Response to discussion Paper on the future of the IRB Approach
Go back
Within every separate asset class, the order of major topics proposed by the EBA seems appropriate. First priority refers to default which is a central factor in models and therefore it should be in the focus of harmonisation. Then risk parameters with LGD before Probability of Default (PD) because more new requirements are expected in terms of LGD. Lastly Credit Risk Mitigation (CRM) because the changes presented by the EBA refer to minor areas and not the whole CRM framework. We explain our rationale here below.
Based on the subjectivity that lies on the Capital Requirements Regulation (CRR) and the necessity for a clear and practical guidance to adopt the regulation, the following topics should be given priority:
Default definition: the subjectivity in paragraph 3 of article 178 of the CRR requires additional clarifications that, when put in place, will demand significant changes in the banks’ internal models. The transformation of historical data sets to address changes in the default definition will require a significant amount of time for implementation and 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. As such, the clarification of the referred paragraph should be one of the first steps due to its implications on the overall IRB process. After all, the default definition is the main pillar of the IRB framework.
PD estimation: the requirement for PD to encompass the whole economic cycle will require further clarifications, as mentioned in the discussion paper. Regardless of such additional guidance, the banks will always have to create new processes for post default rate calculation (which will be based on the historical information in the IT systems) to address the whole economic cycle requirement. The design and implementation of reconstruction models, on top of the observed default rate calculation process based on IT procedures, will require time, not only at the starting moment, but also on a periodic basis, namely on each calibration process.
Clarification regarding the use, or absence of use, of a master scale by banks should also be considered. Expectations regarding their construction and use (e.g. granularity, single PD within grade, etc.) could be considered and clarified.
A review and status of national discretions should be put on the agenda. These may not automatically be removed when new standards are introduced. If an articulation between the EBA efforts and the tasks that are being performed on the national discretions field is not pursued, the current heterogeneity observed in Europe will continue and the doubts underlying the IRB approach will persist.
If changes are kept within a reasonable limit and the EBA avoids burdensome revisions, the timetable could be feasible.
However, we would like to draw the attention of the EBA to the phases that model changes entail:
Firstly, the design and development of changes. This phase is intensive in IT resources among others, as well as project framework, models, data collection, tests, impact studies and documentation. The EBA advice should take into account the approximate cost and time involved. We believe that 2018 is too close in order to complete the full scope of regulatory initiatives to be undertaken including validation by supervisors.
Secondly, the parallel running that banks must undertake and duly coordinate. The nature and depth of the changes require a cautious process of implementation.
Finally, the supervisory approval, which can partially overlap with the parallel running though. It is this phase that banks cannot control. In the experience of banks it can take longer than what the EBA schedule appears to envisage. The problem is compounded by large number of model changes that will require supervisory approval in a short time.
In view of the breadth of models and potential changes we would need between 3 and 5 years of implementation, depending on the required changes and the number of changes to be made at the same time and number of models to be impacted, plus the time needed for supervisory authorisation that must be handled in 6 months.
Harmonisation of the “non-accrued status” of a credit obligation, including the impact on different accounting procedures across the EU banks;
The definition of a “significant perceived decline in credit quality” should be clarified;
EBA comment regarding the fact that clarifications of “multiple defaults” and “return to a non-default status” have been addressed on the EBA consultation on IRB requirements, already published, rise major concerns. This consultation text would benefit from further transparency and practical guidance regarding the following issues:
A more specific approach to identify a cure so that future defaults are effective new events;
A more transparent approach to count the default;
Clear clarification regarding the concepts of “cure” and “non-default status”. If they are not the same, it is important to be more prescriptive regarding the number of months that should be observed to transfer a credit obligation from a cured status to a non-default status and to assign the following periods to each concept:
o Time between the first trigger of default and the ceasing of existence of any active default trigger;
o Time between the ceasing of existence of any active default trigger and the moment in which the obligation is in a “non-default status”.
Further guidance regarding the PD to be used in each status: Which objective conditions must be met to transfer a credit obligation from a cure to a non-default status? Should this be dependent on historical PD for loans in similar situations? If yes, what should be the minimum PD values observed (on a comparative basis with default and non-default obligations) to be able to perform that transfer?
We would also like to remind of the importance of expert judgment specifically when dealing with non-retail counterparties. A deep knowledge of the portfolio in line with internal risk monitoring policies and economic reality should be factored in. Too stringent harmonisation could impede sound practices and the use test.
Consideration should also be provided to acceptance criteria for a divergence between a definition of default used for modelling purposes and that used in live ‘production’ systems. Modelling challenges may require use of a divergent approach that is nonetheless compatible with regulatory objectives (i.e. is conservative) and such flexibility, subject to a comply-or-explain approach, would be welcomed.
The common problems faced by banks when adjusting historical data were:
Incompatibility of IT systems and other technical problems notably in the numerous banks that have merged portfolios and whole entities during the historical period which data is to be rebuilt. In many cases, data simply was not available;
Implementation of new fields in databases is a time-consuming and difficult task involving huge amount of manual work where data cannot be inferred automatically.
These difficulties can erode the quality of risk parameter estimation and hence the comparability of internal model results among institutions. We therefore strongly suggest to limit requests for adjustment of historical data. In case of doubt, it is better to design a reasonable timeframe for the build-up of data year after year.
Going backwards more than 2 years would be all the more complicated. Banks should be given the possibility of building up new data series in the years after implementation.
