The five-year cap on NMD behavioural repricing is one of the most discussed parts of the EBA’s IRRBB framework. Most practitioners can quote the number. Fewer can answer the questions that come up once you try to apply it. The cap is a 2022 addition retained over industry opposition during consultation, and the Phase 2 Heatmap Implementation Report (26 January 2026) has tightened the supervisory expectations around it without changing the underlying text. The analysis below is grounded in the EBA framework; PRA and other jurisdictional positions are addressed at the end.

What the cap actually is

What does the EBA five-year cap say?

Paragraph 111 of EBA/GL/2022/14 caps the assumed behavioural repricing date for three deposit types: retail, wholesale from non-financial customers, and operational from financial customers. The constraint is a maximum weighted-average repricing date of five years, applied separately for each currency. Two narrow exclusions: retail deposits with economic or fiscal withdrawal constraints, and non-operational deposits from financial customers, which are outside scope entirely.

The cap is on the weighted average, not on any individual cash flow. It is a ceiling, so shorter assumptions are unconstrained. Worth pausing on what a five-year weighted-average repricing date actually implies for total run-off: under a linear straight-line profile, a 5-year average corresponds to a 10-year run-off horizon. The cap is a ceiling on the average, not on when the last pound of behavioural balance walks off the book.

Does it apply to core deposits or both?

Both. Q&A 2023_6807 confirms that the cap covers the full amount of the aggregate portfolio, core and non-core combined, per currency. This is the most common point of misreading. The cap is a constraint on the aggregate, not on core maturity. Non-core balances sit at their slotted maturity (typically near zero) and pull the average down. The composition of the book matters as much as the modelling of the core.

The aggregation also operates at two layers simultaneously. It aggregates across the three in-scope deposit categories, and within each category across whatever segmentation the bank uses. So a retail book modelled across five current account cohorts can have individual cohorts at seven, eight or nine years, provided the volume-weighted average across all of them, combined with the wholesale and operational financial books, lands at or below five years per currency.

How it works in practice

Can I blend long and short segments to clear the cap?

On the literal text, yes. The cap is arithmetic on the weighted average across all in-scope balances. Anything slotted at or near overnight dilutes the average and creates headroom.

The discretionary lever is core / non-core classification. A book with 70 percent core at eight years and 30 percent non-core at zero averages to 5.6 years. Restate the same book at 60 / 40 and it lands at 4.8. The behavioural thesis on the core has not changed; the classification of stability has.

In practice the cap rarely binds. The non-core dilution effect means the aggregate breaches five years only if the core portion’s average maturity exceeds roughly 7.1 years (assuming 30 percent non-core), and the Phase 2 report confirms that institutions’ behavioural assumptions are broadly aligned with the benchmark. The cap functions less as a constraint that bites for most banks and more as a supervisory tripwire flagging outliers. It is most likely to bind on banks with deep retail franchises whose books are dominated by long-modelled non-interest-bearing balances.

Does the cap apply at group level, solo entity level, or both?

Paragraph 111 does not say explicitly. Our reading, in the absence of explicit guidance, is that the cap binds where the EVE measurement is being produced. This follows from the structure of paragraph 111 itself: the cap is specified as a weighted average per currency, with no instruction in the equation to take a further weighted average across entities. On that reading, each calculation at which an EVE measurement is produced (for most large banking groups, the consolidated calculation plus a handful of material solo entities driven by Article 7 CRR waivers, proportionality under paragraph 19 of GL/2022/14, and supervisor practice) is subject to the cap on its own portfolio.

The arithmetic creates asymmetric binding. A solo entity with a sticky NMD book at 5.8 years can sit inside a group whose consolidated picture lands at 4.7 because other entities pull the average down. Group passes. Solo fails. There is no offset across levels.

The consolidated calculation typically gets the benefit of more portfolio diversification than any individual solo entity does, which makes it less likely to breach the cap. But the benefit only flows one way. Group-level diversification does not cure a binding cap at solo level; the solo IMS still has to manage its own aggregate independently.

How is the cap enforced operationally?

Paragraph 111 specifies an outcome but not a mechanism. Banks are left with discretion over a methodology that materially affects ΔEVE.

The cleanest enforcement in practice is a downstream scaling adjustment, not a modification of the behavioural model itself. The bank runs its full behavioural model to produce unconstrained EVE sensitivities. A scaling step then reduces the balance receiving the term treatment so that the aggregate weighted-average repricing date lands at the cap, with the residual balance reclassified as overnight non-core. The cap-compliant ΔEVE is the unconstrained ΔEVE multiplied by the ratio of the cap to the unconstrained weighted-average repricing date, which is closed-form rather than a full revaluation.

