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?
The definition of exposures should be revised. This is because institutions that use the standardised approach (CR-SA) report a different exposure to IRB institutions. As the introduction of CRR 3 requires the CR-SA to be calculated for all exposures (and for the output floor in the case of IRB institutions), the exposure should be determined uniformly on the basis of the CR-SA requirements.
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)?
Loss data is robust information that can be reliably compared over a very long period. It is therefore particularly well suited to monitoring the stability of real estate financing in individual EU Member States across different economic cycles.
As there is no separate national real estate market for IPRE and non-IPRE, it does not make sense to analyse national real estate markets separately on the basis of this distinction. Furthermore, the hard test serves to demonstrate that a sustainably stable property market exists. The EBA should ensure that data in accordance with CRR Article 430a (3) is published for each national immovable property market.
With regard to Article 126(4) CRR, the EBA should examine whether the increase in the risk weight for commercial property financing from 50% under CRR 2 to 60% under CRR 3 for the senior tranche is actually justified on the basis of the analysis of loss data. Furthermore, it should be examined whether this justification applies to all Member States. Finally, an appropriate differentiation should be developed for a risk-sensitive reduction of the risk weight to the original level, for example using the hard test results based on Member State-specific loss data.
With regard to Article 465(11) CRR, we consider the institution-specific hard test to be a sensible and risk-sensitive criterion for calculating the output floor. Given the long-term nature of residential property financing, it is important to remove the time limit on this transitional provision as soon as possible in order to achieve planning certainty for lending conditions. The differing implementation of the output floor in other key jurisdictions, such as the UK, Canada or the US, also argues in favour of removing the time limit. For the sake of simplification, this transitional provision should not be structured as an option for Member States, but should apply uniformly across the EU. Furthermore, the transitional provision in Article 465(3) CRR for unrated corporates should also be made permanent, as it also plays a role in long-term property financing – specifically for the unsecured portion of the financing.
In addition, a comparable provision should also be introduced for commercial property, provided that the institution-specific hard test demonstrates a low risk profile.
Q3. Which elements of the real estate framework should be further simplified?
The overhaul of the CR-SA under CRR 3 has led to significantly more complex requirements. However, sufficient risk sensitivity has not been established for IRBA institutions. Therefore, on the one hand, the CR-SA should be significantly simplified for CR-SA institutions. A significant reduction in complexity could, for example, be achieved by dispensing with the differentiation between IPRE and non-IPRE positions for CR-SA institutions. At the same time, risk sensitivity for IRB institutions should be increased. The latter could, for example, be achieved by the full implementation of the hard test (maximum loss rates) in all EU Member States and in third countries with substantial financing from EU institutions. This means that the CRR must state more clearly than before that the data for the hard test must be published. Risk sensitivity for IRBA institutions can also be increased by expanding the gradation of risk weights depending on loan-to-value (LTV) ratios, particularly for commercial property financing.
Q4. Which other clarifications do you consider necessary to apply the new ECAI framework?
We welcome a mapping to the existing credit quality steps. This gives institutions certainty in the application of ratings that are explicitly designated as “without government support”. We do not expect any material deviation in capital requirements as a result, since only a few exposures to banks with expected government support exist (see para. 17).
Article 138(g) excludes institutions from further using an ECAI credit assessment in which implicit government support is assumed. However, point (g) does not explicitly require the exclusive use of stand-alone products, such as Fitch’s “ex government support rating”. Instead, this provision allows for the continued use of existing issuer ratings if no implicit government support is assumed in these ratings.
Whether this is the case can be determined based on the “intrinsic ratings” (see para. 14), which are disclosed and published by the credit rating agencies (CRAs) in their rating reports alongside the final issuer ratings.
- The SACP (stand-alone credit profile) of S&P has almost the same format but uses only lower-case letters, e.g. a+ instead of A+.
- The BCA (baseline credit assessment) of Moody’s has almost the same format but uses only lower-case initial letters, e.g. baa instead of Baa.
If these intrinsic ratings do not differ from the final issuer ratings (e.g. a+ = A+), no implicit government support is present, since otherwise the final issuer rating would be higher due to “notching up” (e.g. AA). In this case, the issuer rating of this CRA in question can continue to be used, as it does not contradict Article 138(g) CRR. Alternatively, the intrinsic rating could also be used, as the institution does not “derive” this itself, but uses a value published by the CRA. The institution would merely be assessing compliance with the requirement of Article 138(g) CRR, an obligation it is already subject to.
If such an assessment were not possible, several negative consequences would arise.
- Fitch would effectively have a monopoly with respect to stand‑alone “ratings without the assumption of implicit government support”. Such concentration on a single CRA cannot be in the interest of legislators and supervisory authorities regarding an appropriate risk measurement.