Multiple defaults: material changes. The major difficulties we identified so far relate to:
o The implicit impact of this new requirement that refers to the automatic decision of the return to non-default status. This requirement is very difficult to implement because it does not fit in the institution’s internal process and counterparty monitoring. Expert judgment may be the key reference to decide whether an obligor is again performing or not and this situation cannot be resembled retrospectively. The multiple defaults concept seems to be more appropriate for retail exposures.
o The identification of default: the reconstitution of the default history to clearly identify the number of default to consider at a specific point in time.
o The LGD estimation: the collection of data that refer to the same default, i.e. losses and recoveries (cash flows and potential guarantees or collateral enforcement).
o The implementation of new parameters (length of the cure period for example) into information system and rating models.
PD estimation: planned guidelines will introduce material changes (further alignment of the long run average of one-year default rate with definition of economic cycle, identification of stressed years, how to cope with the absence of the time series of adequate stress conditions to capture a downturn and combination of different data sources). We need clarification of the following sentences “observed differences in the number of rating grades and/or use of continuous rating scales should encourage application of harmonised rules”.
LGD estimation: We understand that a complete review of LGD and Credit Conversion Factor (CCF) parameters is under way. Concerns are:
o The LGD and CCF should reflect the downturn conditions if they are more conservative than the long run average (this often requires institutions to estimate 2 parameters). Further specification will be done on the nature, duration and severity of economic downturn, length of data series, treatment of incomplete workout, collateral and guarantees, margin of conservatism and costs. Plus coordination with the EBA.
The dimensions of the intended changes to the parameters described in section 4.3.2 are not fully clear. There seems to be proposed fundamentally changes, by indicating supervisory methods compared to current institutions specific models and/or parameters (default rate, LGD and PD).
As a conclusion, it is crucial that EBA and BCBS works are well coordinated in order to avoid to multiply changes and to optimise institutions workload and costs. This will help to enhance the sound harmonisation EBA is expecting.
We have difficulties to estimate the materiality of changes in our IRB models because most of them are not at that time clearly identified. Somme of the changes might entail an entire review of existing models. A good trade-off has to be found between the need for harmonisation and comparison and the sustaining of already approved models.
In any case, there seems to be fundamental changes that should be developed and coordinated with the Basel Committee. It should be reiterated that this requirement may have significant distortionary and inappropriate macroeconomic effects in jurisdictions which have experienced the most severe economic stresses in the recent period and it should be revisited.
Despite the fact that some detailed comments will be provided during the consultation period of each regulatory product, from a more high level perspective, the requirement that “the estimated PD of the construction method should not be lower than the ones obtained by using the observed data“ does not seem appropriate because 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 (if the delinquency behaviour of the observed data represents the downturn period of the complete economic cycle).
Cross default: The cross default issue is of outmost importance, namely for corporate LGD purposes, for which different procedures exist across EU Banks. The major differences and problems arise in the jurisdictions where it is authorized 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. As such, the choice between a facility approach versus client approach could lead to a significant transformation of recovery data sets. Furthermore, the tracking of the relationship between numerous facilities of the same obligor require time and it is complex in nature, reason why it should not be left aside from the EBA guidelines.
The denominator of the one-year default rate. As stated in paragraph 50, it is currently proposed that the denominator includes the exposures which are not in default at the beginning of one year period (“time 0”). In that sense, we would like to know if it is required to weight the non-default obligors with whom the bank has no exposure during the period of one year after time 0.
We think that the changes in the definition and estimation of the parameters of the models should not be considered as material in the sense of RTS on assessment methodology by which competent authorities will assess in detail the models used for the purpose of the IRB approach and verify how risk parameters are estimated, since they stem from a decision of the regulatory authority. At the minimum, we advocate for a more rationalised application of this RTS during the transition period which could also be more flexible.
Back-testing procedures: The main objective of back-testing is to validate if the time series used to compute risk parameters are in line with the most recent observation periods. If this result in a statistical difference between the risk parameters of the model and the risk parameters based on more recent time series, it should originate a review of the model risk parameters. These assumptions/ procedures could lead to contradictory procedures and results when compared to the ones obtained through a TTC approach. In this sense, further clarification how to design, conceptualize and implement the relationship between the back-testing (precise rules and objectives) and the risk parameters measurement (economic cycle, PIT vs. TTC, among others) should be provided and should be a priority as they affect all the remaining aspects of section 4.3.2. The link between back-testing and the stability of parameter estimates should especially be addressed. In particular, some guidelines are welcome on how many back-testing failures are acceptable before the model needs a complete review. In this context, a minimal required level of model’s explanatory / discriminatory power may be a useful benchmark. Finally, the practical meaning of “the expected range of estimation errors” used to define the confidence intervals for parameter estimates in back-testing needs to be explained, and quantified in statistical terms to make it less subjective.
Separation between defaulted and non-defaulted exposures;
LGD in default should include the information on the time in-default and recoveries realised so far;
Each difference between Expected Loss Best Estimate (ELBE) and specific risk adjustment should be duly justified.
These changes impact modelling practices and information system (data to store and to introduce in information system).
Different practices in the estimation of LGD in-default;
Different practices in the estimation of ELBE;
Difference in the level of calculation of the IRB shortfall.
We think that the provision set in the text will enhance consistency as far as the level of calculation of the IRB Shortfall is concerned. We understand that it has to be calculated at an aggregate level separately for the portfolio of defaulted exposures and the portfolio of exposures that are not in default.
