Response to discussion on the simplification and assessment of the credit risk framework

Q1. For the purpose of reporting under CRR Article 430a, which definition of loss should be used?

NA

Q2. Should the loss data (CRR Article 430a) be used for the assessment of RWs of real estate exposures under CRR Article 126(4) and CRR Article 465(11)?

NA

Q3. Which elements of the real estate framework should be further simplified?

• Article 430(a) is used to determine both the preferential risk‑weight treatment for IPRE exposures under the Standardised Approach (SA) and the eligibility of immovable residential property collateral for LGD calculations in the IPRE corporate exposure sub‑class. However, as highlighted in the Discussion Paper, the scope of Article 430(a) is very broad, namely:
  • under the SA it includes all IPRE and non‑IPRE residential and commercial property values in the numerator, and all exposure values in the denominator; and
  • under IRB LGD calculations it includes all IPRE and non‑IPRE residential property values and exposure values in the denominator.
• In our view, using such an expansive scope dilutes the risk sensitivity of a ratio intended to reflect conditions in the IPRE market.
• Therefore, our proposal is to limit the scope of Article 430(a) exclusively to immovable properties secured by IPRE collateral (and therefore not also non-IPRE collateral). This narrower definition would ensure that the resulting ratio more accurately reflects developments in the IPRE market and is better aligned with the metric’s intended purpose.

Q4. Which other clarifications do you consider necessary to apply the new ECAI framework?

We welcome a temporary extension to use existing ECAI credit assessments. However, we would like clarity on two related aspects:

• Deadline: The original deadline of 31 December 2029 listed in CRR article 495e has recently been moved earlier to 1 January 2027 by the ECB (refer Article 24a of Regulation (EU) 2025/1520 of the ECB). We would like to clarify that the temporary extension means that the ECB regulation should be disapplied to the extent of this conflict (the earlier date), and that the original Level 1 deadline of 31 December 2029 still applies.
• Range of ECAIs: At present, compliant ratings (those excluding implicit government support) are available from only one major ECAI, namely Fitch. Reliance on a single provider creates concentration and potential systemic risks. We therefore consider that the above deadline should be further extended until compliant ratings for a broad range of obligors are available from more than one major ECAI. Furthermore, the onboarding of new ECAIs requires additional implementation period of at least six months to allow institutions to complete ECAI nominations, contractual negotiations, and necessary IT feeds, rating mapping, and systems changes.

Q5. Should the consolidation of regulatory products for credit risk be a priority or should the regulatory stability be preferable instead? Have you identified any redundancies in IRB products?

• Overall, we strongly support regulatory consolidation and simplification, provided it is strictly limited to clarification, harmonisation, and removal of redundancies, without introducing substantive changes that could affect existing models or compromise regulatory stability.
• We do not view consolidation of regulatory products and regulatory stability as mutually exclusive objectives. On the contrary, consolidation can – and should – be achieved in a way that preserves regulatory stability.
• At a minimum, consolidation efforts should focus on integrating and streamlining existing regulatory guidance without altering its underlying substance. By design, such an approach would enhance clarity and accessibility while avoiding unintended impacts on approved models or capital requirements. This would ensure that consolidation does not introduce instability.
• In our view, consolidation should be a clear priority. The current regulatory landscape for IRB approaches remains highly fragmented and complex, requiring institutions to interpret and reconcile multiple guidelines, RTS, and Q&As. This complexity often leads to inconsistent interpretations across banks and supervisors, and contributes to inefficiencies in model development, validation, supervisory review, and implementation processes.
• There are clear redundancies in the current framework. For example, overlapping guidance across EBA guidelines, regulatory technical standards, and subsequent Q&A clarifications often means institutions must consult multiple documents simultaneously. Recent consultations (e.g. EBA/CP/2025/10) explicitly refer to the need to read proposals in conjunction with prior “IRB repair” materials, highlighting the piecemeal nature of the current framework. Consolidating these elements into a single, coherent rulebook – or a more harmonised set of guidance – would significantly reduce duplication and improve usability and consistency across institutions and jurisdictions.
• At the same time, consolidation plays an important foundational role in enabling future improvements. A simplified and coherent regulatory baseline is essential to support further refinements, targeted simplifications, and the introduction of additional tools (in this DP) such as “fallback” approaches. In particular, the introduction of fallback methodologies, alongside consolidated guidance, offers a pragmatic way to support future model development while limiting disruption to existing, approved models.
• However, it is critical that consolidation does not inadvertently introduce substantive changes. There is a material risk that even well-intentioned consolidation could trigger recalibrations, additional validation requirements, or supervisory reassessments. Given the significant investments banks have recently made to implement IRB repair and related changes, further changes in the short term could impose a disproportionate operational burden, strain resources, and introduce unnecessary volatility in risk estimates, and capital outcomes. As mitigation, where such inadvertent or unavoidable regulatory changes occur during the consolidation, clear transitional arrangements must be provided.
• From a governance and planning perspective, a stable regulatory environment remains essential. Frequent or implicit changes can undermine confidence in the framework and make it more difficult for institutions to maintain compliant, validated models and stable capital planning processes.
• Finally, alignment between regulatory products issued by different authorities (notably EBA and ECB) is equally important. Reducing inconsistencies and potential contradictions across these bodies would further enhance clarity, comparability, and regulatory certainty. Also see our response to Q17.