The benefit is that the unconstrained behavioural view remains available for internal economic analysis and hedging design, while the constrained view feeds the regulatory metric. The same scaling logic applies cleanly at every consolidation level. Other methods exist (uniform compression, top-slicing, segment-level compression) and produce different ΔEVE outcomes; the choice should be documented as an assumption in its own right.

Does the cap apply to my SOT calculation?

Not directly, but operationally yes. The cap is not specified in the SOT RTS itself. Article 98(5a) CRD does not empower the EBA to impose behavioural assumption constraints in the standardised outlier test, and the EBA confirms in the RTS commentary that the SOT does not impose its own constraints in compliance with Article 98(5) CRD V. However, the same RTS commentary clarifies that for modelling and parametric assumptions not specified in the RTS, institutions shall use those they employ in their internal IRRBB measurement and management. Because IMS behavioural assumptions are themselves subject to the cap via paragraph 111 of GL/2022/14, the cap binds the SOT through the back door. Q&A 2023_6807 confirms the cap is a constraint on the IMS portfolio; what this means in practice is that the SOT inherits the cap rather than being exempt from it.

Does it affect NII or just EVE?

The cap is functionally an EVE constraint. EVE is a run-off measure where behavioural maturity directly determines where on the curve the cash flows sit. NII under the SOT and most internal frameworks is a constant balance sheet projection over twelve months; the long-tail behavioural maturity is irrelevant to that calculation, since the constant balance sheet logic rolls the unrepriced portion regardless of whether the modelled stability is five years or fifteen.

Wider context

How does the EBA cap compare to BCBS 368?

The two sit in opposite locations within their respective frameworks. BCBS 368 puts caps in the standardised framework only, with the bank’s IMS principles-based and no numerical ceiling. The EBA inverts this: the standardised approach exists under a separate RTS, but the headline five-year cap is an IMS-level requirement that applies to everyone.

The BCBS caps are also dual (a cap on the proportion of NMDs classified as core, and separately on the average maturity of that core), producing an implied aggregate ceiling that varies by category (around 4.5 years for retail transactional, lower for other categories). The EBA cap is unitary on the aggregate of core and non-core combined.

Do other regulators apply the same cap?

The PRA does. SS31/15 contains a substantively equivalent 5-year cap on retail and non-financial wholesale NMDs, per currency. Scope is marginally narrower than the EBA version (operational deposits from financial customers are not in scope), and the PRA has not yet built out the segment-level scrutiny architecture that Phase 2 introduces. The PRA has also not published an equivalent of Q&A 2023_6807, so the core / non-core aggregate point is not formally confirmed at the UK level.

The Federal Reserve has not adopted an explicit cap. US IRRBB supervisory guidance remains principles-based, reviewed through the standard examination process. The US has not implemented the BCBS IRRBB standard through formal regulation at all.

The EBA and PRA have both gone beyond Basel by placing the cap at the IMS layer. The Fed has stayed closer to the Basel position. There is no globally consistent answer for cross-jurisdictional banks.

What changed in the January 2026 Phase 2 report, and what should I do about it?

The text of paragraph 111 did not change. The Phase 2 report reaffirms the 5-year cap as the supervisory default, observes that most institutions’ behavioural assumptions are broadly aligned with it, and clarifies what supervisors expect from institutions that go beyond it: a credible link to business model, product or client segment characteristics, robust behavioural evidence, hedging coherence with the modelled assumption, and Pillar 3 disclosure with quantitative impact.

The report also notes ongoing dispersion in NMD behavioural modelling across institutions as a concern for cross-bank comparability, and signals that Pillar 3 disclosure quality will be a continuing area of supervisory focus.

For institutions sitting comfortably under the cap, this changes little in practice. For institutions modelling above it on any cohort, the supervisor’s expectations are now clearly set out. The cap remains the main regulatory benchmark for EVE measurement in Europe, but for institutions that exceed it, what matters more than the cap itself is the wider set of expectations around it: segmentation, hedging alignment and disclosure.

Appendix: cap-compliance scaling methodology

The main article notes that the cleanest enforcement of the cap is a downstream scaling adjustment rather than a modification of the behavioural model itself. This appendix sets out the methodology in full: the scaling formula in conditional form, the scope of the adjustment relative to total ΔEVE, and the validation argument for why no full revaluation is required. The audience here is practitioners implementing the cap in production and model risk teams reviewing the implementation.

