As stated at the beginning, on the basis of a first assessment, we assume that all IRB models will require material changes, which means that in each case an IRB acceptance test will be necessary.
The freedom of methods enables institutions to choose the appropriate rating approach for the respective portfolio and the information available on it. Also the methodology (shadow rating approaches v scorecard-based procedures) would be likely to influence the choice of approach.
In practice, there are few pure through-the-cycle or point-in-time approaches, but many hybrid forms, so explicit stipulation of a rating philosophy, as called for in point 78, seems problematic. Point 78 should therefore be deleted.
The parallel use of different rating approaches for the PD estimate is as a rule not logical for Bausparkassen, with their focus on financing of home ownership and accordingly homogeneous loan portfolios.
Explicit stipulation of a specific rating philosophy does not usually occur. We refer to the response to Q 5.5 (=Question 7).
We reject the additional requirements, since the requirements for security margins are in our view sufficiently regulated in the CRR. Security margins are therefore already comprehensively taken into consideration in the IRB models.
The conceivable deficiencies of the estimates in practice frequently cannot be differentiated unequivocally and unambiguously in accordance with points 24 and 25 and the categories (A to D) and subcategories described there. As a rule, neither the identification and definition of types of errors nor their quantification according to these granular categories and subcategories are possible in a meaningful way.
The requirements governing a process for the identification, quantification, documentation and monitoring of the various types of errors would in our view lead to considerable implementation expense without thereby allowing perceptible reduction of the RWA variability of comparable portfolios.
We reject the requirements concerning the representativeness of data because implementation would be very costly without any recognisable added value.
In our opinion, there are valid situations which justify disregarding a dataset for methodological reasons, for example if it relates to products which are no longer sold and so belong to an expiring portfolio. In this case, excluding the dataset would usually even improve the forecasting quality.