Response to consultation on Regulatory Technical Standards that specify material changes and extensions to the Internal Ratings Based approach
Question 1. Do you have any comments on the clarification of the scope of the revised draft regulatory technical standards to specify the conditions for assessing the materiality of the use of an existing rating system for other additional exposures not already covered by that rating system and changes to rating systems under the IRB Approach?
The FBF appreciates the effort to clarify the scope of the draft revised RTS. The data elaboration improves the interpretation of the Level 1 text.
However, we note that :
• Mechanical recalibration following backtesting or mechanical updates to risk factors due to the addition of an extra year of default should be notified after implementation to reduce the burden. If this is not the case, in particular when machine learning techniques are used, the numerous changes which are structurally part of the models would generate prior notifications / approvals thus congesting the process.
The FBF proposes precising recital (2) to reflect these distinctions:
- New proposal
"Modifications to rating systems as defined in Regulation (EU) No 575/2013 may have a potential impact on internal risk estimates used for the calculation of the risk-weighted exposure amount and, as such, include changes affecting the scope of a rating system, the rating methodology for IRB systems, the definition of default and the validation framework, as well as changes to relevant processes, data and model usage. Updates to data used in the development and calibration of rating systems requiring human intervention should therefore be covered by this Regulation. In the case of mechanical recalibrations following regular updates of data, such changes could be subject to ex-post notifications. In addition, changes to remediate data quality issues (e.g. amending missing/incorrect LTV input data) in order to improve the modelling framework are not covered by this Regulation."
--------
• As an alternative proposition, it could be stated that if during the validation/authorization phase of the model the supervisor has accepted/foreseen the use of continuous updating mechanisms as a core element of the authorized machine learning model, the update of these mechanisms does not trigger a material model change.
• Regarding the inclusion of relevant ESG risk factors in IRB models, only when future regulation regarding these aspects will be settled, the FBF suggests a transitional arrangement to incorporate ESG-risk drivers (if considered relevant and material) in order to avoid delays. This could involve a period where banks submit ex-ante notifications to include ESG risk factors without needing to perform a complete analysis of the ranking or distributions of obligors.
• As envisaged in the EBA statement on CRR3 implementation, changes mandated by regulation (such as reversion to a less sophisticated approach) that do not affect model performance should be permanently outside the scope.
Question 2. Do you have any comments on the clarifications and revisions made to the qualita-tive criteria for assessing the materiality of changes as described in the Annex I, part II, Section 1 and Annex I, part II, Section 2?
• The FBF welcomes the EBA's initiative to improve the distinction between changes that result in a significant change and those that require ex-ante notifications for the Definition of Default (DoD).
• The EBA requirements are understood as leading to encompass in material categorization an extensive range of changes to definition of default which are related mostly to operational implementation issues. Not only these changes may not impact rating systems but such changes do not prevent the well-functioning of risk management functions. For instance, in a change of the batch frequency for defaulting exposures from monthly to daily (with exact date and number of days being stored in any case), the impact is only on the frequency at which the new defaults are stored in the systems. Such change is an an improvement of the framework, does not modify the result of the identification of default and do not impact the rating systems. This is why in the first place, we would introduce a distinction between changes with an impact on ratings systems and changes with no impact on rating systems. In addition, the list of cases underlined in the RTS would not be exhaustive enough to cover all possible cases and will be subject to various interpretation, except for the change of level of application of definition of default (which will probably imply a rebuilding of the model). In this perspective, the most objective way to assess the impact would be to rely more on quantitative metrics (e.g.: impacted default volumetry).
• The FBF proposes to amend the RTS (Annex I, Part I, Section 1) to clarify changes to the definition of default in accordance with Article 178 of Regulation (EU) No 575/2013:
NB : please refer to accompagning document for Before/After
New proposal:
---------------------
3. Changes in the methodologies and rules to the definition of default according to Article 178 of Regulation (EU) No 575/2013,
(a) they change the level of application of the definition of default for retail exposures according to Article 178(1), second subparagraph of Regulation (EU) No 575/2013;
b) they change the default classification in the reference dataset or scope of application of a rating system in a significant manner, the measure and level of which will have been defined by the institution. Institutions could define such metrics for instance on the relative impacted volumetry of default.
-----------------
Such change is also accompanied by the deletion of the quantitative criteria, as impact on risk parameters will be captured by other criteria of the RTS when assessing the materiality of changes on models following a change in the definition of default. In other words, the change is assessed in two-step approach: first on the change in definition of default, second on the impact on models.
• Changes in the methodology used to assign exposures to different exposure classes (in accordance with Article 147 of the CRR) are reclassified as ex-ante notifications in the new RTS. The FBF believes that these changes should be excluded from the scope of the RTS when they do not affect the models. Indeed, the EBA mandate according to article 143(5) of CRR is to write “standards to specify the conditions for assessing the materiality of the use of an existing rating system for other additional exposures not already covered by that rating system and changes to rating systems under the IRB Approach”. The EBA therefore presumes that rating systems and exposure classes are interlinked, meaning that any change to exposure classes will impact rating systems. However, such link is not always the case in practice
• As mentioned in the criteria 2(f) from Annex I, Part II, Section 1 of the new proposed regulation, change in the fundamental methodology for estimating PD/LGD now encompasses methodology for deriving appropriate adjustments and should be considered as a material change (ex-ante notifications otherwise). For these cases, it is within the remit of the bank to define what constitutes a “change in the fundamental methodology”. In this exercise, difficulties may appear as from a supervisory perspective, this fundamental feature could not solely be based on RWA impact. Depending on the size of the exposures/models for the bank, the limit between material change and non-material change is objectively captured through RWA impact, rather than through more subjective criteria determined by the bank. This is why the limits between material and non-material regarding methodologies for estimating PD/LGD should be based on other quantitative criteria (RWA outcomes), which metrics are defined by the bank. Such possibility should be written in regulation.