Q6. Do you consider that the integration of environmental and social risks into the credit risk framework could be further enhanced without undermining its simplicity? Which areas, if any, would you prioritise for further work or clarification?

Overall Position: Enhancing Integration While Preserving Simplicity

• We believe that the integration of environmental and social (E&S) risks into the credit risk framework can be further enhanced without undermining simplicity, provided that enhancements build on existing mechanisms, respect data limitations, and remain aligned with the framework’s core design principles.
• Overall, we support further integration of environmental and social risks into the credit risk framework, provided that:
  • Pillar 2 remains the primary tool for forward-looking risks,
  • Pillar 1 integration respects its data-driven design constraints,
  • enhancements rely on existing modelling mechanisms,
  • data standardisation is strengthened, and
  • implementation is prioritised and phased.
• This approach enables meaningful and robust integration while preserving simplicity, comparability, and the integrity of the credit risk framework

Role of Pillar 2: Primary Channel for Forward-Looking Risks

• Existing supervisory guidance provides sufficient direction for incorporating E&S risks, particularly under Pillar 2 frameworks such as ICAAP and stress testing. These tools are well suited to capture the forward-looking, long-term, and systemic nature of E&S risks. Recent publications (e.g. ESA joint guidance JC 2025 78) further reinforce how these risks can be embedded in stress-testing frameworks.
• Given the structural characteristics of E&S risks, Pillar 2 should remain the primary channel for addressing forward-looking risks, concentrations, and institution-specific exposures, enabling proportional implementation without undermining the simplicity of Pillar 1.

Pillar 1 Design Constraints and Need for Clarification

• Further clarification is required regarding the integration of E&S risks into Pillar 1, particularly IRB models. There is an inherent tension between:

1. the requirement to include all material risks, and
2. the reliance on observed historical data and a 12-month risk horizon.

• E&S risks often materialise over much longer time horizons and are not yet sufficiently reflected in historical data. This creates structural limitations in their reliable quantification within IRB models that would, amongst other undermine IRB-model back-testing and can result in double-counting impacts estimated under Pillar 2.
• Supervisors should explicitly acknowledge this constraint and clarify that E&S risks traditionally assessed under Pillar 2 are not expected to fully migrate into Pillar 1 in the near term. Attempting to force such integration prematurely risks undermining alignment with Basel design principles, reducing comparability across institutions, and weakening model robustness.

Organic Integration Through IRB Model Monitoring

• IRB model monitoring already plays a key role in identifying emerging patterns or early signals of E&S risks. Over time, as these risks materialise in observed default behaviour and become statistically meaningful, they can be naturally incorporated into IRB models.
• This organic evolution ensures that Pillar 1 models remain evidence-based while gradually capturing E&S risks when supported by sufficient data.

Structural Challenges to Integration

• The integration of E&S risks faces several structural challenges:
  • Data limitations: ESG data are incomplete, inconsistent across sectors and geographies, and lack sufficiently long time series.
  • Uncertain transmission mechanisms: The empirical relationship between E&S factors and credit risk parameters (PD, LGD, CCF) is still developing, particularly for long-term transition risks.
  • Time-horizon mismatch: E&S risks often materialise over longer horizons than those used for credit risk calibration.
  • Regulatory fragmentation: Expectations are spread across Pillar 1, Pillar 2, and disclosure frameworks, increasing complexity.
• These constraints require a cautious and prioritised approach to further integration.