It is hard to have an opinion on the two other sources of discrepancies because the RTS does not contain much information on it. Requirements on LGD in default will be more precisely set in the planned guidelines through the definition of downturn LGD. However, there is at that stage no specific indication on the estimation of either the ELBE for defaulted exposures or the add-on that captures the unexpected loss that might occur during the recovery period.
For the sake of clarity, the first bullet point of paragraph 68, “direct estimation of downturn LGD and ELBE for defaulted exposures” should be replaced by “direct estimation of LGD in-default and ELBE for defaulted exposures”, as the downturn LGD is a concept applied to non-defaulted exposures, being LGD in-default the correct concept to capture the LGD for defaulted assets in a downturn/unexpected loss perspective.
We also recommend the alignment of default definition in all regulations within EU by the regulator (e.g. IFRS, Basel II, EBA ITS). Currently EBA ITS requires to disclose non-performing exposure which are classified as performing under Basel, e.g. loans with 2 forbearance measures which were never 90 days or more overdue, no default rating etc. To reduce complexity, internally and also externally for analysts, it would be appreciated if the definition in the various regulations would be aligned, at least within EU.
It is important to clarify whether the application of this new treatment proposed for the PPU and roll-out consideration will have retrospective effects or not.
However, to achieve such a situation, models should not be subject to numerous material changes over a limited time period, facing the risk of losing internal and external credibility.
The proportionality principle requires further and practical considerations in the EBA paper to understand, in an objective way, to which institutions it should be applied and, in those cases, what are the aspects that could be adjusted or reduced regarding the internal risk management process (process validation, number of risk models and segments, number of entities, timelines for implementation, etc.).
Most institutions organisational structure will be impacted following the requirements that relate to internal governance. Stress testing processes might also evolve to consider IRB stress tests as well as Pillar 2 stress tests. A wider use of risk parameters might lead to some changes in internal risk management processes.
It has been the experience of banks that application procedures and regulatory response to applications for model approvals varies widely. This too is a source of divergence in risk estimates between banks. Convergence towards a timely and efficient regulatory approval capability is to be encouraged, preferably with time commitments from the competent authorities. Untimely approval undermines the constant investment made by banks in their risk measurement systems and in model use and risk management.
We disagree with the publication of benchmarking exercise results whether public disclosure or on the EBA website for the following reasons:
The requirements defined in the final version of the BCBS disclosures already include numerous templates that provide a wide range of meaningful information to external stakeholders;
These exercises rely on hypothetical portfolios that have nothing to do with institutions’ real business;
They mainly help competent authorities or the EBA at understanding models and comparing them for regulatory purposes;
Publishing such information will increase complexity of the disclosed information and spread confusion in external stakeholders’ mind and therefore entailing misleading interpretation and analysis;
Difficult to establish a link with Pillar 2 disclosure. May raise more questions that understanding institutions performance.
We also remind that we strongly disagree with the BCBS project requiring banks that are using the IRB approach to disclose hypothetical capital requirement according to the standardised approach for credit risk.
This approach is simplistic and will again create confusion in external stakeholders’ mind and lead to irrelevant questioning and complex debates which would be difficult to solve without the disclosure of proprietary or confidential information.
Disclosure is a very important topic and we understand the need of external stakeholders to understand institutions finance performance. We also agree that adequate disclosure will enhance transparency and reinforce internal model credibility.
We have the following concern as far as the additional ad hoc disclosure is concerned:
• We question the appropriateness of ad hoc disclosures on the EBA website. As mentioned above, we think that these requirements will increase the number of disclosed information and may introduce confusion in stakeholders mind. Moreover, we wonder whether the EBA does not go beyond its mandate by so doing.
• We think that for this ad hoc disclosure to be relevant, it has to be consistent with BCBS requirement. There must be a bridge between disclosed data (i.e. stakeholders shall be able to switch from one set of information to the other one without having to reconcile the information).
• There is no information on the frequency but we suggest it will follow the BCBS defined frequency.
• The disclosed data should really add value to the regulatory disclosure. Such high granularity is most often totally useless to most market users, but it provides valuable information on market shares and customers to competitors (notably to non-EU banks).
Apart from this, we support the EBA proposal to issue guidelines for reporting ad hoc information to be harmonised and therefore avoiding any discrepancies.
Institutions shall publish information consistent with the BCBS disclosure requirements as at January 2015 from end 2016. This ad hoc disclosure should not be made mandatory before the enforcement of BSBS disclosure and the implementation of all new regulatory requirements. One should not assume it represents a limited burden on institutions which have already to implement an unprecedented number of regulatory changes and of disclosure requirements in a very short timeframe. In addition, discrepancies may emerge from change in regulatory requirements as far as models are concerned.
We are also concerned that the EBA's project to anticipate this BCBS disclosure for EU banks creates once again an unlevelled playing field with non-EU banks.
We would like to insist in the idea that, in our opinion, a simplification in the reporting framework will be a positive move in terms of efficiency and administrative and organisational costs for entities and ultimately for supervisors.
Banks acknowledge the scarcity of data to fill up the supporting databases for low default portfolio (LDP) models. We note that EBA evokes in the Discussion Paper the possibility of applying the Standardized Approach for LDP. But shifting LDP models toward the Standardized Approach will increase the capital charge with no apparent economic reason. Therefore, we recommend the EBA to study the possibility of an intermediate model.
