The Basel framework requires banks to report their EVE sensitivity under prescribed shock scenarios, with an outlier threshold set at 15% of Tier 1 capital. The idea is to identify banks taking excessive gap risk. But the reported number is driven almost entirely by one modelling assumption: the behavioural repricing duration assigned to non-maturing deposits.
We tested this. Across six major UK retail banks, we estimated core sight deposit balances and asked: what happens if you shift the assumed duration by just one year?
Four of the six breached the threshold outright. The remaining two sat at or just within the limit. A one-year shift consumed 97% to 141% of each bank’s SOT EVE capital limit.
One year. A bank modelling its current accounts at 3 years versus 4 years is not making an unreasonable choice in either direction, yet that single year of difference eats the capital threshold. You can go from compliant to outlier by changing an assumption that nobody can verify. So what is the SOT actually telling us?
The Evidence
We sampled six major UK retail banks and building societies using Pillar 3 disclosures, annual results, and BoE aggregate deposit data. Core deposits were estimated on a sight-deposit-only basis, excluding time deposits and cash ISAs. UK system-wide, sight deposits represent 61.5% of total deposits (£1,927bn of £3,131bn). After prudential haircuts (10% NIB stability haircut, 10% initial plus 40% beta pass-through on IB*), core retention is approximately 38.8% of total deposits. EVE impact calculated under the Basel GBP 275bp parallel shock.
The table below shows what happens when you extend assumed deposit duration by one year. At four of the six banks, that single year blows through the SOT capital limit. The other two are left with headroom measured in tens of millions.
| Bank | Est. Core Deposits (£bn) | 1yr Duration Shift EVE Impact (£bn) | SOT EVE Limit (£bn) | % of SOT Limit Used | Headroom (£bn) |
|---|---|---|---|---|---|
| Bank A | 192.8 | 5.30 | 4.77 | 111% | -0.53 |
| Bank B | 172.0 | 4.73 | 4.74 | 100% | 0.01 |
| Bank C | 95.0 | 2.61 | 2.19 | 119% | -0.42 |
| Bank D | 112.1 | 3.08 | 2.73 | 113% | -0.35 |
| Bank E | 71.3 | 1.96 | 2.02 | 97% | 0.06 |
| Bank F | 111.8 | 3.08 | 2.17 | 141% | -0.90 |
Each year of NMD duration consumes 15-21% of Tier 1 capital in EVE sensitivity. The SOT EVE limit represents just 1.9% to 2.8% of core sight deposits. The deposit assumption alone takes up almost all of the threshold. There is barely any room left for actual gap risk between the deposit profile and the assets those deposits are invested in.
The entire metric is driven by the false precision of a single assumption that is, at its core, a guess.
The Pillar 3 disclosures tell the same story from the other end:
| Bank | Reported ΔEVE (% T1) | Duration Shift to Breach (Rates Up) | Duration Shift to Breach (Rates Down) |
|---|---|---|---|
| Bank A | -13.4% | ~1 month shorter | ~0.3yr longer |
| Bank B | -9.4% | ~4 months shorter | ~0.2yr longer |
| Bank C | -10.0% | ~3 months shorter | ~0.4yr longer |
| Bank D | -10.0% | ~4 months shorter | ~0.5yr longer |
| Bank F | -6.5% | ~5 months shorter | ~0.6yr longer |
Bank A is one month from breaching the SOT on rates up. One month. That is not headroom. But Bank A is not necessarily running more risk than its peers. A more complex NMD model captures a larger gap between deposit behaviour and the structural hedge, which produces a bigger reported sensitivity. A bank with a simpler model might show more headroom, but at the end of the day, the complex models are still guesses. They could be equally or more wrong than a management assertion.
This creates a perverse dynamic. In many regions, regulators are pushing banks towards increasingly complex modelling approaches for deposit behaviour. But the SOT thresholds will not support that complexity. The more complexity you bring into your deposit assumptions via a model, the more likely you are to breach. You either adopt the complex model and breach the SOT, or keep it simple and pick up a finding on model sophistication. Banks are caught between two regulatory expectations that pull in opposite directions.
Bank F shows the opposite problem. It reports the lowest EVE sensitivity in the peer group at -6.5% of Tier 1, sitting at 43% of the limit. Looks comfortable. But it has the highest deposit sensitivity per year of duration at 21% of Tier 1, because its capital base is small relative to its deposit franchise (£14.5bn T1 against £288bn total deposits). A shift of 5 months in deposit duration would breach. The reported number looks safe. The underlying sensitivity says otherwise.
Why the Assumption Can’t Be Fixed
This would matter less if we could actually measure deposit duration. We can’t.
The Basel Committee said as much when it moved IRRBB to Pillar 2 in 2016. Industry feedback was unanimous: standardising deposit behaviour would produce misleading results. The Committee agreed, concluding that “the heterogeneous nature of IRRBB would be more appropriately captured in Pillar 2.” Fair enough. But the EVE outlier test survived, with a threshold that depends on the very assumption the Committee said couldn’t be standardised.
The framework gives you duration ceilings (5 years for retail transactional, 4.5 years for non-transactional, 4 years for wholesale) but no methodology for how to get there. A bank could defensibly assign 6 months or 4.5 years to its retail savings book. Both are compliant. The EVE difference between them could range from 60% to 85% of Tier 1 capital based on the banks we sampled.
Banks spend significant time and money on historical decay studies, pass-through regressions, and cohort survival analyses to defend their assumptions. But all of that work tells you what happened. It does not tell you whether the franchise strength that produced that outcome will hold when conditions change. A building society with deep community roots is making a different claim about deposit stability than a digital challenger funded through comparison sites. Both might show similar historical retention. Only one has a forward-looking argument for why it should continue.
