The model component approach does not seem the adequate methodology for several reasons:
1) Complexity of the approach
Overall, the guidelines are not easy to understand and seem very complex to implement for any kind of portfolio.
Economic downturn is part of the IRB model framework: the downturn estimation is therefore back-tested and where necessary recalibrated. When concerned, model changes are notified to the supervisor given classification of the model change policy. Therefore, should the model component approach be applied, it would be unduly burdensome and complex to maintain such a framework in the model lifecycle process and would create a labyrinthine system.
2) Excessive conservatism of the model component approach
Generally, it seems that applying the worst level observed for a model component, considering each exposure separately, would imply overstated LGD, CF and RWA, disconnected from reality. It will probably imply levels of unexpected losses which would be higher than any possible losses.
While leading quantitative and qualitative studies to analyze correlation between model components and economic factors could be in some cases relevant (set aside the effort which is requested from panel of economists and model experts), the choice of the final downturn scenario is based on a conservative choice (the worst of the worst) which annihilate the primary effort to identify the relevant economic factors.
Also, it seems unclear whether joint impact should be assessed with a panel of economists (included in the panel of experts), bringing further effort developing the model component approach, especially if the final choice is a conservative choice. Overall, adequate estimation does not mean conservative estimation. Conservatism should be brought to risk parameters only by margins of conservatism (MoC).
3) Applicability to portfolio structure
The basic of banks business is to manage diversified portfolios. We note that diversification and portfolio effects are not taken into account here.
Ignoring diversification is not reinforcing the regulatory framework:
• By not assessing the diversification quality of banks portfolios, risks can be overstated.
• By not taking into account diversification quality in the overall UL risk, regulatory capital assessment is biased and disconnected from reality. It would imply overstated UL and RWA.
The guidelines consider correlation of PD and LGD which is not necessarily the case. For SL for example, a higher leverage would imply a higher PD but not necessarily a higher LGD, as the residual asset life would be a key driver of the LGD: the residual asset life will enable to restructure the loan with postponement of the maturity, which can strongly reduce the possible loss.
The approach with model components seems to consider that defaults are necessarily due to macroeconomic features, yet idiosyncratic risks should also be taken into account. A default on a project finance for example can be due to performance issue on a turbine, or a leakage, etc. Deals can be relatively protected from macroeconomic impacts, as they can benefit from offtake (sales) contracts, with buffers of cash flows generated by the asset financed much higher than the debt service to be paid. Defaults during economic crisis periods don’t necessarily imply high losses, as explained before, depending on the capacity to restructure the loans on the residual asset lives, losses can be very low or null.
Request of separate assessment of Downturn for LGD and CF: the Downturn impact should be assessed considering altogether default, LGD and CF in order to have consistent data and scenarios. Otherwise it will imply overly punitive results.
Moreover, the proposal does not seem applicable to LDP.
The type of model should also be taken into account, and the approach for Downturn be adapted to theoretical models like asset value based models or cash flows models. For these types of models, banks have developed specific approaches to take into account the Downturn elements. These approaches should be taken into account in the guidelines.
Proposal from FBF:
We are in favor of applying one of the alternative approaches laid oud in the RTS: the distributional approach or the institution specific reference value approach.
Regarding distributional approach based on observed recoveries:
It seems the best estimate which could be used as it is based on volatility of historical losses/recoveries and would reflect internal data. Also, clear breakdown should be made between downturn estimation and MoC in the observed volatility of observed recoveries.
Regarding reference value approach:
When the reference value is fixed by the supervisor: the value could be difficult to calibrate: the one-size-fits-it-all value could be detrimental to a risk-sensitive approach (regarding for instance portfolio structure).
When the reference value is institution-specific (e.g.: average LGD for the worst two years) : it could be an interesting alternative. However, we would welcome clarification on the consistency between the SSM and the EBA positions (comparable requirements with Guide for TRIM). Also, we are not in favor of a reference being considered as a hard floor: the institution should be able to use values different from reference values upon thorough justification. The approach using the average LGD for the worst two years would not be applicable to non statistical models. Theoretical models require different approaches as explained above.
Regarding other approaches which are not retained by the FBF:
Downturn discounting rate with fixed add-on: we would welcome clarification of what a downturn discount rate is and the economic rationale behind. Also, for theoretical models, based on asset values or future cash flows to be generated by the assets financed, uncertainty of recoveries and their downturn nature are taken into account in the asset values simulations, through strong haircuts or volatilities applied. Applying a downturn discount factor for those models would imply a double counting of downturn elements. It would imply an unduly punitive overstatement of LGD.
We think that some flexibility should be left to banks in order to accommodate the granularity and diversity that banks may have in their classification. Indeed, on a certain type of exposure, a LGD-model may not necessarily have the same downturn period as for a CCF-model on the same exposure.