Consideration and clarity regarding the use of dual TTC and PIT PD approaches should be given within the PD estimation product.
The margin of conservatism must not be considered as a separate measure to increase the own funds. The margin of conservatism should be related to weaknesses in the models. It should not be possible to impose large add-ons to models. The design of the models must be respected, and must not be subject to add-ons based on unquantifiable supervisory judgements. There should be an upper limit related to e.g. a percentage of the internal estimates that could be added to the internal models.
Clarification of Pillar 1 IRB Stress Testing requirements and the relationship and delineation between this and Pillar 2 Stress Testing requirements.
2. What would you consider the areas of priorities?
We think that it makes more sense to break down the revision by asset class, because within institutions there are different model development standards for each asset class and different model development and review functions.Within every separate asset class, the order of major topics proposed by the EBA seems appropriate. First priority refers to default which is a central factor in models and therefore it should be in the focus of harmonisation. Then risk parameters with LGD before Probability of Default (PD) because more new requirements are expected in terms of LGD. Lastly Credit Risk Mitigation (CRM) because the changes presented by the EBA refer to minor areas and not the whole CRM framework. We explain our rationale here below.
Based on the subjectivity that lies on the Capital Requirements Regulation (CRR) and the necessity for a clear and practical guidance to adopt the regulation, the following topics should be given priority:
Default definition: the subjectivity in paragraph 3 of article 178 of the CRR requires additional clarifications that, when put in place, will demand significant changes in the banks’ internal models. The transformation of historical data sets to address changes in the default definition will require a significant amount of time for implementation and 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. As such, the clarification of the referred paragraph should be one of the first steps due to its implications on the overall IRB process. After all, the default definition is the main pillar of the IRB framework.
PD estimation: the requirement for PD to encompass the whole economic cycle will require further clarifications, as mentioned in the discussion paper. Regardless of such additional guidance, the banks will always have to create new processes for post default rate calculation (which will be based on the historical information in the IT systems) to address the whole economic cycle requirement. The design and implementation of reconstruction models, on top of the observed default rate calculation process based on IT procedures, will require time, not only at the starting moment, but also on a periodic basis, namely on each calibration process.
Clarification regarding the use, or absence of use, of a master scale by banks should also be considered. Expectations regarding their construction and use (e.g. granularity, single PD within grade, etc.) could be considered and clarified.
A review and status of national discretions should be put on the agenda. These may not automatically be removed when new standards are introduced. If an articulation between the EBA efforts and the tasks that are being performed on the national discretions field is not pursued, the current heterogeneity observed in Europe will continue and the doubts underlying the IRB approach will persist.
3. Do you consider the proposed timeframe reasonable? In particular do you consider reasonable the proposed timeline for the implementation of the changes in the area of: a. definition of default; b. LGD and conversion factor estimation; c. PD estimation; d. treatment of defaulted assets; e. CRM?
An informed decision about this question can only be provided after the specific and detailed changes that might result from the entry into force of the named regulatory products become known.If changes are kept within a reasonable limit and the EBA avoids burdensome revisions, the timetable could be feasible.
However, we would like to draw the attention of the EBA to the phases that model changes entail:
Firstly, the design and development of changes. This phase is intensive in IT resources among others, as well as project framework, models, data collection, tests, impact studies and documentation. The EBA advice should take into account the approximate cost and time involved. We believe that 2018 is too close in order to complete the full scope of regulatory initiatives to be undertaken including validation by supervisors.
Secondly, the parallel running that banks must undertake and duly coordinate. The nature and depth of the changes require a cautious process of implementation.
Finally, the supervisory approval, which can partially overlap with the parallel running though. It is this phase that banks cannot control. In the experience of banks it can take longer than what the EBA schedule appears to envisage. The problem is compounded by large number of model changes that will require supervisory approval in a short time.
In view of the breadth of models and potential changes we would need between 3 and 5 years of implementation, depending on the required changes and the number of changes to be made at the same time and number of models to be impacted, plus the time needed for supervisory authorisation that must be handled in 6 months.
4. Are there any other aspects related with the application of the definition of default that should be clarified in the GL?
In our opinion, the qualitative indicators specified in the discussion paper cover the main aspects of the definition of default. However, the guidelines should provide practical guidance for recognition of those indicators in order to mitigate the current subjectivity. Besides the additional clarifications regarding restructured loans (paragraph 3d of article 178), stated by EBA, as the main objective of the future guidelines in this matter, the following situations (paragraph 3 of article 178) should also be subject to review, and a more practical guidance should be provided: Harmonisation of the “non-accrued status” of a credit obligation, including the impact on different accounting procedures across the EU banks;
The definition of a “significant perceived decline in credit quality” should be clarified;
EBA comment regarding the fact that clarifications of “multiple defaults” and “return to a non-default status” have been addressed on the EBA consultation on IRB requirements, already published, rise major concerns. This consultation text would benefit from further transparency and practical guidance regarding the following issues:
A more specific approach to identify a cure so that future defaults are effective new events;
A more transparent approach to count the default;
Clear clarification regarding the concepts of “cure” and “non-default status”. If they are not the same, it is important to be more prescriptive regarding the number of months that should be observed to transfer a credit obligation from a cured status to a non-default status and to assign the following periods to each concept:
o Time between the first trigger of default and the ceasing of existence of any active default trigger;
o Time between the ceasing of existence of any active default trigger and the moment in which the obligation is in a “non-default status”.
