The principle of proportionality is referred to in the text but it suffers from a lack of definition and consequently it cannot be leveraged to provide additional clarity regarding supervisory expectations later in the text. In turn, the CRR states that the regulation should respect the principle of proportionality, in particular regarding the diversity in size and scale of operations and the range of the institution’s activities (e.g. simplest possible rating procedures, proportionality to the nature, scale and complexity of the risks associated with an institution's business model and activities).
The consequences of the application of the proportionality principle with regard to validation processes and number of required risk models and credit segments should also be considered under the light of the proportionality principle and, in particular, how all this will affect smaller institutions.
We propose that the principle is expanded upon, perhaps through a system of categorisation, and these categories can be linked to supervisory requirements elsewhere in the text (e.g. stronger governance requirements can be required of higher category institutions). This will assist in ensuring the appropriate and consistent application of requirements across jurisdictions and between institutions.
It is also important to determine whether the requirements will be applied on a solo basis or a consolidated basis.
The assessment of sequential implementation needs further explanation. For smaller portfolios or to entities we believe that a proportionality principle should be applied with a lighter assessment framework. We understand that roll out plans will have to be submitted for approval. We would like to know how, on which basis and how often this approval will take place.
With regard to the sequential implementation of the IRB Approach, the document sets a period of five years maximum for all exposures and business units. It is not clear if this change applies only to new AIRB applications, or if this change also affects ongoing roll-out plans. It needs to be clarified further how it should apply to existing applications. In the opinion of EBF members, this requirement should not apply retrospectively to the plans that have already been approved.
As it is set out in article 145 of CRR, an institution has to apply the requirements for the use of rating systems during at least three years before it can use the IRB approach for regulatory purposes. We would like to know if this minimum period of three years use test is required when the entity develops new ratings to replace the ones already in use. In case the new rating systems have been implemented for less than three years, we would like to know if it is sufficient to prove that similar systems have been in use for a longer period.
In general the required independence of the validation function makes sense, however it is not clear as to what extent the level of separation of the functions should differ based on the proportionality principle. There is an opportunity here to utilise the proportionality principle but only if clearer indications are provided. The independence of the validation function, including separate reporting lines, may require considerable re-organisation by the bank. It could also lead to staffing issues, both in terms of hiring and retaining qualified staff. Creating a completely separate validation function may cause difficulties such as high turnover of qualified staff as the work can be seen as repetitive. Availability of experienced staff to conduct validation functions is limited and expensive. It goes without saying that such an important function should not be covered by inexperienced staff.
This chapter could include a 'comply or explain' approach, in which an institution that has come to a different solution from the one set out by the supervisors has the opportunity to explain why it has chosen that solution and to convince the supervisors that it is at least as efficient as the solution suggested by the supervisors.
These requirements would not fundamentally impact the validation function. This however depends on the interpretation of the proportionality principle.
Article 11, 3 (b): It is unclear as to whether this requires the validation function to test the deployment or ensure that the appropriate testing has been conducted. It is proposed that the responsibility on the validation function here is that the appropriate testing of the deployment has been conducted but not necessarily that those controls should lie within the validation team.
Article 17(1)(a) (The Role of Internal Audit): There is an apparent overlap in the lines of defence that may put unreasonable pressure on resource requirement, both in terms of numbers and of expertise. The result of the requirements may result in considerable investment of resources with little benefit. It would be proposed that a representative sample of rating systems would be more appropriate, considering the materiality of these rating systems. In particular the requirement that “all” rating systems are reviewed at least annually is particularly onerous. We propose that Article 17(1)(a) is revised to explicitly advocate a risk-based approach commensurate with the role of an internal audit function.
We consider that the requirement stablished in article 49(3)(b) in order to determine the PD when the observed period does not cover a full cycle is against the Through-the-Cycle (TTC) philosophy of the Basel agreement. In the ASRF (Asymptotic Single Risk Factor), on which the Basel framework is based, the PD used must be independent of the Economic Cycle (EC). Contrary to this philosophy, the application of article 49(3)(b) would make the PD dependent on the cycle, because the level of the observed default frequency (ODF) would be conditioned to the period of observed internal data. In our opinion, if the internal data is not representative of a complete economic cycle, the entity has to estimate a PD level not dependent of the cycle, and that estimated PD level could be lower than internal ODF as a result of the fact that the observed period of internal data is downturn.
How should default rates be determined for the artificial period? Default rates will vary depending on the institutions’ risk profile and it is therefore not considered appropriate that institutions apply the same default rates. This needs to be clarified in detail, to avoid situations where the supervisors set a fixed PD for downturn years for all banks, irrespective of their risk profile. A main purpose of internal models is to play a central role in the bank’s risk management by reflecting an accurate relationship between the bank’s risks and its exposures. This will not be the case if the regulators impose unrealistic restrictions on the modelling parameters. The level of default rate for IRB banks should also not result in disproportionate default rates compared to banks operating under the SA method.
