, China’s CIPS as an alternative to the incumbent messaging layer, tokenised deposits, stablecoins pitched as the plumbing of a faster world. The shared assumption is that settlement latency — the interval between a buyer releasing funds and a seller holding final, usable money — is a meaningful cost, and that compressing it from days to seconds will unlock something large.
I went and measured it. The premise is wrong for the trade that actually carries the value. Not slightly wrong, not wrong at the margins. Wrong about which margin is binding. And once you see the arithmetic, you cannot unsee it.
Let me state the conclusion before I defend it, because I dislike essays that make you wait. In the large-value, trade-credit-financed trade conducted by listed firms across the BRICS and BRICS-relevant economies, the seller’s money is not tied up in the payment rail. It is tied up in the receivables cycle — the months a sale sits on the balance sheet as an unpaid claim before cash arrives. Payment-rail latency is one to five days. The receivables cycle is forty-five to a hundred-and-ten. Faster finality acts only on the days; it cannot touch the months. The people selling milliseconds are solving a problem these firms do not have.
What the money is actually doing
Start with the object, because the whole argument lives in a definition that most of the payments conversation never pauses over.
When a firm sells on terms — and serious cross-border trade is almost always on terms — it ships the goods, recognises the revenue, and books a receivable. That receivable is a claim, not cash. It sits on the asset side of the balance sheet until the buyer pays, which by contract might be thirty, sixty, or ninety days later, and in practice often longer once collection friction and documentary reconciliation are added. During that interval the seller has performed but has not been paid. It has, in effect, lent the buyer the value of the shipment. That loan has to be financed, whether out of the firm’s own equity, a revolving credit line, a trade-finance facility, or a discounted invoice. The financing has a cost. That cost — the carry on working capital tied up across the receivables cycle — is the real price of moving goods on credit.
The standard balance-sheet measure of this duration is days sales outstanding. You take receivables, divide by sales, multiply by 365, and you get the average number of days a unit of sales remains an unpaid claim. It is not a precise settlement clock — it is an annual stock-over-flow approximation, affected by seasonality, revenue-recognition policy, the customer mix, and year-end timing — but it is the right order of magnitude for the holding period, and it is the figure every treasurer and credit analyst already lives by.
Now hold that next to settlement latency. Settlement latency is the last step of the timeline: the gap between the buyer instructing payment and the seller having final funds. A faster rail compresses that gap. What a faster rail cannot do is move the sixty days the claim sat as a booked receivable before payment was even released. Those days are the negotiated credit term plus the collection and documentary friction that precede the transfer. They are not a payments problem. They are a financing-and-contracting problem wearing a payments costume.
This is the whole essay in one sentence: the receivables cycle, not the rail, is the thing the seller’s capital is locked inside, and a payment rail can only ever reach the sliver at the very end.
The one number that matters
Reduce it to a ratio and the point becomes unanswerable. Call the holding period T — the full time the money is tied up — and call the rail latency L — the slice a faster rail can remove. The fraction of the cost that settlement speed can address is L divided by T. I call that λ. It is the share of the cycle that finality speed can reach, and everything turns on its size.
If T is ninety days and L is two, λ is a little over two percent. If you drove latency to zero — instant, final, free settlement, the maximum the reform agenda could ever deliver — you would remove a little over two percent of the working-capital carry and leave the other ninety-eight percent exactly where it was. That is not a rounding error you can argue your way out of. It is the structure of the problem.
And here is the property that closes the last escape route. λ is rate-invariant. The carry on the whole cycle scales with the financing spread; so does the carry on the latency slice; the spread is in the numerator and the denominator, and it cancels. This matters because the natural objection from someone defending the speed agenda is to reach for an interest-rate assumption — “but at high rates the latency cost is large.” No. At any rate, the latency share of the carry is L over T. You cannot inflate the importance of settlement speed by assuming a higher cost of capital, because a higher cost of capital raises the cost of the months by exactly as much as it raises the cost of the days. The ratio does not move. This is the single most useful thing in the analysis, and it is the reason the conclusion is robust to the parameter people would otherwise fight about.
So the policy-relevant statistic is not a dollar figure that depends on a dozen assumptions. It is a pure ratio, λ = L/T, and it is small whenever T is months. The only question left is empirical: how long is T, actually, for the firms doing the trade?
The data
I did not want to argue this from stylised facts, so I built it from firm-level accounts.