Further guidance regarding the PD to be used in each status: Which objective conditions must be met to transfer a credit obligation from a cure to a non-default status? Should this be dependent on historical PD for loans in similar situations? If yes, what should be the minimum PD values observed (on a comparative basis with default and non-default obligations) to be able to perform that transfer?
We would also like to remind of the importance of expert judgment specifically when dealing with non-retail counterparties. A deep knowledge of the portfolio in line with internal risk monitoring policies and economic reality should be factored in. Too stringent harmonisation could impede sound practices and the use test.
Consideration should also be provided to acceptance criteria for a divergence between a definition of default used for modelling purposes and that used in live ‘production’ systems. Modelling challenges may require use of a divergent approach that is nonetheless compatible with regulatory objectives (i.e. is conservative) and such flexibility, subject to a comply-or-explain approach, would be welcomed.
5. Do you have experience with adjustments of historical data? What are the methods that you used to adjust historical data, including both internal and external data?
The broadest experience in adjusting historical data was in the preparation of the IRB models for the first authorisation at the onset of the Basel II standard. Back in 2005, banks had to retrieve new data that was needed as an input to the new models. In some cases, banks had to build up time series using proxies. Manual processes consumed huge resources. In particular, the databases required for the estimation of the LGD requested much information that was not typically recorded in the banks IT systems.The common problems faced by banks when adjusting historical data were:
Incompatibility of IT systems and other technical problems notably in the numerous banks that have merged portfolios and whole entities during the historical period which data is to be rebuilt. In many cases, data simply was not available;
Implementation of new fields in databases is a time-consuming and difficult task involving huge amount of manual work where data cannot be inferred automatically.
These difficulties can erode the quality of risk parameter estimation and hence the comparability of internal model results among institutions. We therefore strongly suggest to limit requests for adjustment of historical data. In case of doubt, it is better to design a reasonable timeframe for the build-up of data year after year.
6. To what extent is it possible to adjust your historical data to the proposed concept of materiality threshold for the purpose of calibration of risk estimates?
It seems to be unfeasible to rebuild historical data to apply the threshold retroactively. It entails heavy workload: data collection, change management, changes in parameters and rating systems as well as IT development.Going backwards more than 2 years would be all the more complicated. Banks should be given the possibility of building up new data series in the years after implementation.
7. What is the expected materiality of the changes in your IRB models that will result from the proposed clarifications as described in section 4.3.2?
The EBA programme in this Discussion Paper deals with 4 topics: multiple defaults, default rate, PD estimation and LGD estimation. We understand that requirements are almost set as far as default rate and multiple defaults are concerned. Multiple defaults: material changes. The major difficulties we identified so far relate to:
o The implicit impact of this new requirement that refers to the automatic decision of the return to non-default status. This requirement is very difficult to implement because it does not fit in the institution’s internal process and counterparty monitoring. Expert judgment may be the key reference to decide whether an obligor is again performing or not and this situation cannot be resembled retrospectively. The multiple defaults concept seems to be more appropriate for retail exposures.
o The identification of default: the reconstitution of the default history to clearly identify the number of default to consider at a specific point in time.
o The LGD estimation: the collection of data that refer to the same default, i.e. losses and recoveries (cash flows and potential guarantees or collateral enforcement).
o The implementation of new parameters (length of the cure period for example) into information system and rating models.
PD estimation: planned guidelines will introduce material changes (further alignment of the long run average of one-year default rate with definition of economic cycle, identification of stressed years, how to cope with the absence of the time series of adequate stress conditions to capture a downturn and combination of different data sources). We need clarification of the following sentences “observed differences in the number of rating grades and/or use of continuous rating scales should encourage application of harmonised rules”.
LGD estimation: We understand that a complete review of LGD and Credit Conversion Factor (CCF) parameters is under way. Concerns are:
o The LGD and CCF should reflect the downturn conditions if they are more conservative than the long run average (this often requires institutions to estimate 2 parameters). Further specification will be done on the nature, duration and severity of economic downturn, length of data series, treatment of incomplete workout, collateral and guarantees, margin of conservatism and costs. Plus coordination with the EBA.
The dimensions of the intended changes to the parameters described in section 4.3.2 are not fully clear. There seems to be proposed fundamentally changes, by indicating supervisory methods compared to current institutions specific models and/or parameters (default rate, LGD and PD).
As a conclusion, it is crucial that EBA and BCBS works are well coordinated in order to avoid to multiply changes and to optimise institutions workload and costs. This will help to enhance the sound harmonisation EBA is expecting.
We have difficulties to estimate the materiality of changes in our IRB models because most of them are not at that time clearly identified. Somme of the changes might entail an entire review of existing models. A good trade-off has to be found between the need for harmonisation and comparison and the sustaining of already approved models.
In any case, there seems to be fundamental changes that should be developed and coordinated with the Basel Committee. It should be reiterated that this requirement may have significant distortionary and inappropriate macroeconomic effects in jurisdictions which have experienced the most severe economic stresses in the recent period and it should be revisited.
8. Do you consider the direction of the proposed changes adequate to address the weaknesses and divergences in the models across institutions?