- The requirement to use new stand‑alone “ratings without the assumption of implicit government support” would constitute a regulatory benefit for the CRA, which could then sell two products instead of just one, with the same content, to institutions.
- Institutions and their service providers would have to accept a doubling of licensing costs without obtaining any added value beyond the information already available.
For the reasons set out above, we consider
a) a new mapping to be unnecessary,
b) an obligation to produce or obtain new stand‑alone “ratings without the assumption of implicit government support” to be neither required by law nor appropriate, and accordingly,
c) the use of intrinsic ratings or issuer ratings (where these correspond to the “intrinsic rating”) to be appropriate in order to simplify the credit risk framework.
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?
For institutions with established models, stability in ‘regulatory products’ is preferable, provided that this does not entail any regulatory changes.
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?
In our view, the current regulatory framework already provides sufficient guidance with regard to the consideration of environmental and social risks in credit risk modelling.
In practice, environmental and social risk factors are often already captured indirectly through existing risk drivers and financial variables used in rating models. Introducing additional prescriptive requirements could therefore create the expectation that ESG-related aspects should systematically have a separate or additional impact on creditworthiness, even where such effects cannot be robustly identified in the data.
From our perspective, institutions should not be required to:
- assume an additional credit risk impact of ESG factors beyond what is already reflected in existing model variables; or
- explicitly quantify the contribution of ESG factors to PD estimates where such effects cannot be robustly identified or reliably measured.
Further regulatory prescriptions in this area could increase modelling complexity without necessarily improving the risk sensitivity of the framework.
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)?
Institutions should not be deprived of the choice between ‘continuous’ and ‘discrete’ models, as both types should be selected according to their respective scope of application. In the case of ‘continuous’ models, evidence regarding discretisation could, where appropriate, make the different models comparable. It remains unclear to what extent, as per Discussion Box 4, a prescribed discretisation should necessarily result in a reduction in ‘unwarranted RWA variability’.
Furthermore, we would like to emphasise the need for a clear conceptual and regulatory distinction. This distinction should be made between genuinely continuous rating models and discrete master scale approaches, including those that may appear highly granular in certain portfolio segments.
The reference to “continuous or very granular rating scales” (cf. para. 30) in the discussion paper is insufficiently specified. Please clarify what constitutes a "very granular" rating scale, and specify whether discrete master scales with more rating grades or uneven grade distributions are included within this definition, to prevent unintended regulatory spillovers.
From our perspective, it is essential to clarify that any additional or harmonised testing requirements for continuous models must be limited to those that produce inherently continuous probability of default (PD) estimates. Discrete master scales, even those with relatively fine granularity in certain segments, should not be treated as continuous models for regulatory purposes.
This distinction is especially important for established rating frameworks. For example, centralised master scales in institutional protection schemes may use discrete rating classes based on data availability or portfolio characteristics, instead of continuous modelling. Imposing requirements for continuous models on these discrete systems would not improve comparability or risk sensitivity. Instead, it would trigger substantial and disproportionate implementation efforts, such as redesigning rating architectures.
Against this background, we strongly recommend:
- to provide a clear and operational definition of “continuous” versus “discrete” rating models,
- to explicitly exclude discrete master scales from the scope of requirements targeting continuous models, irrespective of their degree of granularity, and
- to ensure that no indirect obligation arises to adapt existing master scales solely due to their perceived granularity in individual portfolio segments.
In the absence of such clarification, there is a material risk that the proposed framework could lead to unintended consequences. This includes significant one-off implementation costs without corresponding benefits in model performance or supervisory comparability.
Institutions should not be deprived of the choice between ‘continuous’ and ‘discrete’ models, as both types should be selected according to their respective scope of application. In the case of ‘continuous’ models, evidence regarding discretisation could, where appropriate, make the different models comparable.
Q8. Which requirements should apply in the application phase of continuous models (e.g. overrides)?
We would like to highlight several concerns regarding the discussion on harmonising testing requirements for continuous and discrete rating models.
From our point of view, the discussion paper appears to take a rather critical view of continuous models. However, the use of discrete rating scales inherently involves a certain degree of arbitrariness, particularly with regard to the number, width and location of rating classes.
This raises several issues in the context of the objectives of comparability and simplification:
The comparability of model outcomes across institutions would not necessarily improve through the use of discrete scales, as model outcomes depend on the specific definition of rating classes, which may differ across institutions.
- Similarly, the level of risk-weighted exposure amounts (RWEA) may depend on the design of rating classes. Such dependencies do not arise in the same way for continuous models.
- For institutions currently using continuous models, a discretisation of model outputs would likely require recalibration or at least a comprehensive review of existing calibration functions. In our understanding, such changes could qualify as material model changes within the meaning of Annex I, Part II, Section 1.2(d) of Commission Delegated Regulation (EU) No 529/2014. To avoid disproportionate implementation costs, it would therefore be important to clarify that such adjustments would only require prior notification to the competent authority, rather than a full model approval procedure. Otherwise, the transition could lead to significant implementation efforts for both institutions and supervisory authorities.