The receivables side comes from Compustat Global — standardised, consolidated annual statements — for the operating firms domiciled in eight economies central to the China–BRICS settlement debate: China, India, Indonesia, Brazil, Saudi Arabia, the United Arab Emirates, South Africa, and the Russian Federation. That is 12,241 firms and 73,533 firm-years over FY2019 to 2025. I computed days sales outstanding for every firm-year, screened to operating firms with genuine sales and a cost of production, and applied each ratio’s sane validity range so the statistic is computed on the firms that identify it rather than on accounting artefacts. The financing side comes from Dealscan — 13,031 syndicated loan facilities for borrowers in the same economies — collapsed deterministically to one all-in-drawn spread per facility, which gives the price at which these firms actually fund themselves.
The medians are not subtle. Days sales outstanding runs from about forty-five days in the Russian Federation to about a hundred-and-eight in China, with the others — India in the low seventies, Indonesia in the mid-fifties, Brazil around eighty, Saudi Arabia in the mid-eighties, the UAE near a hundred, South Africa in the low fifties — strung between. Financing spreads run from roughly 145 basis points over base in Saudi Arabia to 441 in China. Put the duration and the spread together and the annual working-capital carry on a unit of trade value lands between about twenty-nine and a hundred-and-thirty basis points, depending on the economy.
Against all of that, set a latency of two to five days. The latency share λ comes out at a median of about 2.6 percent at a two-day finality and about 6.5 percent at five days. The largest value anywhere in the panel — in the Russian Federation, which has the shortest cycle and therefore the least room — is eleven percent. Eleven percent is the ceiling, in the one economy most favourable to the speed story, at the most generous latency assumption. Everywhere else, the number a faster rail can touch is low single digits.
This is the empirical heart of it. The cross-border payments reform agenda is optimising a variable whose maximum achievable effect, in the trade that matters, is in the low single digits of a cost that is itself only one of several the agenda does not address.
It is not just the median
A median can hide a tail, and a careful reader will immediately ask whether the distribution has a meaningful mass of firms with very short cycles — firms for whom settlement speed really would dominate. So I looked at the whole distribution, not the midpoint, and reported the exact share of firm-years below each threshold that the theory makes relevant.
The thresholds come straight from the ratio. A two-day latency only reaches even ten-percent relevance once the holding period drops below twenty days; a five-day latency needs the period below fifty. So the right question is: how many firm-years actually sit below those lines?
Below twenty days, the share ranges from 2.6 percent in Brazil to 17.2 percent in Indonesia. Below fifty days — the five-day-latency threshold — it ranges from about nineteen percent in the UAE to about fifty-five percent in the Russian Federation. And below ten days, the region where finality speed would genuinely dominate the cost, the share is at most 9.1 percent in any economy. The overwhelming mass of firm-years sits well to the right of the latency band, in every single economy. The distribution does not rescue the speed thesis. It buries it.
I find it useful to picture it. Lay each economy’s distribution of receivables periods along a horizontal axis measured in days. Shade the one-to-five-day band where rail latency lives. In every economy, the bulk of the distribution — the interquartile box, the median, most of the tails — sits dozens to well over a hundred days to the right of that shaded sliver. The gap between where the money is and where a faster rail can reach is not a matter of interpretation. It is visible at a glance: months on one side, days on the other.
“But the cycle is mostly voluntary”
The sharpest objection to all of this is not about the data. It is conceptual, and it deserves a real answer rather than a brush-off. It goes: you are comparing latency against the entire receivables cycle, but most of that cycle is a voluntary commercial decision. The seller chose to offer ninety-day terms. That credit period is a product feature — a financing service bundled into the sale — not a friction. Comparing a settlement cost against a voluntary credit term is comparing a cost against a choice.
This is a good argument and it is partly right. Some of the cycle is genuinely chosen. But it does not get the speed agenda where it needs to go, for two reasons.
First, even the voluntary part is a cost the seller bears and finances, and the question on the table is which instrument reduces it. A faster rail does not shorten a negotiated credit term by one minute. So whether the term is voluntary or not is beside the point when you are asking what payment-rail speed can accomplish: the answer is nothing, because rails do not touch terms.
Second, and more carefully, I separated out the part of the cycle that is plausibly involuntary — the collection lag and the documentary and reconciliation friction that sit on top of the contractual term — and computed the latency share against that frictional residual alone, bounding it conservatively two ways because the residual is not directly observed. Even measured against only the involuntary friction, the latency share is in the mid-single digits. The conclusion survives the strongest form of the objection. Settlement speed is a small lever even against the part of the cycle nobody chose.
Does the composition trick you?
The other serious challenge is statistical rather than conceptual. Days sales outstanding varies a great deal across industries — a heavy-equipment maker selling on long terms looks nothing like a supermarket — and the mix of listed industries differs across economies. So perhaps the long cycles are an artefact of which sectors happen to be listed where, and the cross-economy story is really a sector-composition story in disguise.