Enhancing Pillar 1 Within Existing Design Principles

• Enhancements should focus on leveraging existing modelling mechanisms rather than introducing new complexity:
• Transmission Channels via Financial Risk Drivers
  • Where financial drivers are already used in IRB models, institutions should assess how E&S indicators (e.g. emissions intensity, environmental fines, physical-risk exposure) correlate with these drivers over time. This helps evaluate whether financial metrics act as transmission channels for E&S risks (as outlined in EBA/REP/2021/18 and EBA/REP/2023/34).
  • For non-retail portfolios, financial drivers should be prioritised for inclusion – even where their immediate explanatory power is limited. Despite practical challenges (e.g. data gaps, low frequency, limited histories), their inclusion provides an essential foundation for capturing E&S risks in the future and avoids over-reliance on behavioural or non-financial variables.
• Segmentation Improvements: Institutions should test segmentation by geography and sector, given that E&S risks vary significantly across regions and industries (e.g. flood risk, drought exposure, transition policies). Where segmentation yields meaningful risk differentiation, these segments should be considered for model calibration.
• Targeted Expert Judgement and Overrides: While the classic transmission-channel approach may show limited materiality at overall portfolio level, this does not imply that E&S risks are irrelevant – pockets of obligors may still be materially exposed, especially in models covering broad segments. For these obligors, manual overrides should be used.
• Explicit Inclusion of E&S Risk Drivers: Where data allows, E&S risk drivers can be incorporated into IRB models via the same pathways as traditional financial drivers. While this may increase the pool of candidate variables, it does not inherently increase model complexity if properly prioritised.

Targeted Use of Margins of Conservatism (MoC): 

• MoC should be applied in a targeted and proportionate way:
  • Category A (data limitations): Where E&S drivers are included but data are incomplete.
  • Category B (representativeness): Where observed E&S impacts in the application portfolio are not reflected in estimation data. Importantly, given Pillar 1 design principles, only observed impacts in the application portfolio should be considered – long-term forward-looking scenarios should not be incorporated under this category.

Data Standardisation and Taxonomy Alignment: 

• A key priority is improved standardisation of E&S data and definitions. Currently, inconsistent taxonomies, lack of verification standards, and differing levels of granularity undermine comparability across institutions.
• Alignment is needed on:
  • common E&S risk taxonomies,
  • standardised definitions of physical and transition risks,
  • harmonised severity scales and data methodologies.
• Without this, identical risks (e.g. drought exposure) may be measured differently across banks, leading to inconsistent modelling outcomes.
• A common taxonomy, coupled with standardised severity levels and risk drivers, is essential. This view is also broadly consistent with the proposals set out in the EBF report Climate & Environmental Credit Risk Data and Modelling (October 2025).

Prioritisation of ESG Components: 

• Finally, since the ultimate objective is to ensure that the full suite of “ESG” risks is eventually captured (as per EBA/REP/2023/34)—whether under Pillar 1 or Pillar 2—additional supervisory guidance is needed on which components of ESG should be prioritised first.
• A prioritisation approach similar to that outlined in the ESA Joint Guidance (JC 2025 78) would help. Given the breadth and heterogeneity of ESG factors, focusing first on the most material and most quantifiable components (for example, specific environmental transition or physical‑risk drivers) would foster greater consistency across institutions, provide clearer supervisory expectations, and reduce the risk of fragmented modelling practices.
• This would:
  • improve consistency across institutions,
  • reduce fragmentation, and
  • provide clearer implementation expectations.

Q7. Which requirements should apply in relation to the measurement of the performance of continuous models (e.g. Back-testing)? How could testing requirements be facilitated and enhanced for continuous models that are compliant with CRR, Part three, Title II, Chapter 3, Section 6 (Requirements for the IRB approach)?

