The correspondent banking network has been contracting for more than a decade. The most recent publicly available CPMI and SWIFT series, which runs through December 2022, shows active correspondents declining in every region, with cumulative falls since 2011 ranging from roughly 19% in Northern America to around 47% in the Americas excluding Northern America.
If you don’t live in this terminology, the arrangement is simple. A correspondent bank holds deposits for another institution and moves money internationally on its behalf, which means the second bank doesn’t need its own account relationships in every market. That second institution is the respondent. When a correspondent withdraws, the respondent hasn’t lost a supplier. It’s lost part of what it can offer its clients.
The industry’s answer has settled into something close to consensus. Faster rails, better messaging standards, tokenised settlement. Every one of those addresses a real friction and we’ve written favourably about several of them.
Our concern is narrower. These developments are being asked to carry more weight than they can bear.
Why These Relationships End
It’s worth being precise about the causes, because they determine what can be done about it.
Correspondent relationships end for several connected reasons. Compliance costs rise. Transaction volumes on a corridor turn out to be too thin to justify the overhead. Risk appetite shifts. Institutions rationalise their business models. Sometimes there are genuine concerns about the quality of AML and sanctions information available from the other side.
What that list has in common is that most of these decisions are taken at portfolio level, against a corridor or a category, rather than in response to any individual client. A well-run bank with a solid corporate book can find its access reduced without a single thing changing on its side.
There’s a second effect that gets less attention. As relationships have fallen away, message volumes have kept rising, which means the same flows now move through fewer institutions. Where a market is served by a handful of correspondents, losing one isn’t a matter of finding another.
What the Evidence Shows
For a long time this was argued from principle. It isn’t any more.
Research by Lea Borchert, Ralph De Haas, Karolin Kirschenmann and Alison Schultz, conducted with the EBRD and published this year in the Review of Financial Studies, traced what happens to companies when their bank loses correspondent access. The main analysis covers four markets, Bosnia and Herzegovina, Croatia, Hungary and Turkey, with a broader exercise extending across eighteen emerging European countries.
The findings are consistent and they persist. When a firm’s main bank lost correspondent relationships, the firm became significantly less likely to keep exporting, and the effect was larger four years later than immediately after the event. Among businesses that survived, revenues and employment were both lower than comparable firms whose banks were unaffected. Smaller companies fared worst, for the straightforward reason that they found it hardest to move their banking elsewhere.
These are strong estimates from a specific set of markets and shouldn’t be read as a universal law. But the direction is difficult to argue with.
The finding that matters most for anyone thinking about network structure is this one. It wasn’t simply having a correspondent relationship that protected a firm. It was whether its bank had alternatives. Companies banking with institutions that held relatively few correspondent relationships saw far sharper falls in export activity than those whose banks were more diversified.
The researchers reached that conclusion independently. We’d have made the same argument, and it carries more weight coming from them.
What New Rails Change, and What They Don’t
It helps to be clear about which layer of the problem each development addresses.
Rails move and settle value. Institutional use is expanding and the numbers are no longer trivial, with Visa reporting in April 2026 that its stablecoin settlement pilot had reached a $7 billion annualised run rate across nine blockchains. Worth noting that Visa still describes it as a pilot.
Standards improve information exchange. ISO 20022 became the required format for Swift cross-border instructions when coexistence ended in November 2025, and it allows richer, more structured data to travel with a payment. The benefit depends on that data being captured properly at source, and the work isn’t finished. Fully unstructured address formats aren’t due to be retired until late 2026.
Banks provide something different again. Licensed access in a market, a balance sheet, the appetite to take a position on a counterparty, and the capacity to support trade finance instruments.
These layers are complements, not substitutes. New rails can genuinely reduce cost, delay and information friction, and the BIS has argued that tokenisation could reduce duplication across correspondent networks rather than displace them. What no rail supplies on its own is regulated local access, institutional risk appetite, or a working relationship between two banks that trust each other’s controls.
The companies in that study didn’t lose the ability to send a message. They lost access to banking channels that supported international payments and trade finance, including, depending on the transaction, the advising or confirmation of letters of credit.
Why More Relationships Isn’t the Whole Answer
Here’s where the argument usually stops, and where it shouldn’t.
If diversification limits the damage, the obvious response is for banks to hold more bilateral correspondent relationships. That helps. It also gets expensive quickly, and it recreates the original problem in a different form, because every additional relationship carries its own due diligence burden, its own service standards and its own points of failure.
Volume of relationships isn’t the same as reliability of access. What turns a collection of bilateral links into something dependable is coordination. Common standards, so information moves between institutions in a form each can actually use. Clear service responsibilities, so nobody’s client falls into a gap. Shared operating processes and a route for escalation when something goes wrong.
That’s the work a governed network of independent banks does, and it’s the part that can’t be bought as a product. The coordination matters precisely because the institutions being coordinated are separate, locally licensed and locally accountable. That independence is what gives each of them standing in their own market. It’s also what makes common standards necessary rather than automatic.
Network-level governance won’t guarantee correspondent acceptance or remove regulatory risk. What it can do is improve the consistency, timeliness and comparability of the information institutions use when assessing one another, which is precisely the weakness that drove much of the withdrawal in the first place.
Four Questions Worth Asking
For anyone responsible for international capability, this is a useful diagnostic.
- How many viable service routes exist in each of your priority markets, and what happens if you lose one?
- Which specific services disappear if a single partner withdraws? Payments, trade finance, local account access, or all three?
- Can KYC and service information move consistently between you and the institutions you rely on, or does each relationship work differently?
- Is it clear who owns the client experience when a transaction crosses two or three institutions?
Most banks can answer the first question. Fewer can answer the last one.
This applies whether or not your own access is under pressure.
Plenty of institutions reading this sit comfortably on the secure side of the network, with correspondent arrangements that have never looked fragile. That doesn’t take them out of the picture. If your client is expanding into a market where access has thinned, your capability there is only as strong as the institution you rely on locally, and you may not know how many alternatives that institution has behind it. Concentration works in both directions. The bank that feels least exposed is often the one furthest from knowing where the exposure actually sits.
Where That Leaves Independent Banks
The institutions still serving these markets properly are usually the ones based in them. A bank with real regulatory standing at home, genuine knowledge of local counterparties and an established corporate book doesn’t need convincing the market is worth serving. What it needs is dependable access to the rest of the world, on terms that don’t rest on a distant institution’s periodic review of the corridor.
Three of the four markets at the centre of that research are represented within the IBOS network, through Raiffeisen in Bosnia and Herzegovina, Privredna Banka Zagreb and Raiffeisen in Croatia, and K&H and Raiffeisen in Hungary. That doesn’t mean network membership removes correspondent risk, and we wouldn’t claim it does. It does illustrate what coordinated local coverage offers when bilateral access becomes less dependable, which is more routes, more relationships and more people who understand the market.
The No.1 Question Worth Asking
There’s little basis for assuming that technology alone will reverse the structural pressures behind correspondent retrenchment. What has held up across the same period is the presence of independent banks with real standing in their own markets, and the arrangements that connect them to each other. The economics that drove those decisions haven’t changed, and a faster rail doesn’t alter the underlying risk and return assessment behind them.
So the question for any bank supporting internationally active clients isn’t which rail their payments will travel on in five years. It’s simpler and less comfortable. When access to a market goes, what can you still do for the client who needed it, and who do you know well enough to ask?