Incompleteness of data covering the LRA, and/or one or more ECs, is a common problem for institutions. The requirement, “Application of different reconstruction methods should not lead to less conservative calculation of long-run averages of one-year default rates estimated from the observed data.”, is extremely problematic. Additionally this may act as a barrier to entry and decrease competition within these markets. The estimations may be limited to a timeframe that is not representative of an LRA or EC and therefore produce distortions in RWA outcomes and fail to incentivise institutions from improving their risk management practices.
The laws that require companies to publish their annual statements usually provide some flexibility regarding the exact date of publication, which typically allow for up to 3 months after the end of fiscal year. Thus, the information could not be available to comply with the 12 months requirements set in article 25(2)(b). This fact needs to be addressed accordingly.
Further clarifications would be necessary on the following issues:
The first sentence in Article 49 (3) states that PD estimation by obligor grade should be based on the long-run average (LRA) of one-year default rates. The term “by obligor grade” is not used in the rest of paragraph 3. Instead reference is made to “the default rates in given type of exposures”. In the text for consultation purposes the term “the default rates of the portfolio” is used. Hence, further clarification on the following is needed: Is LRA referring to the overall average for a given type of exposures across obligor grades or to an average default rate within each obligor grade, for a given type of exposures? In other words, is LRA defined across or within obligor grades?
Is point (b) an alternative to point (a) or is it mandatory after the reconstruction period is performed (as stated in point (a))? If it is mandatory, it is not clear if the requirement that “the reconstruction method estimated should not lead to less conservative PD than the ones calculated with the observed data” is applicable after the reconstruction method or, after the reconstruction method plus the conservative adjustments (specified in (b)).
There does not seem to be real value in option (i) of the reconstruction method (a) because the observed data is already representative of a complete economic cycle and consequently, no reconstruction method should be required. If the consultation text is maintained, the estimation of default rates not available for the period will probably lead to the construction of an overall period (observed + not available) different from the observed period, being this last one, by assumption, already a good representation of a complete economic cycle. Additional clarification would be needed as to the denominator of the one-year default rate (article 49(2)(a). We would like to know if it is required to weight the non-defaulted obligors that have no exposure during a period of one year.
Regarding option (ii) of reconstruction model (a), the construction of an artificial period to achieve a complete economic cycle can originate an aggregate lower PD than the one estimated based on the observed data. This possibility conflicts with the requirement set in the proposal that “the estimated PD of the construction method should not be lower than the ones obtained by using the observed data”. We propose the removal of this last statement to ensure consistency throughout the document. If the purpose of the reconstruction model is to effectively capture historic default rates that are higher than those currently experienced, then this could be expressed more clearly in the RTS.
Further guidance on how to assess the representativeness of the historical data period should also be provided in order to avoid different interpretations from local supervisory authorities and to minimise the possibility of adopting an approach that is later found not to abide by the regulatory requirements.
There needs to be a better comprehension of the purpose of the two components of the Expected Loss and Unexpected Loss, namely EAD and the loss % (PD and LGD) for non-defaulted exposures:
EAD defines the quantum of exposure;
PD and LGD defines the riskiness.
We agree that for high-default portfolios in which there is an absence of any EAD concentration risk (granularity risk) that a default-weighted LGD calculation is appropriate and is consistent with the approach to PD.
However, for low default portfolios and or those portfolios containing concentration EAD risk then it is prudent to take into account in some way the riskiness of each exposure.
We recognise that an institution is required to take into account concentration (granularity) risk in its ICAAP and Competent Authorities are required to assess this risk in the Pillar 2 assessment.
In practice, IRB institutions segregate their corporate portfolios by size, such as Large Corporates, Medium Corporates and SMEs and Retail portfolios between mass-market and private banking. Thus the modelling of PD and LGD for these differentiated portfolios takes into account the actual losses and actual LGD.
However, for some institutions where it may be impractical to segregate portfolios, and or where it may have a concentration risk, then it would seem prudent to take into account this risk in Pillar 1 through an LGD estimate that reflects this risk, if an institution chooses to do so.
Therefore to the extent that an institution includes this risk in its estimate of ELGD, we consider that an exposure weighted LGD estimate should be permitted and in turn should have an impact upon a Pillar 2 assessment.
It is important to mention that significant differences are not expected between the two methods (default weighted vs exposure weighted) as the institution follows a cross-default approach on both defaulted customers identification and workout practices. The problem arises in jurisdictions where it is authorised a facility-based default definition for retail or it is required that the LGD calculation follows a facility approach, where the weighting scheme based on number of cases can distort significantly the estimates, lacking the adherence to real losses.
