LBBW (Landesbank Baden-Württemberg)

In principle, we support the proposal concerning an independent validation function. We too attach great importance to the validation function's independence. It should be prevented, however, that this results in requirements being imposed that go beyond what is necessary for ensuring that independence.
We agree that developing and validating rating systems requires highly qualified staff. However, we see difficulties in attracting that highly qualified staff when requiring the independent validation function as proposed in the RTS.

Article 4(3)
We regard the provision of Article 4, section 3 of the draft RTS as particularly problematic because it would prevent institutions from using jointly developed IRBA systems based on pooled data (referred to as pool models") in future, which are very common in Germany.

The key idea of pool models is to pool data from several institutions, based on which a central outsourced unit (the "pool service provider") develops, maintains, reviews, enhances, and operates collective rating systems. The institutions involved participate in the development and enhancement of the systems. All decisions, for example about changes to the rating systems, are made by joint bodies according to clearly defined rules. Based on the pooled data the quality of the pool model is regularly analyzed by the pool service provider and evaluated by the joint bodies. In addition to the pool analyses the central unit provides statistics on each institution's individual portfolio, based on which institutions will perform independent internal validations and examine the representativeness of the pooled data and the validity of the pool results for their own portfolios (see Article 179, section 2, item b of the CRR).

Such pool solutions present a number of advantages:
• There is much more data available for estimating the models. As a result models are significantly more differentiated and accurate in terms of discriminatory power than individual models developed by each institution.
• Jointly developed rating systems reduce the variability of the estimated risk parameters between the institutions involved.
• Combining the institutions' expertise helps improve the rating systems in general.
• The joint processes applied by the institutions and the central unit lead to more objective decisions about the models; individual interests of the parties involved become transparent.
• The costs of maintenance and review, documentation, further development, IT implementation, and operation are lower.
• Experience shows that there are also advantages for supervisory authorities: they need to audit the models and all changes to them only once and can communicate efficiently with the pool service provider's central points of contact.


The successful collaboration between a central unit and institutions described above is made impossible by the provision of Article 4, section 3 of the draft RTS because this provision bans external service providers that participate in the development of the rating models and risk quantification tasks from being involved in the validation of those models carried out by the institutions. As mentioned above the validation is performed by each institution, but the institutions rely on results and analyses provided by the pool service provider, which developed the models. If the pool service provider was in fact prevented from participating in the validation, pool models would become impracticable: an institution which has outsourced the development of a rating system, an essential task of the credit risk control unit, could no longer comply with the requirement of Article 190, section 2, item f of the CRR ("active participation in the […] validation of models used […]") as, pursuant to Article 4, section 3 of the draft RTS, the external service provider could not be involved in the activities of the validation unit.

We therefore suggest to drop the provision of Article 4, section 3 of the draft RTS. At least the RTS should clarify that a third party involved in the further development and enhancement of a rating system may actively participate in the validation to be able to comply with Article 190, section 2, item f of the CRR even if the validation is carried out by the institutions.

Article 10
In our view Article 10 fails to serve its purpose in its present form as well. The strict organizational separation that is being proposed especially for systemically important banks is not necessary to ensure that rating systems are validated objectively.

We feel that the risk of a conflict of interest between those responsible for model development and those in charge of validation is being overrated since both groups are equally interested in ensuring the validity of the models. In any case the potential for conflicts of interest is considerably smaller than, for instance, between front office units granting loans and the credit risk control unit. In addition, for many years internal monitoring systems have been in place and internal auditors and supervisory authorities have been checking institutions to ensure the validation policies are complied with.

We would like to point out the following issues:
Rating systems often cover low-default portfolios. Therefore it usually takes several years to collect statistically significant data for improving the rating systems. In addition, the professional IT implementation as well as the regulatory approval according to the model change policy process takes some time. Thus even minor model changes often take a couple of months, major changes a year or even more before going life. Furthermore, to verify that such a model change stands the test of reality we ideally need another two years, at least one full year of observation. In our experience, major model changes of a specific rating system only take place every three to five years, minor model changes (e.g. recalibration) every two years.

Model changes require highly qualified staff. In most cases, the crucial skills are not the general knowledge of statistics or methodology, but the detailed knowledge of the specific rating system with its data model, the relevant underlying business and its specialties. A lack of this knowledge leads to rating models hat overlook relevant risks and thus may not be applicable in practice. Specialties of the data may not be adequately taken into account.

Since model changes are relatively seldom, the relevant knowledge can only be acquired during the regularly validation and through the continuous intercommunion with the rating users as part of the credit risk control unit.