This is the kind of objection you cannot wave away; you have to do the work, so I did. I pulled the industry classification and recomputed the receivables cycle four ways: manufacturing only, with construction and real estate excluded, with financial firms excluded, and with the industry effect removed altogether by adjusting each firm-year against its own industry’s global median. Across every one of those cuts, every economy’s median receivables cycle stays between roughly forty-five and a hundred-and-thirteen days. The lowest median under any restriction — the Russian Federation, again — is still more than twenty times a two-day latency. The order-of-magnitude result does not turn on sector composition. Months stay months whichever way you slice the listed sector.
I will be honest about what sector composition does affect, because the discipline of the exercise is to say exactly what survives and what does not. The cross-economy ranking — whether China’s cycle is longer than the UAE’s, and by how much — does shift under industry adjustment. So I treat the ranking and the within-economy size gradient as descriptive colour, not as identified findings. What I claim is the robust thing: the gap between a months-long cycle and a days-long latency, in every economy and under every reasonable cut. Not the league table. The league table is not the point; the order of magnitude is.
One more robustness check, because it pre-empts a methodological complaint: days sales outstanding strictly needs only sales and receivables, not the cost-of-goods screen I used to define operating firms. So I recomputed it the other way — sales only, financial firms removed by industry code instead — and every economy’s median landed within about a day of the original. The screen is not driving the result. The result is in the data, not in my filters.
The dollars
Ratios persuade analysts; magnitudes persuade everyone else. So I scaled the comparison to trade flows, with the explicit warning that this is an illustration of order of magnitude and not a welfare estimate or an observed bilateral settlement cost.
Take China’s goods trade with seven of these partners in 2024 and apply the economy-level working-capital terms. The annual carry on the receivables cycle implied by that exercise is on the order of three-and-a-half to four billion dollars. The slice of it attributable to a two-day settlement latency — the part a perfect, instant, free rail could remove — is about a hundred-and-ten million. Three-point-seven billion against zero-point-one-one billion. The cycle carry is real money, and the point of the paper is emphatically not that it is small; it is that the instrument the speed agenda offers reaches the wrong end of it. Compressing finality removes the hundred-and-ten million at most. The other three-and-a-half billion is the financing cost of the credit term and the non-payment friction, and a faster monetary rail does not reach it.
If you wanted to act on the three-and-a-half billion, you would not build a faster rail. You would do something to the holding period or to the spread — shorten the documentary and collection friction, or deepen the market that discounts the receivable. That is a different toolkit entirely, and it is the toolkit nobody is funding with the enthusiasm reserved for settlement speed.
Where the speed agenda is actually right
Now I want to do the thing that separates an argument from a polemic, which is to say precisely where the opposing view is correct — because it is correct, in a regime this trade does not occupy, and the boundary between the two regimes is the most interesting part of the whole problem.
The entire result is conditional. λ is small because T is large. T is large for a specific, identifiable reason: this trade is relationship-based and credit-financed. The buyer’s promise to pay in sixty or ninety days is itself the instrument; the receivable is an asset precisely because there is a counterparty with a balance sheet standing behind it and a contract worth enforcing. Strip those features away and the whole structure inverts.
Consider a transaction with a counterparty that has no balance sheet to extend credit against, a value too small to negotiate terms over, and a delivery metered in seconds rather than quarters. There is no receivables cycle there, because there is nothing to finance and no one to finance it. The holding period collapses toward the settlement interval itself. T falls toward L, and λ — latency over holding period — rises toward one. In that regime, settlement speed is not a small margin. It is the entire margin. The cost of delay stops being a financing carry on a float and becomes the foregone throughput of an operation that cannot proceed until finality clears.
The limiting case is machine-to-machine settlement: autonomous systems paying each other for compute, bandwidth, energy, data, or physical priority at the moment of use, in fractions of a cent, with no credit relationship and no negotiated term. There, finality speed is the binding constraint, because finality is the holding period. The same arithmetic that makes latency negligible for a ninety-day, million-dollar receivable makes it decisive for a microsecond, sub-cent machine payment.
So the people building for instant, final, micro-value settlement are not wrong. They are building for a different point in the same parameter space — a regime defined by the joint collapse of value and duration. The error is not that they value speed. The error is importing a settlement architecture optimised for the machine regime into the analysis of large-value trade credit, where the parameters are the opposite and the optimisation target is therefore the opposite. An architecture tuned for one is not thereby tuned for the other. Today’s cross-border trade between firms with balance sheets, long terms, and large invoices is squarely in the regime where latency is the smallest lever. Tomorrow’s machine economy may not be. Both can be true, because they are different coordinates, and conflating them is exactly the mistake the current debate makes.