Overall Position: Leverage Existing Guidance and Avoid Additional Complexity

• We believe that the current supervisory framework already provides sufficient guidance for the measurement and performance assessment of continuous models (i.e. direct estimates). Any further requirements should focus on alignment and clarification rather than introducing new or more granular rules, which could increase complexity and reduce coherence.
• Specifically:
  • The current ECB guidance* provides a sufficient and robust framework for the performance assessment of continuous models (*ECB guide to internal models and ECB validation reporting),
  • No substantial additional requirements are necessary, beyond improved alignment across regulatory bodies, and
  • Any enhancements should focus on practical facilitation, such as potential standardisation of discretisation approaches for back-testing.

Need for Regulatory Alignment Rather Than Expansion

• Supervisory expectations for continuous models are already clearly articulated in the EGIM, particularly in paragraphs 250, 285, and 321.
• Rather than expanding the framework, the key priority should be alignment between EBA and ECB guidance. Ensuring consistency across regulatory sources would:
  • improve clarity and usability,
  • reduce the need to reconcile multiple documents, and
  • support consistent supervisory practices across jurisdictions.
• Without such alignment, additional guidance risks further scattering requirements and increasing operational burden without clear benefits.

Applicability of Standard Performance Testing Approaches

• Continuous models can broadly follow the same performance testing framework as discrete models. However, certain tests (e.g. homogeneity or heterogeneity assessments) require discretisation of continuous outputs.
• The EGIM already acknowledges this and provides guidance by allowing performance assessment at sub-range (bucket) level, ensuring that continuous models can be tested in a manner consistent with regulatory expectations.

Discretisation Approaches: Flexibility vs Standardisation

• Currently, institutions have flexibility in defining sub-ranges for discretisation. While this allows proportionality and model-specific adaptation, it may also lead to variation across institutions.
• If further enhancement is considered, one potential improvement could be: the introduction of standardised discretisation approaches (e.g. fixed bucket structures) for back-testing purposes.
• Such standardisation could:
  • improve comparability across institutions,
  • reduce supervisory uncertainty, and
  • streamline validation processes.
• Importantly, precedent already exists – for example, in the ECB validation reporting for LGD and CCF models, where standardised buckets are used for performance assessment.

Q8. Which requirements should apply in the application phase of continuous models (e.g. overrides)?

• We consider that no additional requirements specific to continuous models are necessary for the application phase, including the use of overrides.
• Overrides inherently rely on expert judgement, regardless of whether they are applied in discrete models (e.g. rating notching) or continuous models (e.g. applying a proportional adjustment to the estimated risk parameter). As such, introducing additional model-specific requirements for continuous models would not improve consistency, create a level playing field, or reduce unwarranted RWA variability.
• The existing supervisory framework for overrides – covering governance, justification, and monitoring – Is sufficient and should continue to apply equally to both discrete and continuous modelling approaches.
• While notching is not directly applicable to continuous models, alternative approaches are already feasible, such as:
  • applying relative (percentage-based) adjustments to model outputs, or
  • using discretisation of continuous estimates (e.g. bucket-based approaches) to support override decisions.
• If further facilitation is considered, aligning any discretisation used for overrides with that applied in performance testing could enhance consistency and operational efficiency. However, such measures should remain optional and supportive rather than prescriptive.
• Overall, the focus should remain on consistent governance of expert judgement rather than introducing additional requirements specific to continuous models.

Q9. Which challenges have you encountered in implementing the new CRR definition of facility?

• Overall, we have not encountered material challenges arising directly from the new CRR3 definition of a facility, as our existing internal definitions were already aligned.
• However, some practical implementation challenges arise in ensuring consistent application across models, systems, and processes. These include:
  • determining the appropriate level of facility aggregation for different risk parameters,
  • handling facilities that combine revolving and non-revolving components,
  • managing facilities with both amortising and non-amortising structures,
  • addressing exposure migration between facilities over time (e.g. product switching),
  • linking facilities that are created during default restructuring to prior performing exposures, and
  • reflecting structural changes in facilities throughout their lifecycle.
• Overall, the new definition is not seen as problematic in principle, but its consistent operationalisation across portfolios, time, and data histories requires careful handling.

Q10. Should a consistent and single facility definition be applied across all risk parameters?

