Book a Demo
Across the banking industry, assumptions about non-maturity deposits, or non-maturing deposits (NMDs), are usually discussed as methodology. Validators ask about the fit of a decay curve, and ALCO packs report a core share and a behavioral maturity as settled inputs.
A handful of published figures tell a different story. Each one measures a different way a deposit assumption can drift away from the deposits it describes, and together they show how much judgment sits inside a number most institutions report as a technical calibration.
This article walks through five of those figures: what each one measures, why it looks the way it does, and what it means for the people who approve deposit assumptions.
Why Does One Estimate Carry So Much Weight?
The European Banking Authority (EBA) found in its February 2025 report on its interest rate risk heatmap that, for half of the 120 institutions it reviewed, current accounts and instant-access savings made up more than 50% of all liabilities. These balances have no contractual maturity, so the bank has to estimate how much will stay, for how long, and how far their rate will follow the market.
That estimate travels, as I set out in Non-Maturity Deposits Explained. It sets the liquidity buffer, the economic value of equity (EVE) and net interest income (NII) reported under interest rate risk in the banking book (IRRBB), the size of the hedge book and the price credited to each deposit through funds transfer pricing (FTP). For example, a large European bank disclosed in 2018 that a model update shortening the modeled duration of about €185 billion of deposits by roughly half a year shifted an estimated €360 million a year in FTP, against about €40 million of additional hedging cost. Half a year of assumed depositor behavior moved hundreds of millions of euros between business lines every year.
From 0% to 90%: How Far Apart Are Core Deposit Estimates?
The same EBA report drew on end-2023 data. It found that the share of non-maturity deposits treated as core ran from 0% at the 5th percentile of institutions to 90% at the 95th, with a median of 52% and an average of 44%. For retail deposits alone, the median was 59%. Weighted average repricing assumptions across the same range ran from zero to just over four years.
Some spread is expected, since a regional retail lender and a bank serving technology companies do not have the same depositors. A range from nothing to nine tenths, among institutions under the same rules and often in the same markets, is much harder to explain by depositors alone.
The EBA also compared modeled assumptions with what banks were actually paying depositors in 2023, after the tightening had begun. Most institutions had raised deposit rates, yet behavioral models calibrated on long histories had not been changed, and the EBA concluded that "no clear relationship is observed between the NMD behavioural assumptions calibrated with historical data and the pass-through rates of the NMD in the current IR environment".
This finding is the one I would put in front of any ALCO. Read as a backtest, it is a result the sample as a whole did not pass.
How Often Do Banks Revise Their Non-Maturity Deposit Assumptions?
Dispersion could reflect honest disagreement; staleness is harder to defend. A European Central Bank (ECB) working paper followed 67 significant institutions, holding about 72% of euro area banking assets, from mid-2019 to the third quarter of 2023. About half never changed their NMD assumptions in that period. On average, banks revised their assumptions 1.88 times, and updates did not become more frequent after monetary tightening began in July 2022.
That period was not quiet. In its May 2023 Financial Stability Review, the ECB reported that between June 2022 and March 2023 rates on term deposits rose by about 2.44 percentage points while rates on overnight deposits rose by only about 0.25 points. Between February 2022 and March 2023, companies moved about 9 percentage points of their balances from overnight to term accounts. Much of that money stayed inside the bank, although it left the category that models treated as stable and cheap. The ECB working paper also found that banks with more volatile or rate-sensitive deposit bases did not shorten the maturities they assumed.
For management, the lesson is about process more than mathematics. An assumption that is refreshed only when a supervisor or auditor asks will not move when depositors do.
Why Do Deposit Betas Nearly Double Once Rates Pass 2%?
The third figure explains why so many assumptions went wrong in the same direction. A study in the Journal of Financial Intermediation covering US banks from 1997 to early 2023 found an average deposit beta of 0.49 when the fed funds rate was above 2%, against 0.26 below it. The share of a market rate move passed on to depositors nearly doubles once rates cross that threshold.
Under stable conditions, this relationship stays hidden. For most of the decade before 2022, policy rates sat near zero, deposit rates had nowhere to fall and little reason to rise, and a regression of one on the other found almost nothing. Measured betas came out small because there was almost nothing to pass through, rather than because depositors had stopped caring about rates.
As rates rose, the other half of the relationship appeared quickly. The Federal Reserve Bank of St. Louis measured the cumulative US deposit beta at 0.51 in the second quarter of 2024, up from minus 0.03 in the first quarter of 2022.
A model calibrated only on the low-rate years had measured half of the relationship and applied it to both. A lower assumed beta improves projected NII in a rising rate scenario, and it also lengthens the duration the bank assigns to its deposits, which supports longer hedges. When betas rise instead, the margin compresses and the deposits shorten at the same moment. The same Journal of Financial Intermediation study concluded that because betas rise with rates, deposits become a weaker hedge just as rates increase.
