Response to consultation on Regulatory Technical Standards on assessment methodology for IRB approach

Go back

Question 2: Do you agree with the required independence of the validation function in Article 4(3) and Article 10? How would these requirements influence your validation function and your governance in general?

We understand that the validation function has to be independent from the Credit Risk Control Unit (CRCU) and the staff and management function responsible for originating or renewing exposures.

We strongly approve the required independence of the validation function. This is an essential feature of work efficiency, of guarantee that models are adequately calibrated and of compliance with the regulatory requirements, and of proper management of potential conflicts of interest.

However, we think the wording of the proposition in article 10 could be clarified; it also brings some unnecessary complexity and proposes an organisation scheme for banks that appears overly prescriptive and may not be adapted to all business models

Article 10 seems indeed overly formalistic by only relying on a specific formal organisation. It seems to indicate that only some predefined organisational options would allow the institution to obtain a sufficiently independent validation function. We believe that other adequate governance structures would also achieve the same goal.

Art 10.1 (d) seems also in contradiction with the proportionality principle as it imposes a uniform rule for all global or other systemically important institutions without any distinction between the application of this rule at headquarter level of these institutions and the application of this rule at the level of some small subsidiaries of these institutions.

To summarise, we recommend flexibility and proportionality in order to avoid an inefficient organisational structure where the respect of some formalistic rules would prevail upon the execution of wider and deeper controls.

We realise that a proper equilibrium should be sought between the, in our opinion, legitimate request from the institutions to be allowed to determine their internal organisation themselves on the one hand and the supervisory need to have sufficient assurance regarding the independence of the validation function. We would like to suggest a “comply or explain” approach in relation to an internal organisation that may not completely fit into the one proposed by the EBA.

Therefore, we suggest combining these two visions by adding a new paragraph Art 10.3 or by adding at the end of Art 10.2 the following text:

Organizational options other than those referred to in point (b) to (d) of paragraph 1 can also be accepted provided that:
(i) The institution can prove that all validation function activities are effectively performed, and this in an independent way
(ii) The institution obtains a specific approval for this organisation from its competent authority

Besides, clarifications are needed regarding the role of the CRCU compared to the one of the validation function. Indeed, CRCU should be active in the model validation as per “the areas of responsibility for the Credit Risk Control shall include […] active participation in the design or selection, implementation and validation of models used in the rating process” (article 190 2f of CRR and this draft RTS “Main policy decisions and their rationale” - top of page 10) whereas it is provided that the validation function is in charge of the performance as well as the model validation on an independent basis. We believe some clarity would be useful to the extent to which the reporting lines have to be separated between Credit Risk Control unit and Model Validation unit regarding individual credit risk control and global credit risk monitoring for instance.
Lastly, we also would like to have more confirmation that senior management include CRO or if it will be defined and identified other acceptation within a bank and its subsidiaries in each jurisdiction it may operate in.

Article 4 states that “in the course of the assessment, competent authorities shall verify that the outsourcing by an institution of tasks does not prevent or in other way inhibit the implementation of the methodology referred to in Regulation (EU) n° 575/2013.”

We are wondering what “outsourcing” means. Does that include only external tasks compared to the whole group or is, for instance, a global rating system developed by the parent company (or another Group entity and used in subsidiaries of the Group) considered as an outsourced system from the subsidiary’s point of view?

We understand from EBA Public Hearing on February 9 that separate entities belonging to the same Group would not be considered as external third parties, and hence that modelling or validation tasks performed by another group entity would not be considered as outsourced. We would welcome confirmation of this important point regarding Internal Control requirements.

As a conclusion, these requirements will influence the current validation function and the internal governance in general of almost all institutions. More details should be provided about the Internal Audit role compared to the model validation role. In any case, we think it is important that competent authorities allow for flexibility and take into account institutions constraints in terms of cost/benefits and workplace efficiency in order to assess the implemented structure and its compliance with regulation.

Question 3: Are the provisions introduced in Article 49(3) on the calculation of the long-run average of one-year default rates sufficiently clear? Are there aspects which need to be elaborated further?