• In our view, applying a single facility definition across all risk parameters is not advisable. This is primarily because different aggregation levels are appropriate for different parameters, and because facility structures can be inherently complex.
• In particular:
  • Collateral management, which is central to LGD estimation, often takes place at a higher aggregation level than limit management, which is more relevant for CCF estimation. As a result, different aggregation levels may be required for LGD versus CCF.
  • Facilities may include a combination of revolving and non-revolving components, which often need to be disaggregated for accurate CCF estimation.
  • In practice, there are also custom and heterogeneous facility structures that cannot be adequately captured within a single, uniform definition.
• Enforcing a uniform facility definition across all risk parameters could therefore lead to sub-optimal risk differentiation and less accurate estimation, as it would not reflect how risks are actually managed and measured (see also our response to Question 9).
• Furthermore, any misalignment between the facility definition used in modelling and the aggregation required for reporting purposes (e.g. accounting or regulatory reporting) should not be considered problematic. Model outputs can always be aggregated to higher levels or allocated to lower levels depending on the reporting objective.
• Overall, flexibility in facility definition – tailored to the specific requirements of each risk parameter – is essential and should be maintained.

Q11. Are adjustments proposed in the representativeness requirement for the CCF parameter also suited for PD and LGD risk parameters? Which amendments would be needed to accommodate PD and LGD specificities?

• In our view, the simplifications proposed in the CCF consultation (EBA/CP/2025/10) for assessing representativeness are largely applicable to PD and LGD without fundamental changes. The core elements of representativeness assessment such as scope of application, definition of default, risk characteristics, credit management practices, and market conditions – are common across all IRB parameters.
• These principles can therefore be consistently applied across PD, LGD, and CCF, supporting greater harmonisation. The main exception relates to likely-range-of-variability analysis, which is specific to PD and would need to be considered separately.
• While the simplification efforts are welcome, practical challenges remain, particularly for LGD. For example, unresolved representativeness issues may complicate comparisons between training, testing, and application samples, requiring careful handling to ensure consistency and robustness.
• Overall, we support extending representativeness simplifications across all parameters, provided that parameter-specific features are appropriately accounted for.

Q12: Do you consider further simplification of the representativeness requirement, as proposed for the CCF parameter, as necessary for PD and LGD and if so, what kind of simplification?

See our input for question 11.

Q13. Should these simplifications be pursued? Do you have any preferred approaches with respect to these simplifications?

Principles:

• We support the introduction of these optional simplifications and fallback approaches.
• In our view, offering these simplifications on an optional basis is an elegant way to update supervisory expectations without immediately affecting existing approved internal models. Institutions can adopt the simplified approaches over time, for example when revising models, thereby ensuring smooth and non‑disruptive implementation.
• In our view, the optionality granted to institutions should be unconditional. Institutions should not be required to perform an extended analysis to justify that a fallback approach is appropriate, because such a prerequisite would undermine the intended simplification goal. This position is based on the understanding that the fallback approaches will be conservatively calibrated and therefore not susceptible to arbitrage. The only pre-condition should be the availability of sufficient data in cases where the fallback method relies on data inputs.
• We also support allowing institutions to apply a fallback method to a specific modelling aspect independently of whether fallback approaches are used for other modelling components.
• We propose the following variations to the fallbacks proposed in the DP:

LGD for defaulted exposures (C13):

• We support simplifications for the LGD for defaulted exposures. We also support the use of back testing of these fallback values as stated in par. 51 of the DP. 

Downturn quantification (C12): 

