Project Mariana: Automated Market Makers for Cross-Border Wholesale CBDC Exchange
Project Mariana represents perhaps the most conceptually daring central banking experiment to emerge from the CBDC research programmes of the past five years. Conducted by the BIS Innovation Hub in partnership with the Swiss National Bank (SNB), the Banque de France (BdF), and the Monetary Authority of Singapore (MAS), the project explored whether automated market maker (AMM) protocols — mechanisms pioneered in the decentralised finance (DeFi) ecosystem — could facilitate the cross-border exchange of wholesale CBDCs. The experiment tested the application of algorithmic liquidity provision to the foreign exchange market for central bank money, producing findings that challenge conventional assumptions about the intersection of DeFi and institutional finance.
Conceptual Foundation
The foreign exchange market is the largest financial market in the world, with daily turnover exceeding USD 7.5 trillion. Cross-border FX transactions underpin international trade, investment, and capital flows. Yet the infrastructure supporting these transactions remains fragmented, opaque, and dependent on correspondent banking relationships that introduce delays, costs, and counterparty risks.
Project Mariana hypothesised that AMM protocols could provide an alternative mechanism for cross-border FX conversion between wholesale CBDCs. In the DeFi context, AMMs use mathematical formulas — typically variants of the constant product formula — to determine exchange rates between digital assets, with liquidity provided by pools of tokens rather than by individual market makers quoting bid-ask prices. The application of this model to wholesale CBDC exchange would enable direct, automated conversion between different central bank monies without the need for traditional FX intermediaries.
The hypothesis was deliberately provocative. AMMs in the DeFi ecosystem are associated with permissionless, pseudonymous trading of cryptocurrency tokens — an environment far removed from the regulated, institutional context of wholesale central bank money. Project Mariana’s contribution was to explore whether the mathematical and technical principles underlying AMMs could be adapted to a wholesale, regulated context while retaining the efficiency benefits.
Technical Architecture
The project implemented a three-currency AMM on a DLT platform, enabling the exchange of tokenised Swiss francs (CHF wCBDC), tokenised euros (EUR wCBDC), and tokenised Singapore dollars (SGD wCBDC). Each central bank issued its own wCBDC on a domestic DLT network, and a bridging mechanism enabled the transfer of wCBDC tokens to the shared AMM platform for cross-currency exchange.
The AMM employed a modified constant product formula adapted for the three-currency context. In a two-asset AMM, the constant product formula maintains the invariant x * y = k, where x and y represent the quantities of two assets in the pool and k is a constant. The three-currency extension generalised this formula to three dimensions, maintaining a hypersurface invariant that determined the exchange rates between any pair of the three currencies.
Liquidity was provided by the participating central banks, which contributed wCBDC to the AMM pool. The exchange rate between any two currencies was determined by the relative quantities of each currency in the pool: as one currency was purchased and its quantity in the pool decreased, its price relative to the others increased, following the constant product formula. This automatic price adjustment mechanism eliminated the need for external price feeds or market maker intervention.
The settlement mechanism was atomic: the exchange of one wCBDC for another on the AMM platform occurred as a single, indivisible transaction. The seller’s wCBDC was debited from their account and added to the pool, while the purchased wCBDC was withdrawn from the pool and credited to the buyer’s account, all in a single atomic operation. This atomicity eliminated settlement risk in the FX conversion process.
Findings and Analysis
The project yielded several findings that advance the understanding of how DeFi mechanisms might be applied to institutional finance.
AMMs can produce competitive exchange rates. Under the test conditions, the AMM-determined exchange rates closely tracked the reference rates from conventional FX markets. The constant product formula, when calibrated with appropriate pool sizes and fee parameters, produced spreads that were competitive with traditional FX execution costs for the transaction sizes tested. However, the project acknowledged that the relationship between pool size, transaction volume, and price impact requires further study under production-scale conditions.
Slippage is manageable but requires careful calibration. In AMM protocols, large transactions relative to the pool size cause significant price movement — a phenomenon known as slippage. For wholesale FX transactions, which can be very large, slippage could result in unfavourable exchange rates. The project found that slippage could be managed through appropriate pool sizing, transaction size limits, and the implementation of concentrated liquidity mechanisms that focus liquidity around the prevailing market rate.