Deposits stopped working as a rate hedge exactly when rates rose fastest.
How Should Banks Model Non-Maturity Deposits After 2023?
Two Days Against Four: How Fast Do Deposits Leave Now?
The fourth set of figures concerns speed. The Basel Committee's report on the 2023 banking turmoil compared outflows at distressed banks with the 30-day horizon of the liquidity coverage ratio (LCR):
- A US regional bank concentrated in technology and venture deposits lost about 85% of its deposits in two days, including the outflows its management expected on the second day
- A New York bank with a concentrated commercial deposit base lost about 20% in one day
- The UK subsidiary of the first bank lost about 30% in one day
- A UK mortgage lender lost about 20% over four days in 2007
The comparison with 2007 carries the lesson. A run that took four days to remove a fifth of one bank's deposits in 2007 removed the same share from another in a single day in 2023, and a third bank lost most of its deposits in two.
European data does not support reading these runs as a US story about uninsured, concentrated deposit bases. In an October 2025 speech, a member of the ECB Supervisory Board presented evidence from 110 banks between 2016 and 2024: each percentage point of online banking penetration added about 0.28 percentage points to deposit outflows under extreme stress, with no systematic effect in normal times.
That last detail matters most for modeling. A risk that appears only in stress is invisible in calibration data drawn from calm years.
Do Fewer Outlier Breaches Mean Better Deposit Models?
The fifth figure looks like good news. In the EBA's yearly samples, breaches of the NII supervisory outlier test fell from 39 institutions, 28.06% of the sample, in 2022 to 16, or 16.49%, in 2023, and to 11, or 7.24%, in 2024. EVE outliers fell from 12 in 2022 to none in 2023 and one in 2024. The samples range from 97 to 152 institutions, so the comparison is indicative.
The same EBA work explains much of the improvement. Using end-2024 data, its January 2026 report estimated that without hedges about 36% of institutions would breach the EVE test, against under 1% with them, and about 20% would breach the NII test, against 7%. The hedges, in other words, do much of the work of keeping banks inside the thresholds, and almost all of it for EVE.
This is where the deposit assumption enters twice. Where hedges are sized to the modeled deposit profile, as a replicating portfolio is, the outlier regime rests on the assumption once in the measured exposure and again in the hedge that offsets it. A falling breach rate therefore says that banks have hedged the exposure their models describe. It says less about whether the models describe the deposits.
The caps offer little extra reassurance. The EBA found the median weighted average repricing date of non-maturity deposits was around 2.12 years at end-2023, well below the EU's five-year cap, and its January 2026 report describes that cap as a harmonising benchmark "with limited impact observed so far".
What Do the Numbers Say About How Models Are Built?
Each figure has a ready explanation on its own. Dispersion looks like a data problem, staleness like a process problem, the beta gap like a calibration problem. None of those explanations accounts for why the weaknesses appear together.
In my experience advising banks on ALM and structural risk, the common cause is the order in which the work is done. An expert forms a view of how the bank's deposits behave, and a model is then engineered to reproduce that view and satisfy auditors and supervisors. Assumptions then differ widely because they encode different priors more than different depositors, and they are rarely updated because a model that returns the expected answer gives nobody a reason to reopen it.
I also see most institutions set the interest rate risk and liquidity versions of the assumption separately. The same deposits count as long-term money in the EVE calculation and as a flight risk in the liquidity stress test, with no single document that reconciles the two.
What Do These Figures Mean for ALCO and the CFO?
Taken together, the five figures describe one assumption that can be wrong in four ways: it can differ from peers for no reason the bank can show, sit unchanged while depositors move, rest on a beta measured in the wrong regime, and ignore how fast money now leaves. Caps and outlier tests catch few of these, because they check whether an answer is permitted rather than whether it is right.
Supervisors have already changed the test. Across Basel, the EU, the UK and the US, the rules now converge on assumptions that are justified by the bank's own deposit history, backtested and stress-tested, with the burden of proof on the institution. The most useful question an ALCO can put to its deposit model is a simple one: would this model have predicted what our deposits did in 2023?
Institutions that treat the deposit model as a compliance calibration will keep passing reviews until the day their depositors behave differently from the assumptions. Institutions that let the data lead will see deposits start to move before the run-off shows in the balance.
For the full toolkit, from core balance estimation and deposit betas to replicating portfolios, and how Basel, EU, UK and US rules cap these assumptions, see Mirai RiskTech's whitepaper Non-Maturing Deposits Under Pressure: how banks model non-maturity deposits for IRRBB and liquidity.