There is also a conceptual problem. EVE is a run-off measure: it assumes the bank stops writing new business and the book winds down. Banks are going concerns. In a run-off, deposits leave. The question of how long deposits stick around in a run-off is a different question to how long they stick around when the bank keeps operating. We are using a run-off metric with a going concern behavioural assumption, assessed against a threshold that one year of movement will breach.
The Earnings Tension
The false precision the SOT demands on deposit duration also conflicts with how banks actually manage earnings risk. Banks want to structurally hedge their deposit books at a duration that stabilises net interest income. That hedge tenor may be quite different from a run-off based modelled deposit duration. And as we have established, there is barely any room for gap risk between your EVE deposit assumption and the assets that are hedging those deposits.
This leaves banks with a choice: ignore the SOT limit and hedge to earnings anyway, push back against regional regulators who prefer a modelled run-off approach based on historical observations and match the assets to show no gap risk or a gap within your standard structural hedging range, or hedge your deposits to a run-off view at the expense of earnings volatility. There is no international standard for how to handle this. Banks across jurisdictions are making fundamentally different choices, with no consistency and no comparability.
A Simpler Framework
If the deposit duration assumption swamps everything else in the metric, and you cannot verify it, then stop pretending it is precise. Make the uncertainty visible.
Report EVE with NMDs treated as overnight. No behavioural modelling, no duration assumptions. Just the raw asset-side exposure. Then ask: how much deposit duration does this bank need to get within the limit?
For a stylised bank with £200bn deposits, £200bn assets at 2.5yr duration, £13.4bn Tier 1 capital, and core sight deposits of £78bn, the picture looks like this. When the deposit book is modelled at 2.5 years it matches the asset duration and ΔEVE is zero. As the assumed deposit duration shortens, the mismatch grows and EVE sensitivity climbs:
| Assumed NMD Duration | ΔEVE as % of Tier 1 | % of SOT Limit |
|---|---|---|
| 0.0 yr (overnight) | 39.8% | 265% |
| 0.5 yr | 31.9% | 212% |
| 1.0 yr | 23.9% | 159% |
| 1.5 yr | 15.9% | 106% |
| 1.6 yr (rates up threshold) | 15.0% | 100% |
| 2.5 yr (matched) | 0.0% | 0% |
| 3.0 yr (current) | -8.0% | -53% |
| 3.4 yr (rates down threshold) | -15.0% | -100% |
| 4.0 yr | -23.9% | -159% |
Each year of duration mismatch moves EVE by £2.13bn, or 16% of Tier 1, or 106% of the SOT limit. At overnight, EVE sensitivity is 40% of Tier 1. Over twice the threshold. Every bank would be an outlier on day one. That is the point.
An overnight EVE number is comparable across banks. No assumptions, no calibration debates. It shows you what the asset book actually looks like. The threshold is breached once NMD duration falls below 1.6 years, just 0.9 years from the matched position. A bank modelling deposits at 1.5 years instead of 2.5 years, both defensible choices, swings from flat to 106% of the SOT limit.
The bank is then making an explicit claim: “We need our deposits to behave like 1.6-year instruments to stay compliant, and we believe they will behave like 3.0-year instruments.” How confident are we in that? It depends on the franchise. That is the conversation worth having.
Try it with your own numbers:
Adjust the inputs to see how deposit duration assumptions drive the SOT outcome for any bank.
Stylised model. EVE Impact = Core Deposits × Duration Mismatch × Shock. SOT EVE limit = 15% of Tier 1 capital. Assumes assets and deposits are duration-matched at the asset duration, and the slider adjusts the deposit side only.
Report the overnight EVE. A comparable measure of asset-side gap risk with no assumptions baked in.
Disclose the required duration. How many years of deposit duration does the bank need to reach the threshold?
Assess franchise credibility. Is the required duration plausible given the bank’s business model, competitive position, and customer demographics?
We should be talking about deposit durations in defensible ranges, not pretending that the single point estimate we have selected for our EVE results miraculously keeps us within the threshold. The honest conversation is: “our deposits could plausibly behave like 1.5 to 3.5 year instruments, and here is the EVE range that produces.” Instead, the framework pushes banks to land on a number that works, defend it with backward-looking statistics, and hope nobody asks what happens if the assumption moves by a year.
As currently constructed, the EVE SOT is not comparable across banks and the reported sensitivities are muddy at best. The deposit duration assumption is so enormous relative to the threshold that the numbers lose meaning. Two banks reporting identical EVE sensitivities could have fundamentally different risk profiles, and two banks with identical risk could report completely different numbers. Until the assumption is unpacked, the metric tells us very little worth acting on.
Time to unpack it.
* These haircut assumptions are prudent relative to the findings of our UK deposit stability study.
Our EVE figures are based on simplified balance × shock calculations and ignore interest accrual and rate convexity that would be factored into a full economic value calculation. Analysis based on Pillar 3 disclosures from six major UK retail banks and building societies, Annual Results, BoE Stress Test Actuals, and Bank of England IADB aggregate deposit data (Table A5.2, January 2026). Core deposits estimated on a sight-deposit-only basis (61.5% of UK system total), excluding time deposits and cash ISAs. Tier 1 capital estimates used where exact figures are not publicly disclosed at ring-fenced entity level. Full methodology available in the EVE Analysis working paper.
Stephen Harvey is the founder of irrbb.com and Neuro-XI. He spent 15 years implementing and overseeing IRRBB measurement systems at two G-SIBs and one D-SIB.