The concept of model components seems relatively but not necessarily sufficiently clear, and difficult to apply, with results implying overstated RWA. Should the model component approach be imposed, banks would need examples of application on models and related portfolios.
We ask the EBA to clarify in its draft RTS that this breakdown of LGD between cured and not cured components is only for illustrative purposes and has no objective of prescribing banks how to deal with economic downturn modelling. Also, the relation between identification of model components and the multimodal distribution realized LGDs or drawings seems unclear: we would welcome further economic and banking explanation of the rationale of this link.
It should be noted that the relevancy of “cure rate” will be more difficult to evidence when it comes to LDPs. The other concepts of these guidelines don’t seem adapted for LDP.
As regards to operational questions, as already indicated in our previous comments, we consider the model component approach quite complex to implement, and will therefore provide limited benefits.
The proposed model component approach seems to consider that LGD can be assessed by a linear formula based on several model components. Nevertheless LGD is not necessarily modeled with linear formulas. The proposed approach can be highly difficult to achieve.
Model components: it is unclear whether banks would take the worst observation for each model component (which is overly conservative), or define scenarios with only one or several model component being the worst.
For the definition of scenarios, these scenarios should be realistic scenarios with values taken for the different model components being consistent.
Anyway as mentioned before, taking an approach of “worst” model component on an exposure by exposure basis, ignoring diversification is overly conservative and would probably imply strong and unrealistic RWA increases, which should be tested before imposing it to the banks.
Also it is indicated in the explanatory text page 27 that the worst realization for economic factors would be based on the institution historical observations. Yet the data available might not be sufficient for calibration of a model. Institution data can be used for back-testing purposes, but not for calibration of the model, notably for LDP.
Yes, in particular the requirement that a panel of experts responsible of assessing the dependency between the selected economic factors and the model components has to be independent from the modelling team is particularly concerning.
Besides, in addition to the analysis of statistical correlation, the draft RTS states that banks shall also take into account the expected correlation based on economic reasoning, benchmarking and stress scenarios, both on a qualitative and a quantitative basis. However the draft RTS does not precise neither which assessment, qualitative or quantitative, will be predominant, nor how to benchmark with other banks. We think these uncertainties will not contribute to reduce variability in banks’ RWA.
No specific remarks.
A one-year period could be a suitable for some cases. However, we would welcome clarification of the economic rationale behind the choice of a one-year period:
• Economic crisis often do not last exactly one year ;
• Choosing the “worst” value of economic factor seems contradictory when economic expertise is introduced
Set aside the methodology chosen, should downturn period be defined, the one-year period is therefore not the most adequate choice based on economic judgement.
We question the proposed approach especially the requirement of a minimum period of 20 years of historical values, which seems too long:
• Concerning economic factors, structural breaks can be evidenced in the series throughout the years, switching from one regime to another (e.g.: unemployment rates, interest rates);
• Our understanding is that using 20 years for historical values of model components is not relevant, especially for availability reasons and because the definition of default will not be homogenous during such a reference period. This shows up limited interest of such a method. Also requiring to have at least two economic cycles would not necessarily be achievable.
We would welcome clarification of the economic rationale behind, especially regarding what is laid out in our answer to question 7.
The illustrative case evidences the use of cure rate as a model component: it would be useful to clarify the appropriate period of time to consider that a loan becomes performing again, and what is the definition of “cure rate”.
Finally, structural breaks in the series of economic factors or model components could lead to profiles switching regimes. Therefore, the absolute level of economic factors should be interpreted conjointly with the related period of time: the relevancy of using long-run average values or worst values over the period should be evidenced.
Once again, we consider the overall proposed approach quite complex in terms of implementation for a limited benefit.
We consider that it will be quite difficult to look back on a 20 year period of internal data.
We are not convinced that the same approach for the identification of the final downturn scenario proposed for LGD could be adopted also for the purpose of downturn CF estimation.
Given the complexity of the guidelines, an example of application on a given portfolio and model would be useful. Otherwise no specific comments that were not already made.
Yes, simpler alternatives approaches should be considered, but not only for proportionality purposes (see also answers to Q1). Particularly for model component approach: it seems quite complex; might lead to unrealistic outcome; will mobilize important resources for limited benefit.
EBA should rather contemplate a more flexible approach to account for downturn, by leaving the ability for banks to apply either a simple approach (supervisory add-on for downturn) or a more granular approach (distributional approach or reference value approach), on specific portfolios.
Current methods applied by banks should be taken into account by the EBA in order to avoid for the bank too many works to adapt their models to these guidelines.
Same comments as for Q14.
This question is indeed of the upmost importance. We understand that the EBA needs to have a clear view on the different modeling techniques used by banks for the Downturn assessment. We suggest to refer to survey results which are already produced or will be produced by several banking associations on the downturn estimation.