• In our view, the existing guidance on downturn identification and quantification—such as EBA/RTS/2018/04, CDR(EU)2021/930, EBA/GL/2019/03—is extensive, but at the same time extremely difficult to apply and comply with in a fully unambiguous manner.
• We agree that using the reference‑value method for LGD and CCF parameters provides a solid basis for a fallback approach. As defined in the LGD downturn guidelines EBA/GL/2019/03 (and the forthcoming CCF estimation guidelines following EBA/CP/2025/10), the reference value is calculated as the average of the two years with the highest observed LGD or CCF values.
• Practically it means that the reference-value method, which is currently a non-binding challenger for the downturn quantification (as per EBA downturn GL and EGIM 2025), is elevated to be the primary calculation method, should the institution select this fallback method for downturn quantification.
• The advantages of this method include eliminating the need to identify downturn periods and significantly reducing modelling complexity.
• However, the method may unintentionally introduce bias by always selecting the most adverse outcomes – regardless of whether these reflect genuine economic downturn conditions (which is the intention with downturn quantification) or are driven by institution‑specific idiosyncratic events (which is not the intention). Furthermore, data quality issues, or non-representative defaults (e.g. fraud or missing cash flow cases) and small sample sizes may further distort outcomes.
• To mitigate this risk, we propose adjusting the reference‑value calculation to use more than two years for the average calculation, e.g. the three worst years. This would reduce sensitivity to outlier years while maintaining an appropriately prudent calibration. A further mitigation is allowing well-justified deviations where institutions can demonstrate that the reference value does not reflect a plausible downturn scenario.
• Additional considerations include the need for a minimum of seven years of observations for the fallback method to be applied credibly, on the basis that seven years broadly represent the average duration of an economic cycle. Furthermore, for the purpose of annual calculations, calendar years – rather than rolling 12‑month periods – should be used to support the simplification objective.

Fixed CCFs (C14)

• We support optional fallback values for CCFs provided that the AIRB LGD is still supported, i.e. the use of the fixed CCF does not require reversion to the FIRB or SA for the LGD.

CCF 12-month reference period (C15): 

• The fixed 12‑month reference period for the CCF risk parameter introduced in CRR3 (Level 1) represents a constraint that is not fully aligned with the flexibility permitted for selecting reference periods for other risk parameters. We therefore support the proposed additional flexibility in this regard.
• The exact nature of the “additional flexibility” in Section 3.2.6 of the DP is not clear, and seems to be contradicting the most recent specification in EBA/CP/2025/10. We therefore request that EBA confirm or correct the following interpretation: Our understanding is that the “flexibility” is based on EBAs interpretation that the 12-month reference period stipulated in CRR3 Article 182(1)(g) refers to the risk quantification. Therefore, the flexibility allows the risk-differentiation model to use a cohort approach where the reference period may deviate from a fixed 12 months. This is desirable to align with a cohort methodology used for other risk parameters, e.g. the IRB-LGD. However, the risk quantification level must be compared against the outcome derived using a 12-month fixed reference period. The outcome of this comparison may result in further changes to the IRB-CCF risk quantification level to demonstrate Level 1 compliance. In addition, the scope of observations to consider for both the risk differentiation and risk quantification should be determined using the 12-month fixed reference period to align with the 12-month requirement stipulated in Level 1 (CRR3 Article 182(1)(g)) and the requirement to use all observations in risk quantification.
• For future revisions of the Level 1 text, we respectfully suggest that Level 1 should avoid prescribing detailed modelling specifications—such as reference‑period selection methods. These technical elements are more appropriately addressed in Level 2 or Level 3 texts, where they can be updated more efficiently and aligned consistently with other modelling requirements.

LGD direct and indirect costs (C12):

• We do support the proposal of fixed percentages for these costs.
• Defining indirect costs on a relative rather than absolute basis is appropriate, as absolute minimum values would unfairly penalise small exposures (e.g., EUR 1,000) compared with large ones (e.g., EUR 10,000,000). Given that indirect costs are typically immaterial, a simple and pragmatic approach is justified.
• Also see our related responses to questions 14 & 16.

Standardised MoC C (C10):

• In general, we support the standardisation of MoC C as outlined in paragraph 44(b) of the DP, provided that institutions may deviate from this method under specific circumstances (as discussed below).
• With respect to MoC A and MoC B, we are less supportive of standardisation due to the typically idiosyncratic nature of the deficiencies that are meant to be mitigated with these MoC categories, which may differ per model.
• We propose a standard error–based approach in which the statistical uncertainty of PD, LGD, and CCF observations forms the basis for conservatism. For PD, defaults are treated as binomial events, and the standard error is derived from the variance of the observed default rate. For LGD and CCF, the standard error is calculated from the empirical variability of the observations – whether these are continuous values or assigned to predefined buckets. In the bucketed case, the calculation is based on the variance of observations within and across buckets to capture dispersion appropriately.
• This straightforward and transparent measure of sampling uncertainty aligns with the EBA’s intention to harmonise MoC C practices across institutions. It limits modelling heterogeneity while maintaining proportionality, as higher MoCs naturally arise in data‑sparse or more volatile segments.
• At the same time, institutions should be allowed to deviate from the standard error method under specific circumstances. Examples of specific circumstances include extremely limited observations, structural changes in the portfolio, or data quality issues that impair the representativeness of the sample.