The policy that would actually move the number
If settlement speed is the wrong lever, the constructive question is which lever is right, and the decomposition answers it directly. Rank the instruments by the margin each one acts on, and the ordering falls out of the cost structure rather than out of fashion.
First, and largest, is the financeability of the trade claim. Anything that lets a verified receivable be discounted cheaply — invoice verification, receivables registries, enforceable assignment of the claim, credit insurance — acts on the financing spread, and the spread is paid on every single day of the cycle. Lower the cost of turning a receivable into cash and you lower the carry across the entire holding period at once. This is the instrument with the most leverage, because it attacks the price of the whole thing rather than a few days at the end.
Second is documentary and legal transferability. A great deal of the involuntary friction in the cycle is documentary state that cannot be discharged or financed without manual reconciliation — bills of lading, warehouse receipts, the paper apparatus of trade. Legal recognition of controllable electronic trade documents, of the kind the UNCITRAL Model Law on Electronic Transferable Records contemplates, compresses that frictional residual. This is plumbing worth building, but notice it is the plumbing of documents and title, not the plumbing of payment finality.
Third is counterparty-risk compression — export-credit guarantees, buyer payment undertakings, standardised confirmation — which acts on the contractual term by making the credit relationship safer to extend and therefore shorter and cheaper to carry.
And fourth, last, is payment-rail speed. Not zero — it is not nothing — but last, because it acts only on the latency slice, which is the smallest direct component of the cost this trade actually bears.
I want to be clear that “last” is not “useless.” Faster, more resilient settlement can matter for reasons this analysis does not measure: resilience against the weaponisation of payment infrastructure, sanctions exposure, liquidity management, financial inclusion, reach into populations the banking system does not serve. Those can be excellent reasons to modernise rails. They are simply not the same as reducing the working-capital cost of large-value trade, and the case for rail modernisation should be made on its real merits rather than on a working-capital benefit the arithmetic does not support. Match the instrument to the margin, and the margin says: finance the receivable first, digitise the documents second, de-risk the counterparty third, and speed the rail fourth.
What I am not claiming
Rigour is mostly a matter of refusing to claim more than you have shown, so let me draw the lines clearly, because they matter as much as the result.
This is descriptive measurement, not causal identification. I measured the lock-up and contrasted it with the latency. I did not estimate the effect of adopting any particular settlement architecture, and I would not, because the data cannot support it. The most tempting natural experiment — the 2022 exclusion of Russian banks from the conventional rails — bundles the settlement-access shock with sanctions, asset freezes, and counterparty exit so thoroughly that no honest researcher could isolate a clean settlement effect from it. The Russian receivables cycle did shift over that period; that is consistent with a settlement effect and is not evidence of one. I report it and decline to over-read it.
The sample is listed firms, by construction. The segment most exposed to settlement friction — small and unlisted exporters — does not appear in Compustat at all. I corroborated the mechanism in that segment using the World Bank Enterprise Surveys, which show working capital financed heavily through inter-firm credit and access to finance among the most frequently cited obstacles, more so for smaller firms. But survey evidence on financing structure is consistent with the mechanism; it does not measure settlement cost and it does not prove that latency is immaterial for an SME exporter. I will not pretend it does.
And the model is an accounting identity, not a structural estimate of behaviour. It decomposes the cost and shows which component each instrument touches. It does not predict how firms would re-optimise under a new settlement regime. That is a different paper, and it would need transaction-level settlement data nobody has handed me.
What I am claiming is narrow and, I think, solid: in the observed large-value, trade-credit-financed, listed-firm regime, the receivables cycle is months, the rail latency is days, the share a faster rail can reach is low single digits, and that share is rate-invariant and robust to sector composition and to the screens I used. The redirection follows from the measurement. The treatment effect does not, and I do not assert it.
The discipline
The reason this matters beyond one corner of trade finance is that it is an instance of a general failure: choosing an instrument before measuring the margin it is supposed to move. The cross-border payments agenda picked settlement speed — visible, technically exciting, fundable, easy to demonstrate on a stage — and then went looking for the benefit. When you do it in that order, you end up optimising the variable that is easiest to move rather than the one that is binding, and you can spend a decade and a great deal of capital making a number smaller that was never the problem.
Do it in the other order. Decompose the cost. Find the component that actually carries the weight. Then match the instrument to that component. For the trade that moves real value across these economies today, the binding cost is the financing of a months-long receivable, and the instruments that reach it are the ones that make the claim cheaper to finance and faster to clear of its documentary and contractual friction. Settlement speed will have its decade, and it will be the decade of the machine economy, where value and duration collapse together and finality becomes the whole game. That decade is coming. It is not this trade, and it is not now.
Measure the margin before you choose the instrument. Everything else is engineering in search of a problem.