We agree that the differences in the risk estimates across EU Banks should reflect the differences in risk profiles of the institutions rather than the different interpretation of the IRB requirements. Following the indications from EBA in sections 4.3.2 it seems to be that EBA indicates a direction that goes against the institutions specific risk estimates. As an example, a national average default rate would not represent the actual default rate for an institution or the respective portfolios in a bank. This goes against the intention with the IRB framework.Despite the fact that some detailed comments will be provided during the consultation period of each regulatory product, from a more high level perspective, the requirement that “the estimated PD of the construction method should not be lower than the ones obtained by using the observed data“ does not seem appropriate because 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 (if the delinquency behaviour of the observed data represents the downturn period of the complete economic cycle).
9. Are there any other aspects related with the estimation of risk parameters that should be clarified in the EBA guidelines?
There are additional aspects that require further clarification in the EBA guidelines. For example: Cross default: The cross default issue is of outmost importance, namely for corporate LGD purposes, for which different procedures exist across EU Banks. The major differences and problems arise in the jurisdictions where it is authorized 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. As such, the choice between a facility approach versus client approach could lead to a significant transformation of recovery data sets. Furthermore, the tracking of the relationship between numerous facilities of the same obligor require time and it is complex in nature, reason why it should not be left aside from the EBA guidelines.
The denominator of the one-year default rate. As stated in paragraph 50, it is currently proposed that the denominator includes the exposures which are not in default at the beginning of one year period (“time 0”). In that sense, we would like to know if it is required to weight the non-default obligors with whom the bank has no exposure during the period of one year after time 0.
We think that the changes in the definition and estimation of the parameters of the models should not be considered as material in the sense of RTS on assessment methodology by which competent authorities will assess in detail the models used for the purpose of the IRB approach and verify how risk parameters are estimated, since they stem from a decision of the regulatory authority. At the minimum, we advocate for a more rationalised application of this RTS during the transition period which could also be more flexible.
Back-testing procedures: The main objective of back-testing is to validate if the time series used to compute risk parameters are in line with the most recent observation periods. If this result in a statistical difference between the risk parameters of the model and the risk parameters based on more recent time series, it should originate a review of the model risk parameters. These assumptions/ procedures could lead to contradictory procedures and results when compared to the ones obtained through a TTC approach. In this sense, further clarification how to design, conceptualize and implement the relationship between the back-testing (precise rules and objectives) and the risk parameters measurement (economic cycle, PIT vs. TTC, among others) should be provided and should be a priority as they affect all the remaining aspects of section 4.3.2. The link between back-testing and the stability of parameter estimates should especially be addressed. In particular, some guidelines are welcome on how many back-testing failures are acceptable before the model needs a complete review. In this context, a minimal required level of model’s explanatory / discriminatory power may be a useful benchmark. Finally, the practical meaning of “the expected range of estimation errors” used to define the confidence intervals for parameter estimates in back-testing needs to be explained, and quantified in statistical terms to make it less subjective.
10. Do you have dedicated LGD models for exposures in default that fulfil the requirements specified in section 4.3.4.(ii)?
Although there are some guidelines in the RTS Assessment Methodology that relate to LGD in default, institutions are expecting full new guidelines and waiting for them before making any changes in their current practices and models. Indeed, this point is closely linked to downturn LGD (on going guidelines). The guidelines already defined introduce material changes for institutions: Separation between defaulted and non-defaulted exposures;
LGD in default should include the information on the time in-default and recoveries realised so far;
Each difference between Expected Loss Best Estimate (ELBE) and specific risk adjustment should be duly justified.
These changes impact modelling practices and information system (data to store and to introduce in information system).
11. Do you consider the direction of the proposed changes adequate to address the weaknesses and divergences in the treatment of defaulted assets across institutions?
Preliminary analysis seems to include that main weaknesses and divergences relating to the treatment of defaulted assets are: Different practices in the estimation of LGD in-default;
Different practices in the estimation of ELBE;
Difference in the level of calculation of the IRB shortfall.
We think that the provision set in the text will enhance consistency as far as the level of calculation of the IRB Shortfall is concerned. We understand that it has to be calculated at an aggregate level separately for the portfolio of defaulted exposures and the portfolio of exposures that are not in default.
It is hard to have an opinion on the two other sources of discrepancies because the RTS does not contain much information on it. Requirements on LGD in default will be more precisely set in the planned guidelines through the definition of downturn LGD. However, there is at that stage no specific indication on the estimation of either the ELBE for defaulted exposures or the add-on that captures the unexpected loss that might occur during the recovery period.
For the sake of clarity, the first bullet point of paragraph 68, “direct estimation of downturn LGD and ELBE for defaulted exposures” should be replaced by “direct estimation of LGD in-default and ELBE for defaulted exposures”, as the downturn LGD is a concept applied to non-defaulted exposures, being LGD in-default the correct concept to capture the LGD for defaulted assets in a downturn/unexpected loss perspective.
12. What else should be covered by the GL on the treatment of defaulted assets?
In order to avoid any misunderstanding the GL should state clearly that ELBE estimates should reflect current economic conditions and the institutions aren’t obliged to apply downturn stress factors to such modelling as required for LGD for non-defaulted assets.We also recommend the alignment of default definition in all regulations within EU by the regulator (e.g. IFRS, Basel II, EBA ITS). Currently EBA ITS requires to disclose non-performing exposure which are classified as performing under Basel, e.g. loans with 2 forbearance measures which were never 90 days or more overdue, no default rating etc. To reduce complexity, internally and also externally for analysts, it would be appreciated if the definition in the various regulations would be aligned, at least within EU.