Some clarifications would be welcome, especially on the topics below.

o Margin of conservatism (article 47, text for consultation purposes)

The concept of « high estimation error » requiring conservative buffers is too vague. The wording could be reviewed and specify that, once the errors are corrected, the margin of conservatism should be deleted without requiring any authorisation from the competent authorities.

o Method of PD estimation: long run average PD estimation and reconstruction method (article 49)

We support the proposed approach to bring consistency between PD and LGD. However, we have difficulties in understanding the following sentence “some reconstruction methods may be used to account for the missing data, because due to increased uncertainty, additional margin of conservatism should be adopted. In any case the long run average based on the reconstruction method should not be less conservative that the average of one-year default rates estimated from the observed data”.

We think the guidelines are unclear in particular as far as taking into account historical long-run average PD; examples from the EBA would be welcome. The inclusion of a forward-looking perspective is a welcome addition to the framework, and may solve part of the traditional criticisms towards Basel 2 models (which relied too heavily on past data and experience). However we do not understand the rationale behind the requirement that « forward looking » estimates be above Through-The-Cycle estimates: it should be possible for banks adopting a prospective view in order to reconstruct a full economic cycle, to compute and use a PD estimate less conservative than the historical average.

Besides, we think that allowing institutions to use reconstruction methods when they face missing historical data gives more flexibility. We understand that such provision relies on the fact that margin of conservatism should not be used as an alternative to correct the bias of the models on a long term basis.
However, we would like more clarification on the following topics:

• The difference between option a (i) (real historic period is considered representative) and option a (ii) (real historic period is considered not representative). Explanations and illustrations through examples would be appreciated
• The meaning of “estimated” and “average” in option a(ii)

In our view the proposal does not solve the difficulties that institutions face when missing historical data; the EBA proposal introduces a bias by imposing a choice of the reconstruction method for which a deeply documented rationale will be required.

o The alignment of default to accounting definitions also remains an open question.

We understand that the EBA may wish to remain accounting framework neutral, but considering the disparities in local accounting GAAPs (and potentially with IFRS) we would welcome a clarification stating that regulatory default definitions could be aligned with local accounting default definitions.

o Economic cycle

We believe economic factors might not be the only relevant indicators to consider when estimating default rates and that, if there are correlations between economic factors and default rates, these correlations might be different according to business lines or portfolios.
o Stress test

Regarding stress testing and PD models, according to the EBA’s proposals: “the results of the stress tests should be taken into account in the decision making process in the area of risk and capital management processes. In particular the default rates and rating migrations under stress conditions should be taken into account in the assessment of the adequacy of the calculation of the long-run averages of one-year default rates and the dynamics of rating systems”. We think more clarity should be brought on this significant change to the current modelling of default rates, and especially on how it should be embedded in bank daily practices. We understand from the EBA’s public hearing on February 9 that a consolidated approach would be considered sufficient (i.e. top down assessment of the adequacy of the calculations of stressed PDs, as opposed to a bottom up approach where all PDs of all models would have to be stressed individually). We would welcome confirmation of this point.

• The organization of the bank is not necessarily adapted to deal with cycle definition and measurement issues (article 3 art 49 chapter 8)
• The segmentation and granularity of stress testing models do not necessarily match the structure of PD models, thus the required work would seem too regular.

Question 4: Do you agree with the required number of default weighted average LGD calculation method introduced in Article 51(1)(b) and supportive arguments? How will this requirement influence your current LGD calculation method? More generally, what are your views as to balance of arguments for identifying the most appropriate method?

We understand that the LGD should be based on the number of defaults of obligors unless the definition of default is defined at facility level (e.g. for Retail exposures).

As stated by the EBA in the explanatory text, there are arguments supporting both methods of calculating LGD and we believe both should be allowed: number of default weighted average and exposure weighted average. We would like to point out that all LGD models and calculations currently adopted by banks have received the validation of the relevant national competent authority, all are therefore considered pertinent and prudent.
For granular portfolios, exposure weighted average LGD yields more conservative and pertinent results. At a minimum, this method ensures no bias in the expected loss calculation. Furthermore, some institution might use a two steps model for LGD estimation: a cure rate multiplied by a loss given loss. A cure rate can only be computed on an exposure weighted basis. As an illustration, it is this method that has been used for the AQR “challenger model” on retail exposures for collective provisioning. In addition, it is much easier to take into account future drawings in LGD with an exposure weighted methodology when they have not been included in conversion factor.

On the contrary, for less granular portfolios and specifically for low default portfolios, a number of default weighted average might be more pertinent.

If we admit that LGD parameters should be calculated for homogenous pools or facility, using exposure amount as a risk differentiation could lead to non-conservatism behaviour. As for example if it encourages limit increase for risk mitigation.