Question 3. Do you have any comments on the clarifications and revisions made to the qualita-tive criteria for assessing the materiality of extensions and reductions as described in the Annex I, Part I, Section 1 and Annex I, Part I, Section 2?
• The FBF believes that the clarifications and revisions of the qualitative criteria for assessing the relative significance of extensions and reductions constitute an improvement.
However, as mentioned above, changes to exposure classes that do not affect the models should be excluded from the scope of the RTS.
In accordance with paragraph 19 of the explanatory memorandum, the EBA mentions that "As such, in accordance with Article 148(1) of Regulation (EU) 2024/1623, additional exposures that were not risk-weighted by another rating system (i.e. under the standardized approach or by IRB-F if the scope of an LGD model is extended) require in all cases approval by the competent authority and are not within the scope of this RTS".
But Article 148(1) only concerns the approval of the roll-out plans and does not concern the application for approval of a new rating system.
✓ Recital (3) could be misunderstood as now excluding IRB extensions to STD/ IRB-F exposures from the RTS, which were understood as being in the scope of CDR 529/2014.
✓ We propose: roll-out plans (Article 148(1)) may include extending IRBA systems to STD/ IRB-F approaches. When implementing the roll-out plan, such extensions should remain within the scope of delegated regulation 529-2014, allowing assessment of representativeness as well as quantitative criteria to decide for potential ex-ante notification.
✓Automatic material change classification for such extensions would hinder roll-out plan implementation as supervisory expectation from ECB Guide to Internal Models recommends an implementation within a timeframe of 5 years.
Therefore, the RTS should clarify that extending existing IRBA systems to STD/IRB-F can qualify for ex-ante notification, contingent on materiality assessment and representativeness analysis, balancing rigor with efficient IRB roll-out.
Question 4. Do you have any comments on the introduced clarification on the implementation of the quantitative threshold described in Article 4(1)(c)(i) and 4(1)(d)(i)?
• The FBF welcomes the clarification on the implementation of the quantitative threshold. If the interpretation is correct, several different changes should not be assessed individually, such as modifications of a different nature made simultaneously to the same rating system or modifications of the same nature and to the same rating system implemented sequentially over time
• The FBF would like the EBA to clarify the aggregation of RWA impacts for changes affecting multiple rating systems.
With regard to Article 3.3, several cases should be assessed as a single change. The FBF would like to draw attention to certain points: modifications of the same nature and of the same rating system implemented sequentially in time should be grouped together as a single model change. In this case, the FBF understands that it notifies the supervisory authority in advance of a plan of changes (changes that we identify so far) and that it can, in practice, result in a multi-year plan. In this regard, the RTS could specify that it leaves the possibility of introducing a reasonable and limited period for the grouping of changes.
Question 5. Do you have any comments on the revised 15% threshold described in Article 4(1)(d)(ii) related to the materiality of extensions of the range of application of rating systems?
As seen before, we support the EBA's proposed amendments to assess model changes and extensions by combining qualitative and quantitative triggers, with supervisory flexibility to avoid overly mechanical approaches. This is particularly relevant for rating system extensions, given potential issues with the new proposed ratio.
However, the new ratio introduced by the EBA may lead to counter-intuitive results, particularly in the event of simultaneous reductions and extensions. The FBF believes that it is not operational on sub-perimeters or in the event of simultaneous reductions and extensions.
Two scenarios illustrate the issue:
- Definition of the new ratio = (RWEA of B after extension) / (RWEA of A before extension)
RWA Reduction Case 1: If Perimeter A (before): RWEA = 50 and Perimeter B (after): RWEA = 5, then New ratio = 10%
RWA Increase Case 2: If Perimeter A (before): RWEA = 50 and Perimeter B (after): RWEA = 100, then New ratio = 200%
Paradoxically, the ratio suggests more scrutiny for the second case, while the first case represents a more significant change (90% RWA reduction).
Given these challenges, we believe that more flexibility is recommended in assessing materiality, particularly when RWA increases significantly but actual performance (e.g., ranking ability) remains strong.
While waiting for further discussions between French banks and the EBA to define alternatives and to assess their impacts, our preferred option is to maintain the current assumptions and wording of the ratio (status quo).
Question 6. Do you have any comments on the documentation requirement for extensions that require prior notification?
• The assessment report, as listed in Article 8(1) of these RTS, is understood to be a review of the classification of model changes (representativeness) and not a complete model review report by the independent review team. Indeed, the validation report requirement stipulated in Article 8(1) of the RTS presents significant challenges for credit institutions, particularly concerning non-material model extensions. In fact, this approach appears to contradict the EBA handbook where paragraph 96.c explicitly indicates that “for non-material changes to a rating system the review of the changes could be performed during the regular (yearly) validation activities, and the result of the validation function’s assessment would be communicated to the CA via the usual validation report”
. It is therefore preferrable to rely on the framework laid down for the internal independent review.
• The FBF understands that "model performance" is not an anticipatory backtesting exercise and that institutions are not required to submit the results of an initial backtesting exercise when submitting applications for extensions (the first backtesting exercise is carried out after the implementation of the models). The FBF believes that this position will be better understood if it is clearly mentioned in the RTS.