IRRBB Beyond the Headline EVE Number

By Jonas Osman Abdelghafour.

A bank's headline EVE sensitivity is a single number produced by an engine with a very large number of inputs, of which perhaps three genuinely matter. In a stable rate environment that concentration is tolerable. In a volatile one — where deposit behaviour, prepayment speeds and funding basis all move at once, and often in the same direction — it becomes the main source of measurement error in the ALM framework.

What the headline number conceals

Non-maturity deposit modelling dominates. Current and savings accounts have no contractual repricing date, so the bank assumes one. The split between core and non-core balances, the assumed repricing beta, and the amortisation profile of core balances typically move EVE more than the entire asset book's contractual structure. Two banks with identical balance sheets can report materially different EVE sensitivities purely through deposit assumptions.

The uncomfortable feature of these models is that they are calibrated on history. Deposit behaviour observed through a decade of near-zero rates says little about behaviour when rates are volatile, digital switching is frictionless, and money-market alternatives are visible in a customer's banking app. Betas estimated in the last cycle have repeatedly proved too low in the next one.

Prepayment and early-redemption optionality is conditional. Mortgage prepayment and term-deposit break behaviour respond to rates, and static assumptions inside a rate shock are internally inconsistent. If the scenario moves rates by 200 basis points and the behavioural model does not respond, the scenario is not being modelled.

Basis risk is invisible in a parallel shock. Discounting the whole balance sheet off one risk-free curve suppresses the spread between funding curves and asset yields — which is precisely where earnings pressure appears when markets are stressed. Parallel shocks are also the least informative shape: steepeners, flatteners and short-rate shocks frequently produce the binding result.

Pair EVE with a longer earnings view

EVE answers whether the balance sheet has the wrong duration. NII sensitivity answers whether next year's earnings survive a rate move. Both are needed, and the common weakness is a twelve-month NII window that is silent on everything repricing in year two and beyond. A three-to-five-year projection, run off the same cash flows and the same behavioural assumptions as EVE, is the cleanest bridge between the two views — and it removes the argument, familiar to every ALCO, about why the two metrics tell different stories.

Running both off one engine matters more than the precise horizon. Where EVE and NII use different behavioural assumptions, the numbers stop being comparable and the committee quietly stops trusting either.

Supervisory outlier tests are a floor, not the framework

The Basel and EBA framework prescribes six standardised shocks with an EVE outlier threshold against Tier 1 capital, and the EBA has added a supervisory outlier test on NII. These are useful comparability tools and a poor description of a specific bank's risk. Institutions should run internal scenarios calibrated to their own product mix, customer behaviour and funding profile — including scenarios in which behavioural assumptions themselves are stressed, not just the curve.

Making the framework usable

• Run sensitivity on the assumptions, not only on the curve. Report EVE under a range of deposit betas and core-balance profiles; the spread across that range is the honest measure of model uncertainty.
• Document the behavioural evidence. The most common validation finding is an assumption that cannot be defended when someone asks why this number and not another.
• Tie every limit to a pre-agreed action. Hedge thresholds, deposit-pricing triggers and reinvestment tilts should be agreed before a breach, not designed during one.
• Report the drivers. ALCO packs that show which assumptions moved the number are actionable; packs that show only the number invite false comfort.

Volatile rates do not break IRRBB frameworks. They reveal which ones were carrying their assumptions on trust.

Primary sources: BCBS standards on interest rate risk in the banking book; EBA guidelines and supervisory outlier tests on IRRBB; EBA Risk Assessment Report, June 2026.