Non-Maturing Deposits: How Banks Model NMDs for IRRBB and Liquidity

From first principles to machine learning in deposit modeling for balance sheet, interest rate and liquidity risk management.

What Are Non-Maturing Deposits?

Non-maturing deposits (NMDs), also called non-maturity deposits, are current accounts, savings accounts and instant-access deposits that customers can withdraw on any day. They have no contractual maturity, yet most of the balance stays for years, and the rate paid on it follows market rates only partly and with a lag. That makes NMDs the cheapest and often the largest source of funding on a bank's balance sheet, and the hardest to pin down: nothing in the contract says how long the money will stay.

Why Do NMD Assumptions Matter So Much?

The answer to "how long will the money stay?" comes from a behavioral model, and it travels further than most institutions acknowledge. The same estimate of how much of the balance is core, and for how long, sets the liquidity buffer, the EVE and NII reported under interest rate risk in the banking book (IRRBB), the hedges treasury puts on and the funds transfer price each deposit receives. NMD assumptions are the largest single judgment on a bank's balance sheet, although many institutions still treat them as a technical calibration.

What Did 2022 and 2023 Change for Deposit Models?

A decade of near-zero rates made deposits look stable and nearly free. The rate cycle that began in 2022 broke that picture: deposit betas rose with rates, balances migrated to better-paid products, and in March 2023 concentrated, uninsured and digitally connected deposit bases left within days. The Basel Committee records one US regional bank losing about 85% of its deposits over two days.

The lesson does not stop at the US border. The EBA found that the share of retail NMDs treated as core ranged from 0% to 90% across 120 European institutions, with no clear relationship between modeled assumptions and the deposit pricing banks actually applied. An ECB study of 67 significant banks found that only about half changed their NMD assumptions between 2019 and 2023.

How Should Banks Model Core Deposits and Deposit Betas?

Too often, models are built the wrong way around: an expert forms a view of how the bank's deposits behave, and a model is engineered to reproduce it. This whitepaper argues for reversing the order and letting the data lead, using the established toolkit and, where it falls short, more advanced methods.

What Will You Learn in This Whitepaper?

• The NMD toolkit: segmentation, core balance estimation, decay and survival analysis, deposit beta models and replicating portfolios
• Advanced models: betas that move with the rate level, deposits valued as options, stochastic balances and one behavioral engine for IRRBB, liquidity and planning
• Governance: backtesting assumptions against outcomes, independent challenge and one set of assumptions across reports
• Regulation: how Basel, the EU, the UK and the US cap and test deposit assumptions, and where the four rulebooks converge
• AI and machine learning: why a more precise model is still the wrong core model today, and how AI and agents earn their place around a transparent one

Who Is This Whitepaper For?

Treasury, ALM, IRRBB, liquidity risk and model validation teams, and the ALCO members who rely on their numbers.

About the Author

Luis Estrada is Co-Founder of Mirai, where he leads the consultancy arm serving banks and financial institutions. He has more than two decades of experience in finance, risk management and technology, and has taught AI applied to finance at IEB in Madrid.