The EBA’s RTS on materiality assessment defines when IRB model modifications require full supervisory re-approval versus notification. Getting this wrong means either blocked model updates or supervisory credibility damage.
The RTS establishes three tiers of model changes, each with distinct supervisory treatment. The classification depends on quantitative thresholds and qualitative criteria — misclassification creates regulatory risk in both directions.
| Classification | RWA Threshold | Parameter Shift | Supervisory Treatment | Typical Timeline | Severity |
|---|---|---|---|---|---|
| Material Extension | >5% RWA impact on portfolio | New model / new portfolio application | Full prior approval required | 9–12 months | CRITICAL |
| Material Change | >5% RWA impact | >10% shift in key parameters (PD, LGD, CCF) | Streamlined approval required | 6–9 months | HIGH |
| Non-Material Change | ≤5% RWA impact | ≤10% parameter shift | Notification only (ex-post) | 2–4 weeks | LOW |
The 5% RWA threshold is assessed at the affected portfolio level, not at the total bank level. A model change affecting a niche €2bn corporate sub-portfolio can easily breach 5% even if the bank-wide RWA impact is negligible. This catches many banks off guard.
A structured decision flow for classifying any proposed model modification. Each node should be documented in your model change register to create an auditable classification trail.
Even if quantitative thresholds are not breached, a fundamental methodology change (e.g., switching from expert-based PD to statistical regression) is always classified as material. The ECB applies qualitative judgment, and banks that rely purely on quantitative tests face supervisory challenge.
A typical G-SIB operates 50–120 IRB models across four risk parameter types and multiple borrower segments. Each cell represents models that must be individually tracked for materiality classification.
| Model Type | Retail | Corporate | SME | Specialised Lending | Institutions |
|---|---|---|---|---|---|
| PD Models | 6–10 models | 3–5 models | 2–4 models | 2–3 models | 1–2 models |
| LGD Models | 4–8 models | 3–5 models | 2–3 models | 2–3 models | 1–2 models |
| CCF Models | 3–5 models | 2–4 models | 1–2 models | 1 model | 1 model |
| EAD Models | 2–4 models | 2–3 models | 1–2 models | 1–2 models | 1 model |
With CRR3 driving concurrent model updates across PD, LGD, CCF, and EAD, a G-SIB may need 15–30 material model change applications submitted simultaneously. The ECB’s review capacity is finite — late submissions will face queue delays that can stretch approval timelines to 15+ months.
Understanding the ECB’s internal review process is essential for planning model change submissions. Each phase has distinct deliverables and potential failure points that create timeline risk.
Bank submits formal application with model documentation, impact assessment, validation report, and governance sign-off. Completeness check within 2 weeks — incomplete applications are returned.
ECB JST assesses scope, methodology, and data quality at high level. Preliminary questions issued within 3 months. Banks must respond within 4–6 weeks or face timeline reset.
Deep-dive technical review of model specification, calibration, validation, and performance. On-site model review possible. Duration: 3–6 months. Multiple rounds of questions typical.
Formal ECB decision: approved, approved with conditions, or rejected. Conditions may include additional conservatism margins, monitoring requirements, or implementation deadlines. Timeline: 2–3 months.
The 12-month timeline assumes clean submissions and prompt responses. In practice, 40% of applications experience at least one significant delay — incomplete documentation, data quality issues, or methodology questions that require multi-week internal analysis before responding to the ECB.
Based on observed ECB feedback across multiple model change applications, these are the most frequent causes of rejection, delay, or conditional approval. Address these proactively before submission.
The single most common cause of application return. Missing development documentation, outdated technical specifications, or incomplete model performance reports. The ECB expects a self-contained package — reviewers should not need to request basic context information.
Backtesting that covers only benign economic periods, uses inappropriate benchmarks, or lacks statistical rigor. The ECB expects backtesting across multiple economic cycles, with formal breach analysis and documented remediation actions for any threshold violations.
Missing exposure-level data, unexplained data exclusions, or inconsistent data definitions across model development and validation datasets. The ECB has increasingly tight expectations on data lineage, completeness metrics, and treatment of missing values.
Validation reports that merely describe model outputs without providing genuine independent challenge. The ECB expects validators to test alternative specifications, challenge key assumptions, and document disagreements with model developers.
Impact assessments that cover only the base case without sensitivity analysis across alternative calibrations, stress scenarios, and portfolio composition changes. The ECB wants to understand the range of outcomes, not just the central estimate.
Missing or perfunctory board/committee sign-off on model changes. The ECB expects documented evidence that senior management understands the model change, its rationale, and its capital impact. A rubber-stamp approval without substantive discussion is visible to reviewers.
Prioritised actions for banks managing large IRB model inventories through the CRR3 transition period. Focus resources on the highest-impact items first.
Create a centralised view of all pending, in-flight, and planned model changes with pre-assessed materiality classifications, estimated timelines, and ECB submission readiness scores. This visibility prevents approval bottlenecks and enables resource prioritisation.
Deploy automated tools that flag proposed changes against quantitative thresholds (5% RWA, 10% parameter shift) before human review. Reduce misclassification risk and create consistent, auditable classification records.
For novel methodology changes, new portfolio applications, or borderline materiality cases, initiate early dialogue with the ECB JST. Pre-submission engagement can save 3–6 months versus formal application rejection and resubmission.
Standardise model documentation templates to meet ECB expectations out of the box. Include development reports, validation reports, RWA impact assessments, and governance minutes. One-time investment that pays dividends across every future application.
Establish a lean notification process for non-material changes. Template-based filings with pre-populated fields and auto-generated impact summaries. Minimise operational burden while maintaining regulatory transparency.
Ensure front-line model owners understand materiality thresholds and can flag potential material changes early in the development cycle. Late discovery of materiality causes the worst timeline disruptions.
A phased approach to building model change management capability that meets CRR3 transition demands and ECB expectations.
A typical G-SIB needs 8–15 FTE dedicated to model change management during the CRR3 transition period. This includes model developers, validators, documentation specialists, and a dedicated ECB liaison function. Under-resourcing creates the single largest execution risk.