Behavioural non-maturity deposit modelling for ILAAP and IRRBB

Behavioural non-maturity deposit modelling is a central input to both internal liquidity adequacy assessment and interest rate risk in the banking book. Current accounts, savings accounts and other deposits without a contractual maturity may appear operationally stable, but their stability is conditional on customer segment, pricing strategy, rate environment, confidence effects and competition. A defensible framework therefore has to estimate two related but distinct behaviours: the timing and volume of deposit run-off for liquidity purposes, and the repricing profile of balances for earnings and economic value sensitivity. This article sets out a technical approach to nmd behavioural modelling that can be used by treasury, ALM, risk, validation and audit teams when designing ILAAP, LCR, NSFR, contingency funding and IRRBB processes.

Why non-maturity deposits require behavioural modelling

Non-maturity deposits do not have a contractual final maturity that can be directly inserted into a liquidity gap report or an interest rate repricing ladder. A sight deposit is legally payable on demand, but a material portion may remain on the balance sheet for long periods. Conversely, deposits that have historically appeared stable may leave quickly when rates, market confidence or digital switching behaviour change.

This creates a modelling problem for at least four risk processes:

• **ILAAP**: internal liquidity adequacy assessment requires a forward-looking view of stressed cash outflows, survival horizon, funding concentration, liquidity buffers and management actions.
• **LCR**: regulatory liquidity metrics apply prescribed outflow assumptions, but internal monitoring normally requires more granular behavioural analysis to understand vulnerabilities within those prescribed categories.
• **NSFR**: stable funding measurement depends on the character of liabilities and their expected stability over a one-year horizon.
• **IRRBB**: economic value of equity and net interest income measures require assumptions about the maturity and repricing behaviour of non-maturity deposits.

The same product balance can therefore have different treatments across frameworks. In liquidity stress testing, the question is how much balance may leave and when. In IRRBB, the question is how much balance is stable enough to be treated as core and over what behavioural maturity, subject to prudential constraints and internal governance. Those differences should be explicit rather than hidden in a single undifferentiated assumption.

For a broader governance context, see the related Risk Governance Hub articles on liquidity risk management, asset and liability management and IRRBB governance.

Regulatory and supervisory context

The Basel Committee’s IRRBB standard recognises that non-maturity deposits require behavioural assumptions because their contractual maturity is not representative of their economic behaviour. The standard distinguishes between stable and non-stable balances and expects banks to use prudent constraints when determining behavioural maturity assumptions. The EBA Guidelines on IRRBB and CSRBB similarly require institutions to identify, measure, monitor and control interest rate risk in the banking book, including behavioural assumptions for products without contractual repricing dates.

For liquidity, the Basel LCR standard prescribes outflow rates for categories of retail and wholesale deposits. Those prescribed factors are not a substitute for internal analysis. They provide a regulatory minimum metric, while the ILAAP should assess whether the institution’s own liquidity risk profile, concentration, franchise characteristics and stress experience require more severe assumptions. The Basel NSFR standard similarly provides a regulatory structure for stable funding, but internal funding strategy should be supported by more granular behavioural evidence.

The ECB Guide to the ILAAP is also relevant for banks under the Single Supervisory Mechanism. It sets expectations for governance, risk identification, internal quantification, stress testing and integration into management decisions. Although it is not a modelling manual, it is directly relevant to the control environment around NMD assumptions: material assumptions should be documented, challenged, approved and used consistently in decision-making.

A practical implication is that institutions should not treat regulatory liquidity ratios and internal behavioural models as competing calculations. They answer different questions. Regulatory ratios promote comparability and minimum resilience. Internal behavioural models should help management understand the institution-specific sensitivity of its deposit base under business-as-usual and stressed conditions.

Methodology

A robust methodology starts with clear definitions. The model should specify the population of accounts, the treatment of dormant accounts, closed accounts, intra-group deposits, fiduciary balances, brokered deposits, sweep accounts, offset accounts, promotional balances and treasury-managed balances. Ambiguity at this stage often leads to unstable calibration and weak validation outcomes.

