Lorenzo Bocchi, Director, Prometeia
The European Banking Authority (EBA) has just closed the consultation period for its guidelines about probability of default (PD) and loss given default (LGD) estimation, and the treatment of defaulted assets by banks. Prometeia, the global leader in Risk Management software and solutions, has been among the contributors.
The EBA consultation is part of the work of regulators on a global scale to reduce the excessive variability of internal models for credit risk management, not correlated to the inherent risks of banks. With regard to the internal models, alongside the EBA discussion, a regulatory review by the ECB’s Single Supervisory Mechanism and the regulatory proposals of the Basel Committee on the so-called ‘output floors’ are ongoing (as yet not finalised).
The topic EBA is requesting proposals upon to practioners and researchers is a well-recognized issue for Italian banks, that are asking not to be penalised in terms of capital requirements in the event of massive sales of bad loans, as is currently occurring due to the degradation of the LGD parameter, used in the credit risk internal models of the intermediaries. The EBA guidelines, if definitively approved, may impose tighter calculations of the LGD as from 2020.
Prometeia believes that the impact of the EBA guidelines might be relevant, especially due to a few indications on how to treat the sale of bad loans in terms of equity. Hence, it would be apt to provide for adjustments in order to sterilise the effect of sales on the risk parameters: There is but one general reference to this aspect in the document drawn up by EBA. Given the potential impact of the guidelines on risk-weighted capital and assets, Prometeia has also flagged to the authority the possibility of an impact analysis on the effects of the proposed changes.
In addition to that, here is a summary of some of the most relevant points addressed by Prometeia in its feedback to the EBA consultation paper on PD estimation, LGD estimation and the treatment of defaulted assets:
- As PDs must be long-run average, the reference to benchmarking on most recent data might be misleading. Providing that effective default definition changed over time, thus adjustments are required and this shouldn’t lead automatically neither to MoC nor to calibrate on shorter and most recent time series
- The requirement for specific treatments of short term exposure should be oriented in the sense that all defaults might be considered, but no specific underweighting of them for 1-year PD calibration is necessary
- Treatment of multiple default should be excluded, as this is already addressed within default definition; no specific treatment for LGD is required as it would make the definition adopted inherently different from PD definition, but on the other hand sample definition criteria shall guarantee that all defaults considered in PD estimates are also considered in LGD estimates
- Representativeness of information and the “no data exclusion” provisions for LGD might be in contradiction in some circumstances: some data exclusion should be allowed for sake of representativeness
- LGD and LGD in-default treatment of interests and fees is very questionable from the economic point of view, especially as risks of negative LGDs depending on differences among discount and accrual rates are already addressed by a LGD zero-floor
- LGD and LGD in-default data requirements are meaningful and grounded, but a significant time will be required by a series of institutions before being able to fully cover gaps
- Some LGD and LGD in-default guidelines provisions are targeted to a specific methodology for sound inclusion of model sensitivity to collateral coverage, namely secured vs. unsecured modelling; as other methodologies are available and advisable in some circumstances, some changes are suggested in order to make provisions more generally applicable
- The previous points are connected, so that data availability and gaps might orient the best available methodology;
- The adoption of a undifferentiated standard spread for discounting recovery cash-flows is deemed to be really very simplistic; guidelines on sound methodologies would be most appropriate and only subordinately an appropriate differentiation of regulatory spreads
- The identification of positions ‘treated as closed’ for LGD and LGD in-default estimations should be not only based on time but other characteristics might be considered (for instance based on existing supporting collaterals)
- Inferencial techniques for other incomplete workouts should be linked to LGD in-default/Expected Loss best estimate (ELbe) definition; the regulation might be reinforced in the sense of consistent and more specific as possibile methodologies to be adopted
- The admission of ELbe override, within a dedicated process, will allow institutions to cope with quantification specificities of recovery expectations within most complex restructuring operations; this is very appreciated but within a controlled and ring-fenced dedicated process of ELbe validation/override; at this regard guidelines should be reinforced
- The guidelines related to ELbe quantification and downturn should be strongly reinforced, as it is not clear whether it is expected all to be jointly grounded on macro-economic variables and satellite models, in a kind of baseline/adverse stress testing framework. This is advisable for reconciliation of IFRS9 treatment as well with rather limited specificities (for instance discounting rate). A high degree of convergence not only is an available efforts-optimising option available to institutions, but would be very beneficial from the supervisory perspective as well
- Model segmentation and business segmentation criteria are connected but not strictly aligned, thus exposure under the same models are expected to be treated similarly but not necessarily homogeneously
- The impact of human judgement on estimates should be assessed at all levels (qualitative information, input override, output override), not output overrides only