A strict separation on a personal level would exclude the most competent staff (those who regularly validated the rating system) from a model redevelopment. In addition, the validation function will be less attractive since it is separated from the model development and the credit risk control function. Thus it is questionable that it attracts highly qualified staff to the same degree as today.

We agree that it is reasonable to limit incentive to disguise the model deficiencies and weaknesses. We also agree that a fresh view on the rating models and transparency in general are reasonable aims. In our opinion, these goals can also be met by other means as for example a clean cut between validation and development with separated approvals or regularly changes of responsibility for the validation of a specific rating system.

Further, pool models are designed in a way that encourages independent validation: the joint processes applied by the institutions and the central unit lead to more objective decisions about the models; individual interests of the parties involved become transparent. Models are reviewed, enhanced and further developed by the pool service provider. On this basis the institutions contributing to the pool decide on validation issues according to clearly defined rules. This ensures objectivity.

We think that it is absolutely necessary for performing validations in practice that the credit risk control unit is involved both in the design and implementation and in the validation and modification of those models. This unit has the necessary expertise not only to develop suitable and economically reasonable models, but also to assess their reliability.

Moreover, there are interdependencies between the development and validation of models which require regular communication between the individuals responsible for these tasks. Thus, insights from validation may be crucial for further development and validation, in our view, absolutely requires knowledge of how the model was developed. In our opinion, linking validation and further development does not preclude independent validation.

These issues were already discussed in connection with CRD I - which contained the exact same text on this matter - in the 2006 CEBS guidelines on validation. At that time, for the reasons given above, CEBS concluded that development and validation may be assigned to the same unit (section 419).

In our view the organizational integration of the validation function should not depend on the size of the bank. It should be governed by the aim to ensure objectivity rather than any restrictive formal requirements that, while failing to significantly improve objectivity, will seriously complicate validations in practice. This is why we object to a separation of the validation function from the credit risk control function up to senior management level based on a bank's complexity and the risk inherent in its portfolio."
The prohibition in Art. 4(3) of cooperation between model developers and model validators under outsourcing arrangements renders pool solutions common in Germany practically infeasible, especially as it is not clear how an institution can satisfy the requirement of CRR Art. 190(2)(f) CRR, if it has outsourced the development of the rating system as a key task of the credit risk control unit and Art. 4(3) RTS draft does not permit the outsourcing unit to be involved in the validation unit's activities. The provisions in CRR Art. 179(2) and Art. 190(3)(a)-(e) are de facto rendered null and void by the RTS draft provisions and the large number of advantages for institutions and supervisory bodies alike can no longer be used, see our response to Q2.

Having to forego pool-rating procedures would have a huge impact on the institutions affected: massive adjustments to the IRBA systems and for each individual institution's organisational structures would be necessary (especially installing resources for permanent local model development and validation). It remains unclear whether the validity of an individual institution's models can be shown in the long run without recourse to the shared data pool (as the starting point of shared modelling).

The independence demanded for large banks up to the senior management level is in principle not workable. It would necessarily entail, however, not only a massive adjustment of the organisation structures but also a duplication of activities, without significant additional objectivity. Here, we believe there should be no overshooting of the goal, particularly as the conflicts of interest in conjunction with model development and its validation are clearly smaller than, for example, for the organisational separation between front and back offices.

In any case, such a rigid organisational separation between validation and model development would clearly delay the model optimisation implementation processes.
Article 4(3)
The prohibition in Art. 4(3) of cooperation between model developers and model validators under outsourcing arrangements renders pool solutions common in Germany practically infeasible, especially as it is not clear how an institution can satisfy the requirement of CRR Art. 190(2)(f) CRR, if it has outsourced the development of the rating system as a key task of the credit risk control unit and Art. 4(3) RTS draft does not permit the outsourcing unit to be involved in the validation unit's activities. The provisions in CRR Art. 179(2) and Art. 190(3)(a)-(e) are de facto rendered null and void by the RTS draft provisions and the large number of advantages for institutions and supervisory bodies alike can no longer be used, see our response to Q2.

Having to forego pool-rating procedures would have immense effects on the affected institutions: massive adjustments to the IRB systems and for each individual institution's organisational structures would be necessary (especially installing resources for permanent local model development and validation). It remains unclear whether the validity of an individual institution's models can be shown in the long run without recourse to the shared data pool (as the starting point of shared modelling).

Each of these changes can and will, given an unchanged final version of the RTS draft, lead not only to major changes to the prevailing IRB rating systems, but at least in part also to major shifts in the institutions' model landscape.
Axel Kuemmel