Segmentation

Segmentation should reflect behavioural drivers, not only product labels. Common dimensions include:

• customer type: retail, SME, corporate, financial institution, public sector;
• relationship depth: primary banking relationship, salary account, transactional account, single-product customer;
• balance size and concentration;
• rate sensitivity and product pricing tier;
• channel: branch, digital-only, intermediary or platform-originated;
• insured and uninsured balance components;
• operational versus non-operational wholesale deposits;
• currency and jurisdiction.

Segments should be large enough to estimate behaviour with reasonable statistical credibility, but granular enough to avoid masking high-risk sub-portfolios. Where data are limited, expert judgement may be necessary, but it should be governed under a formal framework such as governance of expert judgement.

Balance decomposition

Many institutions distinguish between volatile, non-core and core balances. One approach is to estimate a stable balance floor over a historical observation window, adjusted for stress and business growth. Another approach uses account-level survival analysis to estimate the probability that a balance remains over time. A third approach uses time-series models to link aggregate balances to explanatory variables such as administered rates, market rates, customer acquisition, seasonality and macroeconomic conditions.

The output should be fit for both liquidity and IRRBB use. For liquidity, the model may produce cumulative run-off curves by segment under baseline, idiosyncratic, market-wide and combined stresses. For IRRBB, it may produce an assumed repricing profile for stable balances, deposit beta estimates and maturity caps or floors consistent with policy.

Rate sensitivity and deposit beta

Deposit beta measures the extent to which administered deposit rates move with market rates. A simple beta can be estimated as the change in paid rate divided by the change in a reference rate, but that approach may be misleading where there are lags, floors, nonlinear pass-through, pricing campaigns or customer migration between products.

More robust specifications may include:

• lagged market rates;
• asymmetric pass-through in rising and falling rate environments;
• product-level administered rates;
• customer migration to term deposits or money market funds;
• balance elasticity to rate differentials;
• nonlinearity around zero or low-rate floors.

The model should separate pricing behaviour from balance behaviour. A low beta does not necessarily mean that balances are stable. Customers may tolerate low pass-through for a period and then withdraw when rate differentials become economically material.

Stress calibration

Stress assumptions should reflect both historical evidence and forward-looking vulnerabilities. Historical data may not include the current digital switching environment, changes in deposit insurance awareness, social media amplification, or recent shifts in monetary policy. Conversely, purely judgemental overlays may be difficult to validate if they are not anchored to evidence.

A balanced calibration typically uses:

• internal historical outflow events;
• peer or market stress observations where available from public supervisory or financial reporting sources;
• customer concentration analysis;
• early-warning indicators such as rate-shopping, failed retention offers and large balance movements;
• reverse stress tests to identify run-off levels that breach risk appetite or survival horizon constraints.

The calibration should be connected to stress testing programmes and to the firm’s risk appetite that actually guides decisions, rather than maintained as an isolated treasury assumption.

Worked example: from behavioural balances to liquidity and IRRBB assumptions

The following simplified example is illustrative only. It is not a regulatory calibration and should not be used as a benchmark.

Assume a bank has £10 billion of retail instant-access savings balances. The portfolio is segmented into three groups:

| Segment | Balance | Relationship feature | Observed rate sensitivity | Initial behavioural classification | |---|---:|---|---|---| | A | £4.0bn | primary relationship, salary or mortgage link | low | stable core candidate | | B | £3.5bn | savings-only relationship | medium | mixed | | C | £2.5bn | digital acquisition, promotional pricing history | high | non-core candidate |

For ILAAP liquidity stress testing, the bank estimates cumulative 30-day outflows under a combined idiosyncratic and market stress as follows:

| Segment | 30-day outflow assumption | Outflow amount | |---|---:|---:| | A | 8% | £0.32bn | | B | 18% | £0.63bn | | C | 35% | £0.875bn | | **Total** | **18.25% weighted average** | **£1.825bn** |

The bank then assesses whether its liquidity buffer, monetisation assumptions and contingency funding actions remain sufficient after this outflow, allowing for other contractual and behavioural cash flows. This assumption may be more severe than the regulatory LCR treatment for some balances if internal evidence indicates greater sensitivity.