- At the same time, mapping model outputs to discrete rating grades solely for the purpose of backtesting would not raise comparable concerns and could represent a pragmatic solution.
Finally, it is unclear whether the proposal is limited to PD models or is also intended to apply to LGD and CCF. For LGD, a continuous scale is common due to variable collateral values. We do not see the limitations of continuous models mentioned in Box 2 as applicable to LGD and CCF models.
For example: An LGD model for a low-default portfolio that assumes a linearly decreasing LGD with increasing collateralisation would, according to the proposal, no longer be permissible if certain ranges of collateralisation are sparsely populated in the discretisation. In this case, the continuous model is economically plausible, but the discretisation would force it to be artificially transformed into a discrete model (e.g. bins for the collateralisation ratio). This would result in cliff effects and adverse incentives (additional collateral would no longer affect the LGD classification, so there would be no incentive to obtain more collateral).
Q9. Which challenges have you encountered in implementing the new CRR definition of facility?
NA
Q10. Should a consistent and single facility definition be applied across all risk parameters?
The proposal to harmonise the definition of “facility” (C8) does not appear to provide clear simplification benefits. Default risk and loss risk are driven by different risk factors, which are often most appropriately captured at different levels (e.g. obligor level versus transaction level). A mandatory harmonisation of the estimation level could therefore reduce modelling flexibility without providing clear benefits.
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?
With regard to representativeness requirements applicable to PD estimation, we consider the existing regulatory framework to be adequate and sufficiently robust. Extending the representativeness requirements currently applied to the CCF parameter to PD and LGD estimation does not appear necessary from a risk measurement perspective. Such an extension could introduce additional modelling and documentation requirements, while the benefits in terms of improved risk measurement appear limited. We would therefore not see a need to extend representativeness requirements beyond the current framework.
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?
Simplifications are generally to be welcomed. However, they should not result in an excessive operational burden on institutions.
Q13. Should these simplifications be pursued? Do you have any preferred approaches with respect to these simplifications?
3.2.1 Simplified approach for Margin of Conservatism
MoC categories A and B
Fallback solutions can only be considered effective simplifications if they are readily applicable and do not implicitly create new benchmarks or supervisory expectations for more advanced modelling approaches.
One approach that could provide a practical simplification would be the introduction of materiality thresholds. For example, it could be considered that no MoC needs to be determined where a data deficiency affects only a very small share of the calibration sample (e.g. below 1%). Such thresholds could reduce unnecessary modelling complexity while maintaining an appropriately prudent framework.
MoC category C in low default portfolios
For low default portfolios, highly standardised approaches are often not well suited to the statistical characteristics of the available data and may result in excessively conservative or difficult-to-interpret outcomes.
The benchmarking approach used by the ECB derives an 80% or 90% quantile of the long-run average default rate based on the volatility of annual default rates under the assumption of a normal distribution. However, such approaches may present several methodological limitations:
- the number of observations underlying annual default rates is not sufficiently reflected;
- short-term spikes in observed default rates can significantly distort the measured volatility;
- the assumption of a normal distribution may not adequately reflect the statistical properties of default rate dynamics in low default portfolios.
In this context, institutions should be able to consider the results of analyses related to the likely range of variability of one-year default rates, where these provide a more appropriate methodological basis.
Moreover, with a 20-year historical observation period, the interpretation of an 80% quantile implies that in another 20-year observation period the default rate would remain below this threshold with a probability of 80%. In practice, however, institutions are subject to regular monitoring, validation and recalibration processes, ensuring that parameter estimates are updated continuously.
Purely quantile-based approaches may therefore effectively reflect extreme scenarios rather than realistic parameter estimates, potentially leading to an undue degree of conservatism.
3.2.3 Simplified approach for Downturn estimation
Simplified approaches to downturn estimation are welcomed as an additional option. However, relative to the existing approaches, they should be calibrated to be less excessively conservative so that they can be applied in Pillar II use cases as well (otherwise, there would be no simplification effect). If the requirements regarding back-testing and the selection of downturn periods remain unchanged, the resulting simplifications and reductions in modelling effort will be minimal.
For sovereign LGDs and for CCFs (particularly for corporates), the downturn concept should be fundamentally reassessed, as the causal link between the economic situation and LGDs for sovereigns and CCFs appears doubtful.
3.2.4. Simplified approach for estimation of LGD for defaulted assets
Standardised values for the add-on for additional losses after default would be appreciated. A significant reduction in modelling burden could only be achieved if the requirements for back-testing were lifted or reduced.
