Active correspondent banking relationships have declined more than 20% since 2011, according to BIS CPMI monitoring data. In some regions, particularly the Americas and small island developing states, the decline exceeds 30%. That figure is cited frequently as evidence that correspondent banking is failing. It is not. Banks are making a rational economic decision: they are retreating from low-margin clearing activity because their business model depends on balance sheet yield, not transaction throughput.
The decline in correspondent relationships is not a system breaking down. It is a system working exactly as designed, for institutions whose economics are built around lending. The friction in cross-border payments is not a technology problem or a messaging problem. It is an incentive problem. Understanding this distinction matters for every platform that depends on correspondent banking to move money across borders, because the solutions the market proposes only work if the diagnosis is correct.
The conflict between yield and velocity
A correspondent bank earns revenue from its balance sheet. It lends deposits, captures FX spreads, and retains funds as long as economically rational. These are legitimate, profitable activities. They are also fundamentally at odds with the objective of clearing: moving funds from one place to another as quickly and predictably as possible. You cannot optimise for yield and velocity at the same time.
An institution designed to lend has a structural incentive to hold deposits. An institution designed to clear has a structural incentive to release them. When the same institution does both, clearing becomes subordinated to balance sheet priorities. Settlement queues form not because the technology is slow, but because releasing funds competes with retaining them. The incentive to process quickly is real but secondary. The incentive to manage the balance sheet is primary.
The capital commitment is enormous. An estimated $5 trillion or more sits trapped globally in nostro and vostro prefunding, according to the Oliver Wyman and SWIFT joint report. This capital exists solely because settlement timing is unpredictable. Platforms prefund because they cannot trust that the correspondent will release funds on a known schedule. The prefunding is not a flaw. It is a rational response to uncertainty created by the correspondent's own business model.
Where the delay actually lives
The messaging layer is fast. SWIFT gpi data shows that approximately 90% of cross-border payments reach the destination bank within an hour. Yet the FSB's 2025 consolidated progress report found that only 35% of retail payments and 55% of wholesale payments are credited to end customers within that same hour. The gap between arrival and credit is not a transit problem. It is a processing problem at the edges of the correspondent chain.
The correspondent bank that receives funds decides when to release them. That decision is shaped by liquidity management, compliance review, and operational capacity. None of these are malicious. But they are all influenced by the fact that a bank holding deposits earns more than a bank releasing them. The incentive to process quickly exists, but it competes with every other demand on the institution's balance sheet and operational bandwidth.
This is why proposals to fix cross-border payments by fixing the messaging layer keep falling short. The FSB's own assessment acknowledges that tangible improvements for end-users have been limited despite completing the majority of international policy work under the G20 Roadmap. The messaging works. The infrastructure at each end of the message does not.
What the market proposes and why it falls short
The market has produced a range of solutions, each addressing a real problem. None of them change the economics of the institutions at the start and end of the chain:
- Stablecoins offer programmable settlement on-chain, then hit the same correspondent bank for off-ramp processing. The on-chain speed is real. The bottleneck at the edges remains.
- SWIFT alternatives promise fewer hops between institutions, but the institutions themselves still operate on the same balance sheet model. Fewer hops through the same infrastructure produces marginal improvement, not structural change.
- Treasury dashboards give perfect visibility into settlement queues they cannot shorten. The platform sees the delay in real time. It cannot resolve it.
- Orchestration layers optimise routing around one bottleneck and into another. The routing improves. The economics of the institutions being routed through do not.
Each has value in its specific use case, and none should be dismissed. But these solutions work around the existing infrastructure. They do not rebuild it. The constraint is not the network between the banks. It is the banks themselves. Until the institution at each end of the payment is designed for clearing rather than lending, the bottleneck remains.
The specialist correspondent model
The alternative is an institution designed for clearing from the ground up. No lending book. No balance sheet yield to protect. 100% reserve. The only revenue comes from moving funds, not from holding them. When the business model is aligned with clearing, settlement becomes predictable because there is no competing incentive to delay it.
This is where the etymology of the industry becomes relevant. Nostro means "ours" in Italian: the bank's own money held abroad. The entire correspondent banking system is organised around nostro: the bank's capital, the bank's liquidity, the bank's balance sheet priorities. The alternative is loro, "theirs": funds held on behalf of a third party, in segregated custody, moved through clearing infrastructure designed for movement, not storage.
When the institution's business model aligns with clearing, the downstream economics change fundamentally:
- Settlement becomes predictable. Prefunding requirements drop because the timing buffer disappears. Capital trapped in nostro accounts is freed for growth.
- Custody becomes structural. Compliance shifts from procedural reconciliation of pooled ledgers to verification of segregated named accounts. The audit trail is the architecture.
- FX becomes transparent. Wholesale interbank rates replace embedded markups. The platform sees the cost, benchmarks it, and manages it as a distinct line item.
The gap will widen before it closes
Instant payment schemes are expanding worldwide. The EU's Instant Payments Regulation mandates SEPA instant credit transfers across all eurozone PSPs. FedNow has raised its transaction limit to $10 million. Domestic instant settlement is becoming the norm, and it is resetting what platforms and their customers expect from every payment.
Cross-border settlement is not keeping pace. CEPR research has documented the firm-level consequences of correspondent bank withdrawal, finding that loss of correspondent relationships leads to measurable declines in export performance. The decline is not just an abstract statistic. It has real economic consequences for the businesses and platforms that depend on correspondent infrastructure to operate internationally.
As domestic instant payments set new expectations for speed and certainty, the gap between what platforms experience locally and what they experience internationally will grow. The institutions that close this gap will not be the networks or the messaging providers. They will be the clearers: specialist correspondent institutions that provide multi-currency clearing, named custody accounts, and liquidity management through a single relationship.
Lorum is built on this model. Regulated by the DFSA, operating across 30+ markets, with 100% reserve and no lending book, Lorum provides the clearing, custody, and cash management infrastructure that platforms need to settle with certainty rather than probability. The correspondent banking model is not broken. Its incentives are. The infrastructure should be designed accordingly.