For IRRBB, the same £10 billion portfolio is not automatically assigned the same behavioural profile. The bank estimates that only £5.5 billion qualifies as stable core balance for interest rate risk purposes after excluding high-beta and promotional balances. It assigns the following illustrative repricing profile:

| IRRBB bucket | Stable balance allocated | Rationale | |---|---:|---| | Overnight to 3 months | £1.0bn | operationally liquid component | | 3 months to 1 year | £1.2bn | moderately stable balances | | 1 to 3 years | £1.8bn | observed relationship stability | | 3 to 5 years | £1.5bn | longest behavioural component, subject to policy cap | | **Total stable core** | **£5.5bn** | |

The remaining £4.5 billion is treated as short repricing or non-core for IRRBB purposes. The example illustrates three points. First, liquidity stability and interest rate stability are related but not identical. Second, segmentation materially affects both survival horizon and economic value sensitivity. Third, policy caps and validation challenge are necessary to prevent excessive earnings stabilisation through optimistic core deposit assumptions.

Model validation and performance monitoring

Independent validation should assess conceptual soundness, data quality, implementation, outcomes analysis and governance. This is consistent with broader principles discussed in independent model validation standards and model risk management frameworks.

Validation checklist

A practical validation review should consider the following questions:

1. **Scope and definitions**: Are product populations, exclusions and account treatments clearly defined and reproducible? 2. **Segmentation**: Are segments behaviourally meaningful, statistically credible and stable over time? 3. **Data lineage**: Can balances, rates, customer attributes and account events be reconciled to source systems and finance records? 4. **Historical window**: Does the calibration period include relevant rate cycles, stress periods and business model changes? 5. **Survivorship bias**: Are closed accounts and migrated balances captured, rather than only accounts that remain open? 6. **Outlier treatment**: Are large one-off flows, corporate actions and operational transfers treated consistently? 7. **Rate beta estimation**: Are lags, nonlinearities and pricing floors considered where material? 8. **Stress overlays**: Are overlays justified, approved and subject to sensitivity testing? 9. **Back-testing**: Are realised balances and outflows compared with model projections at segment level? 10. **Use test**: Are assumptions actually used in ALCO reporting, ILAAP, funds transfer pricing, risk appetite and contingency planning? 11. **Limit monitoring**: Are trigger breaches escalated to treasury, risk and senior management? 12. **Change control**: Are recalibrations, methodology changes and management overlays documented and independently reviewed?

Performance monitoring should not wait for annual validation. Monthly or quarterly monitoring can compare actual balances with expected behavioural ranges, track rate pass-through, identify migration into term products, and test whether early-warning indicators remain predictive. Where model performance deteriorates, compensating controls may include temporary assumption conservatism, enhanced ALCO review or restrictions on balance sheet strategies dependent on the model.

Integration with ILAAP, LCR, NSFR and contingency funding

NMD behavioural models should be embedded into treasury risk management rather than maintained as technical artefacts. The ILAAP should explain how assumptions are selected, how severe but plausible stresses are defined, how deposit concentrations are assessed, and how liquidity resources are sized against the resulting cash-flow profile. The analysis should support board and senior management decisions, consistent with sound board risk reporting.

For LCR, behavioural modelling can be used to analyse the drivers behind regulatory outflow categories. For example, a bank may identify that deposits within the same regulatory bucket have materially different sensitivity by channel or balance size. The regulatory calculation remains prescribed, but management information can highlight concentrations that require pricing action, customer retention planning or additional liquidity buffers.

For NSFR, behavioural evidence can inform funding strategy and the internal assessment of stable funding reliance. A portfolio that receives favourable regulatory treatment may still create strategic funding risk if it is concentrated in a narrow customer segment or acquired through rate-led campaigns.