Upon a closer look at models’ results, the EBA may realise that the largest divergences are due to margin requirements applied by national authorities; these margins are a function NCA’s risk sensitivity and are not necessarily due to calculation method bias.

The EBA’s proposal (number of default weighted LGD calculation) could result in important model reviews and recalibrations. We are wondering how institutions will have to handle the maintenance of the current models that have been given supervisory approval with an exposure weighted LGD. The transition period (including the approval of competent authorities) should be further specified. Institutions require time to implement the default weighted method due, among others, to historical data reconstitution.

Furthermore, the EBA should verify if including the exposure as a driver of the LGD segmentation does not influence the credit origination. As such we think it could be more realistic to introduce flexibility by allowing institutions to use the most appropriate approach.

Lastly, we are unsure about the notion of LGD downturn forward looking. We suggest dealing with this topic as part of the forthcoming EBA consultation on LGD downturn.

Question 5: Are the provisions introduced in Article 52 on the treatment of multiple defaults sufficiently clear? Are there aspects which need to be elaborated further?

Multiple default requirements are not clear enough. We understand that one of the objectives is to increase consistency among risk parameters (PD, LGD and CCF) and that institutions will have to define and set their own length of the cure period. However, we need more information on the situation where multiple defaults are spread over various estimation periods, as calculations seem difficult to implement; examples should be provided taking into account different situations (for instance a situation where an obligor defaults before December 31 (reference date), then returns to performing status before this reference date and then defaults again shortly after it). Calculations are especially difficult when estimating LGD while merging different time periods during which the obligor is in default; this implies taking into account flows during obligor performing periods as well as costs allocation rules.

Moreover, requirements linked to the default definition are spread over several references (CP 32, Non Performing Exposures /Forborne Exposures, forthcoming EBA guidelines, etc.). As such, it is hard to have a global view on the definition of default / multiple defaults. This situation is all the more uncomfortable as default is a key component of the IRB approach: any change or evolution in this definition implies heavy work, impact on RWAs, and material changes that need Competent Authorities approval. Hence it is important that regulators seek a coordinated approach on this topic. Therefore, we would really appreciate that the forthcoming EBA guidelines on the definition of default includes all relevant information.

Regarding the cure period definition, it should be clarified if the cure period after default is a period during which the counterpart is marked as “in bonis” in the system of the financial institution, but should a new default occur during this period, it should be combined to the first default for risk parameters estimation purposes. Thereafter a clearer distinction between the exit default criteria and the cure period definition could be made (i.e. explicit reference to chapter 6 in the text). Should a slight difference exist between operational default marking process, and the modelled default (cases when modelling criteria are not exactly in line with operational criteria), we think it should not be interpreted as a weakness in the use test principle by supervisors.

For non-retail exposures, banks commonly use an observation period after default during which the obligor’s situation is frequently reassessed: this process is commonly part of internal procedures. The cure period shall then be defined in accordance to corresponding risk management processes of the institution (which do not necessary cover the same length of time for each obligor), and the “short period of time” may be linked to the duration of the internal process instead of a fixed number of months (assuming of course that the internal process covers a sufficient period of time).

It could also be clarified if the cure period should be necessarily defined as the same for the institution or if the analysis of the “default experience” could lead to different period of time or definitions by transaction or types of counterparties. The content of the “analysis of the default” experience could also be more explicit and some examples of good practices would be much appreciated: such as transitions matrices from first default to a new default status or other specific studies.

For distressed restructuring, institutions do frequently have exit criteria from default that encompass a long period of time and an additional cure period during which the default criteria could be weaker than the initial criteria. However the text does not specify the need to take into account in the LGD estimation the effect of restructuring, in case no default occurs after the restructuring plan is put in place. We hope this issue will be addressed in the forthcoming EBA guidelines on default.

The effect of combining the defaults occurring during the cure period to the first default can also be specified for LGD estimation as taking the EAD of the first default and considering the repayment inflows since the first default. Such a practice could decrease the LGD estimate compared to an estimate relying on the exposure of the last default, and will impose keeping records of all the inflows after the first default even in case of a return to “in bonis” status. Historically some institutions have taken the last default in order to have a prudent estimation of the LGD parameter and such an approach has been validated by their supervisors.

Question 6: Are the provisions introduced in Article 60 on the treatment of eligible guarantors for the purpose of own-LGD estimates sufficiently clear? Are there aspects which need to be elaborated further?

