Primary tabs

Czech Banking Association

The draft RTS specified that the bank has to find model components which drive the bimodal (or more generally multinomial) distribution of LGDs. Our banks are not using model components as it is supposed by the draft RTS (probability of cure model), but they are using overall models with risk factors, which are based on non-defaulted characteristics (e.g. LTV, collateral type, exposure type,…), which are changing over time.

The proposed way of finding model components and its dependency on macro factors will bring significant increase of model complexity, will be time consuming and with almost no benefits.

The approach with model components will require to create new LGD model for estimating downturn, which can be far away from currently one used in the bank. To create such new downturn models for all LGD models in a bank will take many years of development, internal validation and also will create a lot of material model changes which have to be approved by regulators. The regulatory approval of material model changes is currently time consuming on both side (bank/regulator) and take more than 1 year. In the case that all IRB banks will be forced to create material model changes for all models, the regulators will be overloaded and this exercise will take many years.

Moreover, usually all banks tried to find a way how to find a coincidence between macroeconomic factors and LGD values – at least for downturn LGD (this is not new in regulation), ICAAP and stress testing exercises, etc. Our banks undertake a lot of effort in the past to find such relation (trying many ways including model components) and no clear dependency was found. Especially in the Czech banking sector the LGD is strongly driven by bank policies (granting procedures, workout practices, …) than others.

As EBA doesn’t show any study of feasibility of such approach and the draft RTS is only theoretical concept and due to the fact that our banks undertake a lot of effort (including big expenses) in past we see proposed approach as an burdensome one, which only creates a lot of work with vague results at the end.

In such complex approach, proposed by EBA, it will be also meaningful to consider a suggestion to set a materiality threshold for which portfolios such complex analyses should be done. There are also immaterial portfolios with simple LGD models, where it makes no sense to create a quite complex methodology.
The real impact of such proposal will be that almost no dependency will be found and only panel of experts will artificially create a set of dependencies. This assumption is strongly against the nature of the whole idea created in Art. 2
One of the main issue in finding a dependency of LGDs on macroeconomic variables is that one that observed LGD is a results of workout process which usually take many years. The final LGD value is simple saying the final observed loss divided by initial exposure at default. It is not so problematic to distinguish the cash flow from different years, but this cash flow is usually dependent on the workout strategy of the bank and the concrete situation of the client rather than on overall economic conditions. Moreover, the cash-flow, which the bank receives from the clients in the same year is not independent from the year of default. Different situation is for the client who is for a short period in default and completely different for client with many years in default.

More generally, we are afraid that it is almost impossible to create a value of LGD for a certain time. In reality, this is really complicated while there is not only one collateral and on the other side, there are many loans with no collaterals.
The proposed period of twenty years, which, by EBA opinion, should cover two economic cycles is beyond CRR, which requires to have data for LGD estimation for 5/7 years and to add more data only in relevant cases.

Moreover, generally the banks in the Czech Republic do not have data prior 2007 for LGD estimation, as this is the year where our banks first started to use IRB models. Therefore, not taking into account the meaningful of such analyses, banks will be able to analyze the dependence of LGD values and macroeconomic factors only for 8 years at maximum.

Moreover, such long history of macroeconomic factors does not take into account the changes in model components behavior (as defined in Art. 2) – which will be mainly driven by changes in banks processes and therefore will not be stable within 20 years – the drivers which determine model component will be significantly different in year 1 and year 20 and therefore will not be able to fit bimodal distribution of LGD.
The proposed period of twenty years, which, by EBA opinion, should cover two economic cycles is beyond CRR, which requires to have data for LGD estimation for 5/7 years and to add more data only in relevant cases.

Moreover, generally the banks in the Czech Republic do not have data prior 2007 for LGD estimation, as this is the year where our banks first started to use IRB models. Therefore, not taking into account the meaningful of such analyses, banks will be able to analyze the dependence of LGD values and macroeconomic factors only for 8 years at maximum.

Moreover, such long history of macroeconomic factors does not take into account the changes in model components behavior (as defined in Art. 2) – which will be mainly driven by changes in banks processes and therefore will not be stable within 20 years – the drivers which determine model component will be significantly different in year 1 and year 20 and therefore will not be able to fit bimodal distribution of LGD.
We believe that the bank should prove in their own way if there is some dependency on the macroeconomic variables and if not, this should be covered by margin. Such approach should be challenged by regulators and the banks should be able to argue for their approach.

We therefore support the reference value approach, which is close to what our banks currently do and it is not burdensome neither for the institution nor for the regulator. The reference value approach much more relies on institution specificities and will allow institution to use the approach for which the institution has data, capacities, possibilities and is more appropriate.
Petr Vojáček
C