We are overall in favour of Option 2 for non-retail clients. Please refer to question 2 for our proposed alternative approach for retail clients and our detailed analysis.
We propose to set an absolute maximum limit for retail clients. We believe this approach will adequately capture default positions, leaving out “technical defaults” that may bias the modelling framework, and remain operationally feasible. This approach is prudent, represents a pragmatic solution to the challenge of dealing with large pools of retail clients in bank’s retail branches and avoids undue complexity. We remain reluctant to consider a relative threshold here since it would behave in contradiction with current risk management practices. For example let us assume that a borrower draws on his authorization up to a certain amount, and that his credit quality deteriorates: the bank is then likely to decrease his authorization, which will then increase the relative rate of use of the authorization. While the absolute amount at risk for the bank is reduced, a default may be triggered by the relative exposure threshold. Also, please note that the absolute thresholds currently applied are largely lower than the proposed maximum thresholds of 200 € (4€ is an actual threshold for the consumer credit activity for instance).
As far as non-retail clients are concerned, we propose to the set the threshold as the max of a 500 € absolute limit and of a relative limit defined as 2% of the total amount of all credit obligations of the borrower (i.e. “option 2”). This approach reflects the wide variety of clients included in the non-retail population, ranging from SME to large corporates to sovereign entities. We believe this approach would provide for a more accurate measurement of risk parameters. We strongly believe that thresholds should be differentiated for non-retail (hence our proposal above to focus on the maximum of a relative threshold and of an absolute amount). On this basis we also believe that expert judgment should remain allowed for triggering defaults for non-retail counterparties in certain instances.
More generally, we believe that while a competent authority sets a certain threshold for its jurisdiction, banks in that jurisdiction should be allowed to trigger defaults internally with lower thresholds to capture material credit quality deteriorations before regulatory thresholds are hit, leveraging on existing internal risk monitoring procedures and specific knowledge of counterparties. We consider this would also be in line with the regulatory definition of default stemming from the concept of “unlikeliness” to pay.
We consider that an absolute threshold should remain both material and significant enough to prevent, as mentioned in the EBA consultation paper, from recognizing too many defaults that will be cured in a short timeframe. Irrespective of the retained target option, more identified defaults imply more recoveries and thus better LGD and an increased PD.
Irrespective of the final choice of options, if entities are not allowed to use only an absolute amount as a threshold, then option 2 would be preferred for the retail exposures as well as for non-retail ones.
Before implementation of new thresholds into models, institutions need to collect 5 years of historical data according to set levels. In practice, it is not considered as a feasible/reasonable scenario to recalibrate all current historical data taking into account new thresholds.
The draft RTS proposes a transition period of maximum 2 years for institutions to modify their models. A change of default definition that changes the number of defaults has material operational implications and may simply be impossible retroactively –an historical data set ranging from 5 to 7 years may not be adequately corrected, integrated and tested within the risk management system over a 2 year period.
As far as diligences linked to the materiality of changes of a rating system, no approval should be requested by the supervisor subsequent to those, especially if the implementation of different materiality threshold on models has no significant impact. In the latter instance, the competent authorities should thus be notified through ex-ante or ex-post communication. In all instances, we would like to propose one global communication to the supervisor on the change of threshold, as opposed to separate communications model by model.
NCAs should also bear in mind that the definition of retail portfolios is not always homogeneous between banks (owing to historical practices, and to the granularity of internal model approvals granted by supervisors, in connection with the client segmentation then in force within banks). We consider that the thresholds (retail Vs non-retail) should be implemented on the basis of the retail / non retail boundary currently implemented in the bank. Any modification of the perimeter of retail portfolios would entail significant impact over data series and additional operational constraints for banks, overstretching the transitional phase.
We would also like to point out that any discrepancy between regulatory and accounting definitions of default is likely to result in additional operational difficulties. In some jurisdictions convergence may have been achieved across both definitions. In this respect we would simply like to highlight that a coherent framework would help meet regulators’ expectations in terms of comparability of models and practices.
We would like to emphasise again that no approval should be requested by the supervisor/authorities subsequent to the changes contemplated in the draft RTS on the basis of PDs estimation already including margins of conservatism calibrated to take into account uncertainty in the evolution of default rate.
As a consequence, the impact of the implementation of different materiality thresholds is expected to be significant especially from an IT development perspective.
We urge the EBA to consider jointly the answers and comments submitted to this consultation, and to the consultation paper 36/2014, in so far as numerous subjects are inter-related and evolutions contemplated by EBA in some parts of CP36 will impact banks modelling of default in the context proposed by CP32.