IRB model repairs: Guidelines and definition of default

IRB model repairs: Guidelines and definition of default

By Christian Dusterberg, IFRS 9 Senior Advisor, Erste Group.

Christian, can you please tell the Risk Insights readers a little bit about yourself, your experiences and what your current professional focus is?

While I have been working heavily on IRB models before, my recent focus contrinues to be on IFRS9 implementation and go-live, including methodological work on models etc and implications for operative processes in managing credit risk provisions.

At Risk EMEA 2018, you will be joining us to discuss ‘IRB model repairs: Guidelines and definition of default and impact on estimates and risk parameters’ – Why is this a key talking point in the industry right now?

For the first time European regulators are issuing fairly specific standards for IRB modelling, that apply to all EU member countries. Previously the (more or less) specific standards were up to national regulators to define, and hence in practice differed widely. This was correctly identified by EBA as a source for variation between RWA that should be reduced.

So, in principle I believe this is a good move. However implementing these standards over the subsequent years is a huge amount of work (depending on which specific interpretations institutions complied with in the past) not only of developing and implementing and managing any capital impacts, but very importantly also in the application and approvals processes on the side of regulators and their interaction with banks.

Further, the more specific the interpretations for IRB models are, the more likely is the need to tailor models for different uses, which increases complexity. So there would be a “core model” and then certain elements would be tailored for specific uses; taking a straightforward example: discount rates in LGD modelling will likely have to be different for regulatory purposes and for IFRS9 purposes.

 Why is it important to align default for capital and impairment?

As a general principle simplification should be sought wherever possible. If a default definition is not aligned, then this affects all risk parameters – they would have to be estimated distinctly for capital and impairment hence increasing the complexity of the model landscape, and the difficulty in reconciling model outcomes for different uses.

What could be the impacts of changing to the EMA 90-day rule?

This really depends on which jurisdiction an institution was exposed to in the past. It may be very limited, it may be significant.

Conceptually it could even be beneficial (higher PD, lower LGD while keeping EL constant would lead to lower RWA in most situations). However in practice this will not materialize due to the uncertainties (and hence necessary conservatism in parameters) in generating historic data for PD and LGD estimation under a revised default definition.

What might have to be considered with collection and recovering practices?

Here the impact of ECB NPL guidelines will be interesting: If these result in different behaviour in future NPL management then this will need to be considered in models built for the future. And if these result in different impacts of NPL related charges on capital and on impairment, then this will need to be reflected in models differing for the two uses.

If there is a substantial amount of specific charges applying only to capital and not to impairments (to which accounting rules apply which may not allow as much conservatism as the prudential regular prefers for NPLs) then this can drive a further wedge between the different model uses.

Finally, what challenges do you foresee in the future? And have you got any advice for your peers on how to best handle them?

Focusing on the modelling aspects, the key challenges I see in keeping the model landscape simple and managing the different requirement for different model uses effectively, including implementation aspects (e.g. two LGDs for one exposure – one for capital, one for IFRS9).

From a senior management point of view it will also be very challenging to have good quality management reports that deal with the different views on the same portfolio: P&L impacts being driven by IFRS9 models and RWA impacts driven by capital models. This is already a challenge today and this will become more difficult in the future.