Book a Demo
On 9 March 2023, Silicon Valley Bank lost more than $40 billion of deposits in a single day, and its management expected to lose over $100 billion more the next morning. Signature Bank lost about a fifth of its deposits in one day. In 2007 it took Northern Rock four days to lose the same share. In Europe, a large share of corporate money moved from overnight to term accounts without leaving the bank at all. Each of those movements broke a deposit assumption that had looked safe for a decade.
Those assumptions matter more than their quiet decade suggested. The estimate of how much of a non-maturity deposit (NMD) balance is stable, and for how long, 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 hedges treasury puts on and the price each deposit receives through funds transfer pricing (FTP). When that estimate is wrong, every one of those numbers is wrong with it. This article looks at what the 2022 rate cycle and the runs of March 2023 revealed about deposit behavior in the US and Europe, and what each lesson means for the NMD assumptions banks hold today.
What did the low-rate decade teach deposit models?
The first lesson concerned price. With market rates close to zero, deposit rates had nowhere to fall and little reason to rise. Measured deposit betas, the share of a market rate move passed on to depositors, came out small because there was almost nothing to pass through. Depositors had not stopped caring about rates. A model calibrated on those years concluded, reasonably from its own data, that the bank could hold deposit rates still whatever the market did.
The second lesson concerned volume. From 2020, surplus pandemic savings swelled accounts further. Decay analysis run on a growing book shows money that stays and keeps arriving, which supports high core shares and long behavioral maturities. Both conclusions sat comfortably inside the regulatory caps, and both pointed where the business preferred: long-dated funding at almost no cost.
An assumption that never seemed to move drifted from the center of asset-liability management (ALM) to the edge of the compliance calendar, refreshed when a supervisor or auditor asked and judged mainly by whether it passed. Nothing visibly failed, because the calibration data and the world the model described were the same world. The weakness was that the model had only ever seen one regime.
How fast did deposit rates follow the market after 2022?
Pricing moved first. Research published by the Federal Reserve Bank of New York found that the cumulative beta on US interest-bearing deposits reached almost 0.4 by the fourth quarter of 2022, matching the peak of the 2015 to 2019 cycle in about one year instead of three. By the second quarter of 2024, the Federal Reserve Bank of St. Louis measured the cumulative US deposit beta at 0.51, up from minus 0.03 in the first quarter of 2022.
This surprised many models, although it would not have surprised the research. A study in the Journal of Financial Intermediation covering 1997 to 2023 found an average US deposit beta of 0.49 when the fed funds rate was above 2%, against 0.26 below it. Beta depends on the level of rates. A model calibrated only on years below that threshold had measured the low-rate half of the relationship and applied it to both.
Because betas rise as rates climb, the effective duration of deposits shortens exactly when rates rise. A bank that funded long fixed-rate assets on the assumption of sticky deposits saw its net interest margin compress as deposit costs rose, while the value it had counted on from those deposits as a natural hedge shrank at the same time. That is the mechanism behind SVB's interest rate risk losses, and it was present, in milder form, across the industry.
Where did European deposits go before anything ran?
Volumes moved next, most visibly in the euro area. The European Central Bank (ECB) reported that between June 2022 and March 2023 rates on overnight deposits rose by only about 0.25 percentage points, while term deposit rates rose by about 2.44 percentage points. Companies responded: the term share of non-financial corporations' deposits rose by about 9 percentage points between February 2022 and March 2023.
The European Banking Authority (EBA) later found that, across its sample of EU institutions, the share of total deposits held as NMDs fell by about 6% during 2023, while total deposits stayed broadly stable. Much of this money went into term deposits rather than out of the bank. It still left the category that models treated as stable and cheap, and for the NMD model that amounts to much the same thing as an outflow.
A bank watching only its total deposits could conclude that its NMD assumptions had survived the cycle, while the composition underneath told a different story.
How fast did deposits run in March 2023?
Then some deposit bases stopped migrating and ran. The Basel Committee on Banking Supervision's analysis of the 2023 turmoil, first published in October 2023 and updated in its October 2024 progress report, set the outflows against earlier runs and against the 30-day horizon of the liquidity coverage ratio (LCR):
- Silicon Valley Bank: about 85% of deposits in two days, counting the outflows management expected on 10 March
- Silicon Valley Bank's UK subsidiary: about 30% in one day
- Signature Bank: about 20% in one day
- Credit Suisse: about 25% in seven days
- Northern Rock, 2007: about 20% in four days
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. Silicon Valley Bank was the extreme case: according to the Federal Reserve's April 2023 review, about 94% of its parent's deposits were uninsured at the end of 2022. The Basel Committee pointed to faster payments, mobile banking and social media as common drivers of the outflows at the failed US banks and at Credit Suisse, and found that the outflows exceeded the levels assumed in the LCR and the net stable funding ratio by a wide margin.
