Stellungnahme DK zu "EBA RTS-/ITS-Entwurf zum Benchmarking interner Modelle"
Any analysis of the root causes for differences between banks needs to factor in a number of methodological peculiarities: In the studies published to date, the EBA itself pointed out that banks' rating models differ in terms of the mapping to rating classes, the data sources (internal, external, hybrid), the type and nature of the use of historic data (length, weighting) the application of the floors, the definition of default as well as the calibration details (e.g. adjustment by means of the through-the-cycle or pointin-time-methodology, underlying master scales). This may result in different outcomes. Whilst not limited to, such differences may also result from different supervisory requirements under various national jurisdictions. At least if and when they demonstrably lead to significant RWA differences, such requirements should be standardised. Notwithstanding the foregoing, a full harmonisation is not advisable if it would give rise to high costs and if it would render the use of existing histories impossible.
Benchmarking forms part of the internal models' quantitative validation. However, the second material element consists in backtesting. In our understanding, there is an indispensable need for preserving the internal models' existing risk sensitivity; also, the benchmarking approach should not be used as a backdoor for the introduction of a compulsory uniformity. The most likely way of preventing the latter is if backtesting is generally given a higher priority than benchmarking. This should be applicable at least in the presence of methodologically flawless backtesting approaches (which does not apply to each and any internal approach) and it is applicable to cases where - also in the IRBA area - low default problems play no role. What must be avoided at this point is that - although backtesting does not suggest any need for action - corrections have to be carried out on the grounds of benchmarking concerns. The RTS should clarify beyond any doubt that, provided there is a sound reason, also larger RWA differences may potentially be legitimate and that these do not necessarily have to be levelled out. In light of the above, we are of the opinion that benchmarking for portfolios which lack backtesting options can be productive. In the presence of comparable markets and asset classes this would only offer value added for the other portfolios if risk sensitive benchmarks were established (e.g. by defining absolute deltas that ought to be deemed "normal"). However, from the present draft it does not become clear that benchmarking offers value added for portfolios with sufficient backtesting options. […]