For contingency funding, behavioural model outputs should be translated into operational playbooks. This includes defining trigger levels, expected management actions, communication protocols, collateral mobilisation, central bank facility readiness where applicable, secured funding capacity and customer retention measures. Assumptions about the timing and effectiveness of actions should be tested, not merely listed. A contingency funding plan that assumes immediate execution of untested actions will not provide reliable support to the ILAAP.

Funds transfer pricing is another important integration point. If FTP credits business lines for stable deposits, the value assigned should be consistent with validated behavioural maturity and liquidity value. Over-crediting deposit franchises can create incentives to gather balances that are economically unstable or expensive to retain.

Governance and controls

Effective governance requires clear ownership. Treasury or ALM may own the model methodology, but risk should provide independent challenge, model validation should assess model risk, finance should reconcile source data, and ALCO should approve material assumptions. Internal audit may assess the control framework, including whether the model is used consistently in regulatory and management processes.

Key governance artefacts include:

• model documentation covering purpose, scope, methodology, assumptions, limitations and implementation;
• a data dictionary and lineage map;
• calibration evidence and sensitivity analysis;
• validation reports and remediation plans;
• ALCO approval records;
• policy limits for maximum behavioural maturity and core balance recognition;
• monitoring dashboards and escalation thresholds;
• a clear distinction between regulatory assumptions, internal base assumptions and stress assumptions.

Governance should also address incentives. Business units may prefer assumptions that recognise long behavioural maturity and low rate sensitivity because these can increase perceived deposit value. Risk and finance functions should challenge whether such assumptions are supported by observed behaviour and by plausible stress outcomes. The three lines model is relevant here: business ownership, independent risk oversight and audit assurance should be clear, as discussed in three lines in practice.

Limitations

NMD behavioural modelling is inherently uncertain. Past stability may not persist after a change in rate environment, customer communication, digital switching costs, market confidence or competitor behaviour. Models calibrated on long periods of low or stable rates may underestimate balance migration when customers have stronger incentives to search for yield.

Data limitations are common. Account-level histories may be incomplete after system migrations. Product codes may change. Customer relationship indicators may not be available historically. Closed accounts may be missing from analytical datasets. These issues can bias stability estimates upward if not corrected.

There is also a risk of false precision. A model may produce detailed monthly decay curves or long-duration repricing profiles, but the confidence interval around those estimates may be wide. Management reporting should therefore include sensitivities and scenario ranges, not only point estimates.

Finally, liquidity and IRRBB uses can create tension. Longer behavioural maturities may reduce measured earnings volatility in some scenarios but can increase economic value sensitivity. Shorter maturities may be conservative for IRRBB economic value but may not capture franchise value or funding behaviour. The institution should document these trade-offs and avoid forcing one assumption set to serve all purposes without adjustment.

Frequently asked questions

Question 1: Is nmd behavioural modelling required if LCR outflow rates are prescribed?

Yes, for internal risk management purposes. The LCR applies prescribed regulatory assumptions, but the ILAAP should assess the bank’s own liquidity risk profile. Behavioural modelling helps identify deposit segments that may be more sensitive than their regulatory category suggests and supports contingency funding planning.

Question 2: Should the same NMD assumptions be used for ILAAP and IRRBB?

Not without analysis. ILAAP assumptions focus on cash outflow timing under stress, while IRRBB assumptions focus on repricing behaviour and economic maturity. The same data should be reconciled, but the model outputs may differ because the risk questions differ.

Question 3: How often should NMD models be recalibrated?

The frequency should reflect materiality, portfolio change and model performance. Annual recalibration may be appropriate for some stable portfolios, but rapid rate changes, business model shifts, deposit campaigns or monitoring breaches should trigger interim review.

Question 4: What is the main validation concern for core deposit estimates?

A common concern is overstating stable balances by relying on aggregate historical averages. Validation should test whether stability persists at account or segment level, whether closed accounts are included, whether stress periods are represented and whether rate sensitivity has changed.

Professional disclaimer: This article provides general technical information and should not be treated as regulatory, legal, accounting or investment advice.