13. What are the impacts for the institutions that should be considered when specifying the conditions for PPU and roll-out?
We recall that it is necessary to maintain some flexibility in the 5-year deadline for the implementation period of the IRB approach. In particular, it is essential that competent authorities ease rules so that different situations such as mergers and acquisitions can be taken into account. This also relates to change in the roll-out plans. Economic conditions as well the structure of portfolio (sales) may change over time leading institutions to review their plan.It is important to clarify whether the application of this new treatment proposed for the PPU and roll-out consideration will have retrospective effects or not.
14. Do you expect that your organisational structure and/or allocation of responsibilities will have to be changed as a result of the rules described in section 4.3.5?
Using risk parameters on a daily basis, taking stressed situations into account when estimating parameters and implementing several control levels in the development and implementation of internal models are key elements to reduce discrepancies among institutions in capital requirements.However, to achieve such a situation, models should not be subject to numerous material changes over a limited time period, facing the risk of losing internal and external credibility.
The proportionality principle requires further and practical considerations in the EBA paper to understand, in an objective way, to which institutions it should be applied and, in those cases, what are the aspects that could be adjusted or reduced regarding the internal risk management process (process validation, number of risk models and segments, number of entities, timelines for implementation, etc.).
Most institutions organisational structure will be impacted following the requirements that relate to internal governance. Stress testing processes might also evolve to consider IRB stress tests as well as Pillar 2 stress tests. A wider use of risk parameters might lead to some changes in internal risk management processes.
15. Do you agree that CRM is a low priority area as regards the regulatory developments?
When compared to the remaining areas, we agree that CRM is a low priority area in the EBA timeline as it refers to specific points of CRM (a reference is made to 3 focused RTS) and not the fundamental review of the CRM framework. The omission of such guidance, in the beginning of the process, could lead to undesirable multiple changes in IT systems and LGD models to address, in a first instance, LGD calibration and, in a second instance, LGD revision due to changes in CRM rules. Two structural changes instead of one will lead to the loss of synergies obtained in a unique major LGD revision and will create additional costs for banks.16. Are there any other significant intra-EU or global discrepancies?
Model options in the EU and globally should not deviate. In any case, future guidelines should address and clarify what should be assumed (locally and at consolidated level) in the case that these discrepancies exist within an international financial group, where home and host supervision practices may diverge.It has been the experience of banks that application procedures and regulatory response to applications for model approvals varies widely. This too is a source of divergence in risk estimates between banks. Convergence towards a timely and efficient regulatory approval capability is to be encouraged, preferably with time commitments from the competent authorities. Untimely approval undermines the constant investment made by banks in their risk measurement systems and in model use and risk management.
17. Do you agree that the area of disclosures needs to be strengthened, in particular with regard to disclosures related with the benchmarking exercise, for instance by publishing them on the EBA website?
We agree that the disclosure area should be strengthened. However, any mandatory requirement for banks to disclose their benchmarking exercise results, in the Pillar 3 report, does not seem appropriate as it can trigger stress events, if not understood correctly by investors. We highlight that bank counterparties and stakeholders will be highly sensitive to the market’s interpretation of the disclosed results, compared to other risk disclosures, and thus we recommend that, if such a disposition is being studied, the IRB benchmarking results disclosure should be developed with this consideration in mind.We disagree with the publication of benchmarking exercise results whether public disclosure or on the EBA website for the following reasons:
The requirements defined in the final version of the BCBS disclosures already include numerous templates that provide a wide range of meaningful information to external stakeholders;
These exercises rely on hypothetical portfolios that have nothing to do with institutions’ real business;
They mainly help competent authorities or the EBA at understanding models and comparing them for regulatory purposes;
Publishing such information will increase complexity of the disclosed information and spread confusion in external stakeholders’ mind and therefore entailing misleading interpretation and analysis;
Difficult to establish a link with Pillar 2 disclosure. May raise more questions that understanding institutions performance.
We also remind that we strongly disagree with the BCBS project requiring banks that are using the IRB approach to disclose hypothetical capital requirement according to the standardised approach for credit risk.
This approach is simplistic and will again create confusion in external stakeholders’ mind and lead to irrelevant questioning and complex debates which would be difficult to solve without the disclosure of proprietary or confidential information.
Disclosure is a very important topic and we understand the need of external stakeholders to understand institutions finance performance. We also agree that adequate disclosure will enhance transparency and reinforce internal model credibility.
18. Would you support EBA Guidelines targeted at disclosure requirements related with the IRB Approach and taking into consideration the proposals of the Basel Committee on those requirements? Which current disclosure requirements should be given the priority? What should be the timetable for such Guidelines?
The Transparency Exercises have indeed provided harmonized, easy to access information on RWA to users and seems to have been appreciated. According to the feedback received by the banks, their effective use by analysts and investors has been rather limited.We have the following concern as far as the additional ad hoc disclosure is concerned:
• We question the appropriateness of ad hoc disclosures on the EBA website. As mentioned above, we think that these requirements will increase the number of disclosed information and may introduce confusion in stakeholders mind. Moreover, we wonder whether the EBA does not go beyond its mandate by so doing.
• We think that for this ad hoc disclosure to be relevant, it has to be consistent with BCBS requirement. There must be a bridge between disclosed data (i.e. stakeholders shall be able to switch from one set of information to the other one without having to reconcile the information).