Q14. Do you have any comments and suggestions with reference to the calibration of the fall back approaches?

• Avoid excessive conservatism: It is expected that the calibration of the fallback methods may include some conservatism so that the fallback options are not susceptible to arbitrage. However, it is important to avoid introducing excessive conservatism. This may deter the use of these simplifications thereby undermining the objective of institutions adopting simpler methods. Furthermore, excessive conservatism in these simpler methods will compromise the risk sensitivity.
• Direct and indirect costs (C11): The calibration of fallback percentages should reflect that these costs typically vary by exposure class, jurisdiction, collateral type, and other portfolio characteristics. An industry‑wide survey can serve as a useful benchmarking tool to derive fallback values that are representative across institutions.
• Downturn quantification (C12): See our proposal under question 13.
• Where feasible, the calibration of fallback approaches should be based on the institutions own data to improve representativeness and reduce undue conservatism.

Q15. Do you see other potential simplification areas where the modelling burden is not commensurate to the gain in risk sensitivity?

We propose the following potential simplification areas:

Maximum Recovery Period: 

• The estimation of the maximum recovery period typically requires disproportionate analytical effort while contributing only marginally to overall risk sensitivity. As a simplification, we propose a fallback approach whereby the maximum recovery period is set equal to the time‑in‑default corresponding to the 99th percentile of the cumulative average recovery curve (calculated across all default vintages).
• In our view, the result of this fallback method for the maximum recovery period (applicable to the LGD parameter) is also suitable to use as the value for the maximum drawing period (defined in EBA/CP/2025/10) applicable to the CCF parameter. 

MoC consideration for in-default segments:

• With respect to LGD in-default reference-date segments, computing separate AA and MoC values per reference date is operationally burdensome, primarily because later reference date buckets typically contain very few observations. A simplified approach, such as applying a single MoC across all reference dates, would therefore be appropriate, especially given the limited materiality of in default portfolios compared with performing ones.
• Additionally, introducing a fixed EBA percentile for the MoC calculations could be beneficial to ensure a more structured and transparent approach to conservatism.

Downturn quantification:

• Alternative for Type I approach: Rather than requiring the full four aspect analysis in par. 27(a) of EBA/GL/2019/03 (elevated realised LGDs, cure rates, time in default, recoveries) and combining these into a coherent downturn framework, a more pragmatic alternative could be to allow institutions to focus solely on elevated LGD levels. This may still capture the main downturn driver while significantly reducing complexity. This approach aligns with the additional interpretation provided in EGIM (par. 303, Credit Risk Section), which seems to give preference to the realised LGDs under par. 27(a).

LGD Cash flow labelling: 

• NCAs require us to be able to label all individual cashflow components used in the LGD calculation. This does pose a practical challenge for older defaults for which granular cash flow data may no longer be available due to e.g. system changes and/or default dates pre-dating acquisition dates. Instead, the movement in monthly exposures are used to infer monthly cashflows, supplemented with identification of select types of cash flows, e.g. collateral sales. However, NCAs do not regard this as sufficient which then results in additional findings and obligations.
• Where granular cash flow data is not available, and cash flows are inferred from changes in exposures between consecutive dates, we propose that the requirement to label cash flows be clarified as applying to only specific types of cash flows. The list of applicable cash flows would only be those that can add discriminatory value, whereas the rest can be aggregated without reducing risk sensitivity of the LGD estimates. This list should take the diverse aspects of portfolios into account.
• This would significantly simplify the construction of cash flow data for LGD modelling, and align supervisory expectations without reducing risk sensitivity.

Q16. What do you perceive as challenges in your capacity to collect appropriate data, in particular in relation to indirect costs?