The provisions of article 60 on the treatment of eligible guarantors for the purpose of own-LGD estimates are sufficiently clear. However, in the “text for consultation purposes” the sentence below introduces much confusion “If it is assumed that under article 183(4) CRR, the effect of a funded guarantee […], the seeming contradiction can be dispelled”.

We would like to ask for further clarification and examples to illustrate this requirement.

Question 7: Do you support the view that costs for institutions arising from the implementation of these draft RTS are expected to be negligible or small? If not, could you please indicate the main sources of costs?

Huge costs will arise from the implementation of the RTS proposed in this Consultative Paper. The estimation provided by the EBA of one-off costs associated with these requirements is unrealistic.

The main sources of costs are due to:

• Taking into account new additional regulatory requirements as well as adjustments or clarifications provided in the text (chapter 8 articles 28, 51, 52, 54, and 57). These articles have major impacts on risk parameters, including :
o Recalibration of risk parameters
o Reconstitution of historical data
o Performance of internal impact studies
o Upgrades or changes in the information systems
o Implementation of new rating systems and impacts on the use test
o Staff training, among others credit officers and users due to new rating systems
• Data maintenance requirements (chapter 12)
o Ongoing adaptations of the IT systems and improvements of the data quality processes
• Amendments and implementation of new processes such as :
o Roll out plan regular reviews and updates (chapter 2 article 7s)
o Map of rating systems construction, review and updates (chapter 7 article 33)
• Administrative costs (chapter 1 to 14)
o Updates, writing of documentation (processes, procedures…)
• Organisation (chapter 3)
o Adaptations and/or changes of the internal governance

Given the number of forthcoming upgrades, consistency is crucial in order to avoid back and forth changes in the rating framework. Implementation costs may be further increased by the alignment of IRB models to future accounting practices (IFRS 9).

Question 8: What are the main benefits for institutions that you expect by the adoption of these draft RTS?

We support the initiatives aiming at a better comparability of internal modelling approaches between banks and jurisdictions. Clarification of the underlying rules in parameter computations is desirable, and will contribute to improving the transparency and level-playing field. Improvements will be all the more significant as a strong convergence between supervisors will be achieved (hence, as stated in the introduction, we emphasise again that EBA works on these issues should be aligned with the Basel Committee’s current review of Basel 2).

Additional benefits institutions expect from the adoption of this Consultation Paper are the following ones:
• By explaining the area of controls that Competent Authorities will perform, it gives a global and updated view of all the regulatory requirements that institutions are expected to meet;
• It will enable harmonisation of the supervisory assessment methodology across all EU members and therefore reduce or minimise arbitrage;

These benefits will apply provided that assessments and permissions by national competent in the EU ensure an adequate level playing field at all times.

Question 9: Do you expect that these draft RTS will trigger material changes to the rating systems (subject of the RTS on materiality of model changes)? If yes, could you please indicate the main sources of the changes (please list the relevant Articles of these draft RTS)?

Some requirements will trigger changes to rating systems calling for Competent Authorities approval. They mainly concern:

o Changes in the rating methodology for IRB systems
• The most important change refers to the principle introduced in articles 51, 54 and 57 of the draft RTS EBA/CP/2014/36.
o Changes in the definition of default according to Regulation (EU) n°575/2013 Article 178. They relate to :
• The introduction of a materiality threshold (article 28 d of Consultative Paper EBA/CP/2014/36)
• The implementation of the multiple defaults requirements (article 52 of Consultative Paper EBA/CP/2014/36).

We would like to underline that an over restrictive limit of human judgment would be detrimental to the forward looking aspect of the rating process and could jeopardize the link between the actual perception of the risk and the risk parameter.

Taking the above into account will require heavy works such as reconstitution of historical series, recalibration of risk parameters and models across a firm as well as huge evolutions in the information systems. It may potentially trigger numerous material change requests per institution that will need to be handled in due course by the ECB for euro area banks.

These changes will be even more burdensome for banks if they had to be applied retrospectively. Therefore we would like to request flexibility when implementing this RTS by allowing institutions to build historical series as they become available and to gradually include them in their risk parameter estimations. Being fully compliant with these new requirements will take time, banks are conscious that this approach could introduce bias in the parameters estimation but it is a key factor in building long run reliable and complete series of historical data.

Upload files

Name of organisation

French Banking Federation