• There is no information on the frequency but we suggest it will follow the BCBS defined frequency.
• The disclosed data should really add value to the regulatory disclosure. Such high granularity is most often totally useless to most market users, but it provides valuable information on market shares and customers to competitors (notably to non-EU banks).
Apart from this, we support the EBA proposal to issue guidelines for reporting ad hoc information to be harmonised and therefore avoiding any discrepancies.
Institutions shall publish information consistent with the BCBS disclosure requirements as at January 2015 from end 2016. This ad hoc disclosure should not be made mandatory before the enforcement of BSBS disclosure and the implementation of all new regulatory requirements. One should not assume it represents a limited burden on institutions which have already to implement an unprecedented number of regulatory changes and of disclosure requirements in a very short timeframe. In addition, discrepancies may emerge from change in regulatory requirements as far as models are concerned.
We are also concerned that the EBA's project to anticipate this BCBS disclosure for EU banks creates once again an unlevelled playing field with non-EU banks.
19. Would you like to see any modification of the reporting framework implemented in terms of IRB exposures?
There is today a growing trend to standardization as well as homogenisation on this topic. We agree on the need for harmonisation and comparability. However, too much harmonised granularity may lead to the opposite effect, could be detrimental to simplicity and lead to misleading comparison. To maintain this objective of comparability, institution’s internal organisation as well as their economic model should also to be taken into account.We would like to insist in the idea that, in our opinion, a simplification in the reporting framework will be a positive move in terms of efficiency and administrative and organisational costs for entities and ultimately for supervisors.
20. What would you consider an appropriate solution with regard to the definition and treatment (modelling restrictions) of the low default portfolios?
A common understanding of situations where internal models can be challenging to implement recognising the implications of thin data availability is welcome. More analysis or information regarding best practice should be distributed before any decision to limit modelling restrictions on LDP. Caution should be applied as excessive standardisation may not promote prudent risk behaviour. Standardisation may not promote incentives for selecting the best customers.Banks acknowledge the scarcity of data to fill up the supporting databases for low default portfolio (LDP) models. We note that EBA evokes in the Discussion Paper the possibility of applying the Standardized Approach for LDP. But shifting LDP models toward the Standardized Approach will increase the capital charge with no apparent economic reason. Therefore, we recommend the EBA to study the possibility of an intermediate model.
21. How would you ensure appropriate use of the IRB Approach in a harmonised manner without excessive concerns of the so called ‘cherry picking’?
The need for harmonisation must not overrule the principles of having efficient risk sensitive models, even though there are several model options with different outcome. There may be several good reasons for operating different models to different exposure classes. This needs to be carefully assessed to avoid introducing changes indiscriminately. Furthermore, we do not believe that the current governance framework and rating models for internal models allows for “cherry picking”.22. Do you see merit in moving towards the harmonisation of the exposure classes for the purpose of the IRB and the Standardised Approach?
One solution could be to move towards the harmonization of the exposure class definitions between IRB and standardized approach to improve the comparability between standardised and IRB Banks. This situation is especially important if the BCBS consultation document regarding capital floors advances to a framework in which a specific floor is defined for each exposure asset class. A premise for a correct functioning of that framework will always be “equivalent exposure class scopes” between standardized and IRB approaches. This is not only applied in terms of retail definition which is currently different, as mentioned in the discussion paper, but also for other classes that do not exist in standardized approach, namely the specialized lending (which is being proposed to be created under the standardized approach, by BCBS, in the “Revisions on the standardized approach for credit risk” consultation document).23. Would the requirement to use TTC approach in the rating systems lead to significant divergences with the internal risk management practices?
As mentioned in question 2, a TTC approach for IRB will diverge from the current movement on the accounting side to a PIT approach (according to IAS 39 / IFRS 9). This originates constraints when “use test” requirements are put in place, being this, one of the reasons why we support that, although IRB models and Credit provisioning models should be aligned, the use of IRB parameters as input parameters for Credit provisioning models is not a mandatory requirement.Consideration and clarity regarding the use of dual TTC and PIT PD approaches should be given within the PD estimation product.
24. Do you agree that the possibility to grant permission for the data waiver should be removed from the CRR?
We agree that a level playing field should be pursued, an aim that has not been assured with the current text. However, this could be achieved without the costs of removing the current waiver, namely by defining quantitative minimums (for population and defaults) that must be observed to allow a reduction of the minimum historical period from 5 to 2 years.25. Are there any other aspects of the IRB Approach not discussed in this document that should be reviewed in order to enhance comparability of the risk estimates and capital requirements?
It is important the IRB continues to reflect internal data, and that IRB in the future is designed as a tool for including risks that is reflecting banks own risk profile. Unquantifiable external risks should be implemented through other measures and buffer requirements. We consider that changes related to PD calibration, LGD calibration, and the definition of margin of conservatism should be subject a parallel review by the Basel Committee.The margin of conservatism must not be considered as a separate measure to increase the own funds. The margin of conservatism should be related to weaknesses in the models. It should not be possible to impose large add-ons to models. The design of the models must be respected, and must not be subject to add-ons based on unquantifiable supervisory judgements. There should be an upper limit related to e.g. a percentage of the internal estimates that could be added to the internal models.
Clarification of Pillar 1 IRB Stress Testing requirements and the relationship and delineation between this and Pillar 2 Stress Testing requirements.