• A key challenge in collecting recovery-related cost data – particularly indirect costs – is that these costs often originate from highly aggregated, shared service functions within the institution. These services typically support a wide range of activities beyond recovery processes, making it difficult to isolate recovery-specific cost components.
• In addition, some costs that are technically direct in nature may still be processed through shared service structures. Due to operational practices (e.g. batch processing), these costs are not always allocated to individual recovery cases, leading to further gaps in granularity.
• Another challenge relates to data representativeness over time. Current cost structures within shared services may not accurately reflect the cost levels applicable to older recovery cases, which are often required for model estimation. This creates additional uncertainty when aligning historical recovery data with current cost information.
• As a result, attributing recovery-specific indirect costs and unallocated direct costs to individual portfolios necessarily relies on assumptions, introducing a degree of modelling uncertainty.
• Given these practical limitations, the use of fallback approaches is preferable, as they can help to reduce variability across institutions and provide a more consistent and pragmatic solution.
• Finally, flexibility in the application of fallback methods is important. Institutions should be able to opt for fallback approaches for indirect costs, direct costs, or both, depending on the specific data availability and characteristics of the portfolio under consideration.
• Overall, challenges are primarily operational and data-related, and a flexible fallback framework is key to ensuring robustness and comparability.

Q17. Do you agree with the approach proposed by EBA? Do you see further measures as necessary?

EBA and ECB alignment on modelling guidelines:

• In our recent feedback to the EBA on the CCF estimation consultation (EBA/CP/2025/10), we identified several areas where the 2025 ECB Guide to Internal Models are not fully aligned with the EBA guidelines, including the proposals in EBA/CP/2025/10.
• The consequences of the above are inconsistent interpretation, additional validation and supervisory feedback cycles, and delays in model approval and implementation. These consequences are expensive, and lead to frustration since institutions must repeatedly adjust models to satisfy diverging expectations, often without a clear view of the ultimately binding standard. This undermines planning certainty, consumes significant resources, and detracts from the objective of achieving a stable, predictable, and harmonised regulatory framework.
• In our view, although not in the scope of this DP, achieving consistency between EBA and ECB products is a natural and necessary extension of the objective to improve regulatory stability and clarity. Ensuring alignment across these supervisory texts would help prevent contradictions and support a more coherent and predictable regulatory framework.

Rule-based vs. principle-based regulation

• We welcome the EBA’s efforts to simplify the prudential framework. However, we are concerned that the approach outlined in the Discussion Paper of adding more rules and options may ultimately result in greater complexity. Our understanding of the broad simplification proposals in the 2024 Draghi Report on European Competitiveness, which is the original catalyst for this DP, is a clear call to move away from highly prescriptive regulation toward a more principles‑based framework. By introducing additional rules and options, the approach proposed in the DP risks further entrenching a rules‑based system, which appears misaligned with the intent of the Draghi Report.
• We therefore propose that the prudential framework should progressively evolve toward a more principles‑based approach as a sustainable path to simplification, supporting supervisory consistency while maintaining robust prudential standards.

Logistic and linear regression models and the AI Act

• Although not formally within the scope of the EBA products, the AI Act’s broad definition of an “AI system” has created legal uncertainty as to whether widely used logistic and linear regression (LR) models for IRB and credit scoring fall within scope, including through their Use‑Test link to creditworthiness assessments of natural persons, which are classified as high‑risk under the AI Act.
• Depending on the interpretation applied, this has practical implications for internal models governed by the EBA products.
• As these LR models are static, non‑adaptive, and fully transparent techniques implemented to meet IRB requirements, we consider that they should be explicitly excluded from the scope of the AI Act through a clear statement in the relevant EBA product, in order to enhance legal certainty, proportionality, and regulatory consistency.

Zero-Flooring of observed CCFs

• The introduction in CRR3 Article 182(1) of a zero floor for observed CCFs is inconsistent with the Basel Accord’s original design principle, which applies an effective output floor at the level of the current drawn amount (“EAD at no less than the current drawn amount”).
• This change in CRR3 shifts the output floor (after downturn adjustments and MoC) to an input floor (before downturn adjustments and MoC).
• This introduces additional conservatism in the estimated EAD compared to the estimated EAD under the original Basel design principle. This reduces risk sensitivity, undermines product‑specific differentiation, and results in estimates that are no longer suitable as true best estimates, thereby weakening the Use Test.
• We therefore propose clarifying at Level 2/3 how input floors should interact with the best‑estimate principle to preserve risk‑management usability, and ultimately to realign this Level 1 requirement with the Basel framework.

Upload files

Name of the organization