Caveat on interpretation. Paragraph 111 specifies the cap as an outcome but is silent on the operational mechanism for enforcing it. The methodology below is one defensible approach, not the only one. Other methods (uniform compression across the run-off, top-slicing the long tail, segment-level rather than aggregate compression) are equally consistent with the text and will produce different ΔEVE outcomes. The choice of enforcement method is itself a modelling assumption that should be documented, validated and disclosed.

Pre-requisites for the scaling methodology

Before applying the scaling adjustment, the bank’s production IRRBB infrastructure must produce two inputs per currency.

First, the unconstrained NMD contribution to ΔEVE per currency. This is the ΔEVE attributable to in-scope NMD categories (retail, wholesale from non-financial customers, operational from financial customers) under the bank’s behavioural model, with no cap-compliance adjustment applied. Standard production engines produce this as part of the EVE decomposition by portfolio.

Second, the unconstrained aggregate weighted-average repricing date per currency. This is computed with non-core balances slotted at overnight (t=0) and core balances slotted at their modelled behavioural WARD. The aggregate WARD is volume-weighted across all in-scope deposit categories in that currency.

Both inputs are products of the bank’s existing behavioural model. No additional modelling is required; the scaling methodology is a post-processing step operating on outputs the engine already produces. The methodology does not depend on the specific shape of the core run-off (linear straight-line, exponential decay, or any other behavioural profile), provided the modelled WARD is recorded faithfully.

The scaling adjustment

For an NMD book with unconstrained aggregate weighted-average repricing date WARD_agg, the cap-compliant NMD ΔEVE is:

ΔEVE_NMD_compliant = ΔEVE_NMD_unconstrained × min(1, C / WARD_agg)

where C is the cap (5 years under paragraph 111 of EBA/GL/2022/14).

The min(1, …) enforces the asymmetric nature of the cap: it is a ceiling, not a target. For institutions with WARD_agg ≤ 5 years, the scaling factor is identically 1 and the adjustment is a no-op. For institutions with WARD_agg > 5 years, the factor falls below 1 and the scaling reduces the NMD contribution to ΔEVE in proportion to how much the unconstrained WARD exceeds the cap.

The adjustment is applied per currency, in line with the structure of paragraph 111. WARD_agg, the scaling factor, and the adjusted ΔEVE are all currency-specific.

Worked example. For a book with 70% core at modelled WARD of 8 years and 30% non-core, the unconstrained WARD_agg is 5.6 years. The scaling factor is 5 / 5.6 = 0.893. An unconstrained NMD ΔEVE of £100m becomes £89.3m after the cap-compliance adjustment, with the residual £10.7m representing the sensitivity attributable to the portion of core balance reclassified as overnight non-core for cap purposes.

Scope: NMD only, not total ΔEVE

The scaling factor applies to the NMD contribution to ΔEVE only, not to the total ΔEVE across the balance sheet. Term deposits, fixed-rate loans, wholesale funding, the investment portfolio, derivatives, and equity capital all sit outside the scope of paragraph 111 and are unaffected by the cap adjustment.

The cap-compliant total ΔEVE is therefore:

ΔEVE_total_compliant = ΔEVE_NMD_compliant + ΔEVE_non_NMD_unchanged

In practice, the bank’s EVE engine produces a decomposition of ΔEVE by portfolio. The NMD lines are identified per currency, the scaling factor is applied to each, and the adjusted lines replace the unconstrained lines in the aggregation. The non-NMD lines flow through unchanged.

This isolation makes the methodology operationally efficient. The same scaling formula runs unconditionally for every institution: for banks under the cap the min(1, …) returns 1 and the adjustment is a no-op, while for banks above the cap it scales the NMD ΔEVE down proportionally. No conditional logic is needed upstream, and no non-NMD position is recomputed.

Application across consolidation levels

The same logic applies independently at each consolidation level. Solo entity, sub-consolidated, and group EVE measurements each have their own unconstrained ΔEVE_NMD and WARD_agg, computed from their respective in-scope balances. The scaling factor is calculated and applied at each level independently. A banking group can therefore produce cap-compliant EVE measures at all required levels using the same methodology, with each level potentially having a different scaling factor (or none at all) depending on whether its own WARD_agg exceeds the cap. This is consistent with the asymmetric binding pattern described in the main article, where a group’s consolidated calculation can clear the cap while one or more solo entities sit above it.