How Should Banks Model Non-Maturity Deposits After 2023?
Was March 2023 only an American story?
It is tempting to read March 2023 as a US episode about concentrated, uninsured deposits, with little to teach banks built on insured retail balances. The European evidence does not support that reading. Credit Suisse's collapse was a European run, and the digital channel that accelerated the US runs shows up in European data too.
In an October 2025 speech, ECB Supervisory Board member Pedro Machado presented evidence from 110 banks directly supervised by the ECB between 2016 and 2024. Under stress, extreme outflows were about 0.28 percentage points higher for each percentage point of online banking penetration, with no systematic effect in normal times. That last finding matters most for modeling. A risk that appears only in stress is invisible in calibration data drawn from calm years.
Supervisors have drawn the conclusion on both sides of the Atlantic. Michael Barr, then the Federal Reserve's Vice Chair for Supervision, said in February 2024 that uninsured deposits "proved flightier than previously assumed". The EBA, comparing modeled assumptions with how banks actually repriced deposits in 2023, found "no clear relationship" between the two, as I discussed in what the data says about NMD assumptions. The test applied to an NMD assumption has moved from whether it fits within the cap to whether it is right.
Which competitors did the calibration data never see?
The pressure has not eased. US money market funds reached a record $7.98 trillion in the week to 2 September 2026 and still held $7.92 trillion in mid-October, according to the Investment Company Institute. The Bank for International Settlements (BIS) put the stablecoin market at about $320 billion at the end of May 2026. When retail money moves into a stablecoin, it comes back to the banking system, if it comes back at all, as the issuer's wholesale deposit, which carries higher run-off rates than the retail balance it replaced.
In the low-rate years these alternatives paid little more than a current account. The data on which most NMD models were calibrated therefore never saw them compete for the same money.
What does each lesson mean for non-maturity deposit assumptions today?
The pricing lesson is the most direct. A beta estimated on the low-rate decade describes one regime and will understate deposit costs whenever rates sit above it. The assumption needs to let pass-through rise with the rate level, tested against the 2022 to 2024 experience rather than the years before it.
The migration lesson concerns what counts as an outflow. Money that moves from a current account to a term deposit at the same bank no longer behaves like an NMD, so decay analysis that tracks only total deposits will miss it. Segmentation by product, by insured and uninsured balances, by concentration and by channel is where that movement becomes visible.
The speed lesson belongs to liquidity. The Basel Committee's review questioned whether the outflow rates assumed in the LCR, and the 30-day window itself, fit runs of this speed, and the ECB speech warned that run-off rates may underestimate the risk of uninsured, digitally active deposits. In May 2026 the Basel Committee agreed to consider targeted updates to its 2008 liquidity risk principles. Whatever comes of that, a regulatory floor calibrated for a slower world cannot stand in for the bank's own estimate.
The last lesson is the one I see most often in practice. In my experience advising banks on ALM and structural risk, most institutions still run static, rarely recalibrated NMD assumptions, and they set the interest rate risk and liquidity versions separately. The same deposits are treated as long-term money in one report and as a flight risk in another. The events of 2023 hit both reports at once, which is why the two views need to start from the same evidence.
Why must assumptions be tested against regimes they have not seen?
Deposits did not turn unstable across the board in 2023. Much of the European money that moved went into term deposits at the same bank rather than out of it, and the fastest runs hit concentrated, uninsured and digitally connected deposit bases. What failed was a set of assumptions calibrated on a single regime and never asked how they would behave in another, with higher rates, faster payments and competitors paying a market rate for the same money.
This changes the question for the asset and liability committee (ALCO). Fitting within the caps and passing validation show only that an assumption is allowed. After 2023, supervisors and boards want evidence that it came from the deposits themselves and has been backtested against outcomes: survival analysis on segmented balances, beta models that respond to the rate level, and one set of assumptions shared across IRRBB, liquidity and planning.
Institutions that treat the deposit model as a compliance calibration will keep passing reviews until their depositors behave differently from the assumptions. Institutions that test those assumptions against regimes they have not yet seen 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.