3.2.5 Apply the fixed IRB-CCF derogation to a larger scope
The discussion paper refers to the draft guidelines on Credit Conversion Factor estimation under Article 182(5) CRR (EBA/CP/2025/10 Draft Guidelines on CCF Estimation) in a manner that appears to assume their final adoption. These draft guidelines introduce significantly more complex and operationally demanding requirements, which should be carefully considered in the context of simplification.
In particular:
In cases where an internal CCF estimation according to the methodological requirements of CRR 3 can no longer be achieved, the practice of using the SA-CCF, as stipulated in the CRR, should be preferred, as it represents an appropriate and consistent approach. However, the discussion paper introduces that the application of the SA-CCF is supposedly linked to the use of the F-IRB-LGD. We view this linkage as problematic and contradictory to the CRR. The option to fall back on the SA-CCF without using the FIRB-LGD is both substantively preferable and legally required.
This “package deal” between CCF and LGD is asserted in the discussion paper without a clear justification as to why this principle should apply specifically to revolving exposures. For e.g. non-revolving exposures, the use of SA-CCF and AIRBA-LGD is explicitly prescribed, and there are no conceptual reasons why this should not also be applied to revolving exposures.
- The discussion paper and draft guidelines on CCF estimation appear to introduce situations where a minimum CCF of 100% should be applied instead of SA-CCF values. Such an approach appears difficult to reconcile with the provisions of the Capital Requirements Regulation and may lead to disproportionately conservative outcomes.
- Furthermore, the discussion paper would require institutions to model potential future credit decisions, for example for fully drawn credit lines or potential increases in credit limits. This effectively implies that institutions would need to hold capital for credit risks that have not yet been assumed.
- These aspects of the consultation paper, which partly contradict the CRR, imply legal and methodological uncertainty that fundamentally calls into question the continued consistent application of the A-IRBA and make a step back to the F-IRBA appear necessary.
Such modelling assumptions would also be unsuitable for many other modelling purposes, meaning that institutions might need to develop separate models for internal risk management purposes, which would increase rather than reduce modelling complexity.
Q14. Do you have any comments and suggestions with reference to the calibration of the fall back approaches?
Simplifications are generally to be welcomed. However, they should not result in an excessive operational burden on institutions.
For calibrating downturn fallback approaches and possibly the LGD in default, data from the Global Credit Data consortium could be used to derive standardised add-ons for each broad segment. Banks could then be responsible for ensuring an adequate mapping between their internal segmentation and the segmentation specified by the supervisory authority. Institutions should also have the flexibility to adjust the standardised values where the supervisory segmentation does not appropriately reflect their portfolios.
Q15. Do you see other potential simplification areas where the modelling burden is not commensurate to the gain in risk sensitivity?
Simplifications are generally to be welcomed. However, they should not result in an excessive operational burden on institutions.
Q16. What do you perceive as challenges in your capacity to collect appropriate data, in particular in relation to indirect costs?
NA
Q17. Do you agree with the approach proposed by EBA? Do you see further measures as necessary?
We welcome the proposed framework for considering six different criteria regarding the significance, scope, complexity and consistency of new regulations. Another measure that should also be considered is the proportional impact on the wide range of different institutions that are to apply the new regulations. In particular, the first three criteria (criticality of exposures, materiality of miscalibration and simplicity of the rules) have to be assessed differently by institutions and the impact varies depending on the size and business model of the respective institutions. Accordingly, the implementation of new regulations should always be proportionate and take the differences between SI, LSI and SNCI into account.
In view of the conflict of objectives between risk sensitivity and simplification, a distinction should be made between two categories: on the one hand, complexity that institutions can avoid by exercising their options, and on the other hand, complexity that cannot be avoided. If institutions can resolve this conflict of objectives themselves by, for example, opting for the CR-SA, the Foundation IRBA or the Advanced IRBA and thus determining the desired level of complexity themselves, this conflict of objectives need not be resolved by the regulator. The abolition of options that allow institutions to use more risk-sensitive requirements therefore does not constitute simplification from the institutions’ perspective and weakens their competitiveness.
However, the regulator should ensure that risk sensitivity is indeed higher in more complex approaches. This has not been the case for IPRE financing since the introduction of CRR 3, for example, if the hard test is not met or the relevant information is not available. This is because, under the CR-SA, real estate collateral is taken into account when determining the risk weight. In the Foundation IRBA, by contrast, the property collateral may not be taken into account when determining the risk weight without a hard test. This inconsistency must be resolved in the CRR, for which the EBA should issue a corresponding recommendation (for example, LGD 25% for LTV < 60%, LGD 30% for LTV 60–80%, LGD 35% for LTV > 80%). In practice, the risk weight under the CR-SA for lower-risk IPRE financing with lower loan-to-value ratios (LTV) is actually lower than the risk weight resulting from the Foundation IRBA.