The governance of AMM parameters is a central bank function. Unlike DeFi AMMs, where governance is typically exercised by token holders through decentralised voting, the wholesale CBDC AMM requires governance by the participating central banks. Decisions about pool sizes, fee structures, liquidity provision, access controls, and circuit breakers must be made by institutions with monetary policy mandates and accountability. The project demonstrated that this institutional governance layer could be superimposed on the AMM architecture without compromising its operational efficiency.
Cross-border regulatory coordination is essential. The operation of a multi-currency AMM involving three sovereign currencies requires regulatory coordination that goes beyond bilateral arrangements. The project identified the need for a multilateral governance framework that addresses the allocation of responsibilities, the resolution of disputes, and the management of systemic risks arising from the interconnection of three monetary systems through a shared platform.
Implications for FX Market Infrastructure
Project Mariana’s most significant implication is the demonstration that DeFi mechanisms can be adapted for institutional use cases without sacrificing the regulatory safeguards that distinguish wholesale financial markets from permissionless cryptocurrency trading. The AMM protocol, stripped of its association with speculative token trading and reimagined as a tool for central bank money exchange, offers efficiency gains that could transform cross-border FX settlement.
The traditional FX market relies on a network of dealer banks that provide liquidity by quoting bid-ask prices and absorbing inventory risk. This model, while functional, concentrates FX market-making in a small number of institutions, creates dependencies on their balance sheet capacity, and introduces credit risk through unsettled FX exposures. An AMM-based model distributes liquidity provision algorithmically, reduces dependency on individual institutions, and eliminates settlement risk through atomic exchange.
However, the transition from the traditional FX market structure to an AMM-based model would involve significant disruption to established business models and market practices. Dealer banks derive substantial revenue from FX market-making, and the displacement of this activity by algorithmic mechanisms would have implications for market structure, employment, and the distribution of value within the financial system. The pace and form of any transition would be influenced by the competitive dynamics between traditional and DLT-based FX infrastructure.
Relationship to Other Central Bank Initiatives
Project Mariana complements several related initiatives in the CBDC space. Project Jura explored the settlement of tokenised securities against wCBDC, focusing on the delivery-versus-payment mechanism rather than the FX conversion. Mariana addressed the FX conversion component that Jura’s architecture required but did not fully develop. Together, the two projects provide a more complete picture of how cross-border DLT-based settlement could function, encompassing both the securities and cash legs of international transactions.
Project Helvetia established the domestic foundations for wCBDC in Switzerland, demonstrating that the SNB could issue and manage tokenised Swiss francs for securities settlement. Mariana extended the use of SNB wCBDC to the cross-border FX context, adding a new dimension to the SNB’s experimentation with digital central bank money.
The mBridge project, involving the BIS Innovation Hub and central banks from China, Hong Kong, Thailand, and the UAE, has explored a similar multi-CBDC platform for cross-border payments but with a different architectural approach. Where Mariana employed AMM-based exchange, mBridge relies on a common platform with bilateral exchange arrangements between participating central banks. The contrasting approaches provide valuable comparative data on the relative merits of algorithmic versus bilateral FX mechanisms in the CBDC context.
Limitations and Open Questions
The project’s authors were candid about the limitations of the experiment and the open questions that remain. The test environment involved a small number of participants, a limited range of transaction sizes, and controlled market conditions. The behaviour of an AMM under stress conditions — high volatility, rapid capital flows, or a sudden withdrawal of liquidity by a participating central bank — was not tested and represents a critical area for future research.
The impermanent loss phenomenon, well-documented in DeFi AMMs, is also relevant to wholesale CBDC AMMs. Impermanent loss occurs when the relative prices of assets in an AMM pool diverge from external market prices, resulting in losses for liquidity providers compared to simply holding the assets outside the pool. For central banks providing liquidity to a wCBDC AMM, impermanent loss represents a cost that must be weighed against the benefits of the AMM model.
The interaction between an AMM-based wCBDC exchange and the broader FX market raises macroprudential questions. If a significant share of FX transactions were to migrate to AMM-based platforms, the impact on FX market liquidity, price discovery, and the effectiveness of central bank FX interventions would need to be assessed. These systemic considerations will ultimately determine whether AMM-based wCBDC exchange moves from experiment to production.
Donovan Vanderbilt is a contributing editor at ZUG DLT, covering distributed ledger technology law, regulation, and institutional adoption from Zurich. The Vanderbilt Portfolio AG provides research and analysis on Swiss digital asset infrastructure.