Additional clarity would be welcome around the assignment of the status of no longer in default (cured). It would be helpful if a more specific approach was given around when two default events are mutually exclusive, i.e. when a cure is really a cure so that future defaults are new events.
There needs to be further clarity of the requirements for consistency between any cure definitions used to treat multiple default events, and the definition of defaults and non-defaults for PD and LGD modelling and own funds calculations over time.
When a defaulted obligor is transferred back into a non-default status, it will be because any previous loss (if any) will have been written off and a new credit assessment of the obligor will take place taking into account a restructured balance sheet. That obligor is now to all intents and purposes a new obligor that is not connected to its past history.
Some aspects should be clarified:
Articles 52(a) and 52(c) are mutually contradictory. An exposure is either in a default or non-default state at any given time. Once an appropriate policy and practice has been established to establish this state by an institution then the default count follows. A borrower cannot multiply default during a period in which it is deemed to be in default. If there is a strong (statistical) propensity to re-default subsequent to a cure then the process by which that cure was determined should be examined. It is not appropriate that real defaults should be determined retrospectively. Article 52(c) should be re-phrased to clarify between the first instance of default, and the repeated breaching of default thresholds (i.e. 90 day delinquency) within a cure period.
We would like to know if in the case of an obligor that is in cure status, it would be possible to apply a PD different from an obligor in default (100%) but higher than the PDs applied to exposures in a non-default status, using a specific bucket properly calibrated.
Article 53(b): It is not always required that an institution compute both a long-run average and a downturn estimate to know which is the greater. This requirement should be clarified to recognise this.
Article 54: There is considerable divergence between jurisdictions and institutions regarding the treatment of defaulted assets and the relationship to provisions. The use of two possible approaches questions the objective of harmonisation. It is proposed that one approach should be required and the requirements of the single approach should be elaborated upon in detail.
Article 54(2)(e): The requirement for “all currently available and relevant information” should be clarified so as not to be interpreted to require that all possible information be known concerning a defaulted exposure, only that all practically available and reliable sourced data be considered for use in the evaluation of the exposure.
If the banks should use the definition of multiple defaults as outlined in Article 52, this definition needs to be clarified further to ensure consistent application across banks. This should include, for example, the definition of the length of the cure period during which multiple defaults are counted as one default. It should also define the explicit conditions for defining a facility as cured.
Banks applying AIRB have already implemented own estimates for the collaterals they use, including guarantees. The same types of collateral providers (article 201) should be eligible under all approaches; however, the CRM technique and the effect of the guarantees will of course vary between approaches.
We do not share the opinion that the costs of implementing these measures are negligible or small. The proposed methods for PD and LGD estimations will lead to structural changes in the Financial Institutions models, namely for the ones that have IRB models aligned with current (IAS 39) and future accounting practices (IFRS 9). The following aspects will originate significant costs and could be a step back on the ongoing approval processes with the local supervision authorities:
Costs arising from the transformation of delinquency data sets according to the rules defined for multiple defaults. There is a possibility that this could not be achieved for periods before some cut off dates (due to specificities of IT systems or changes in the data recovery processes within the banks) which will lead to loss of statistical information and major pressure over the 5 year minimum threshold for observable data.
Unclear rules regarding the PD adjustment to reflect a complete economic cycle and the introduction of requirements regarding the reconstruction of periods for models in which the observed data is not representative, will lead to structural changes in the current PD models under the IRB approach, with major consequences across different business lines that depend on these models (e.g. pricing and loan granting).
A default weighted LGD approach will lead to an overall reconstruction of LGD models and to the evaluation of new segments, with consequences on the provision side of the Banks’ P&L which should be, in some extent, aligned with IRB models in order to overcome excessive shortfall deductions on CET1. There is also a significant level of uncertainty regarding the degree of the impact that such changes could originate on the current capital requirements and on the P&L side of the Institutions (due to changes in provisions).
If the time period of a maximum of five years for the sequential implementation of the IRB Approach is changed retroactively to apply to already approved plans, this could have a significant cost.
Additional clarity has been provided in some areas however further work could achieve greater convergence between supervisory practices, with consequent benefits for consistency of outcomes for capital requirements and clarity for implementing institutions.
The intensive supervisory review proposed by the RTS creates a risk that supervisors will pursue a policy of optimality of rating systems. Such a policy may require material changes in many rating systems or even minor changes but with similar costs to institutions.
The requirements of Article 49(3) will require revisions of many rating systems.
The lack of clarity regarding the definition of default and cure (Articles 26 – 29) may also lead to revision of rating systems.
Re-organisation due to changes to the validation function could result in a material change.
It is necessary for a precise assessment of changes that EBA makes clear how to deal will the interactions between this draft and earlier guidelines of national competent authorities.