Long-run Loss Given Default
Long-run Loss Given Default is the arithmetic average of realised LGDs over a historical observation period weighted by a number of defaults.
ECB TRIM Requirements
When performing this calculation, institutions should observe the following points.
- In the event of definition of default applied at obligor level, where two facilities of the same obligor are assigned to the same facility grade or
pool, two options are seen as compliant for calculating the average. The first is to compute the average weighted by the total number of facilities within that facility grade. The second is to first take the exposure-weighted average realised LGD at the obligor level and then take the arithmetic average LGD weighted by the number of defaulted obligors within the LGD grade. Institutions should demonstrate that the approach they use does not distort the actual observed loss.
- Under paragraph 160 of the EBA GL on PD and LGD, the realised LGD of each observation should be floored at zero for the purpose of LGD estimation. In cases where LGD estimates for specific facility grades or pools are low or even zero (in exceptional cases), in order to ensure that these estimates are accurate and not driven by (systematic) errors or distortions institutions should ensure that their estimation process is accurate. In particular, they should ensure that there is a sufficient number of observations supporting the estimate and that these outcomes are carefully monitored and scrutinised.
- Under paragraph 162 of the EBA GL on PD and LGD, institutions should apply an appropriate treatment to extremely high values of realised LGDs much above 100%, at the level of data quality, risk drivers, assignment to grades or pools or assignment to calibration segments. To ensure that the estimates are accurate, institutions are not expected to cap realised LGD values (i.e. to replace the observed value by a pre-defined value when the observed value is above the pre-defined one).
Institutions can calibrate LGD estimates to the LRA LGD calculated at the level of the Calibration Segment under paragraph 161(b) of the EBA GL on PD and LGD. When calibration segments are used for the purpose of LGD estimations, institutions are expected to base their decision on a sound rationale, in particular on quantitative evidence. It is the ECB’s understanding that, to comply with Article 181(1)(a) of the CRR, institutions should also calculate the LRA LGD at a more granular level than the calibration segment (i.e. individual LGD grades or pools if estimation is discrete or ranges of LGD grades if the estimation is continuous). The level should be appropriate for the application of the model. In addition, institutions should ensure that there are no systematic deviations when comparing the estimated LGDs with the LRA of realised LGDs at this more granular level, i.e. the direction of divergences should be random.
Combination of Different Components
Where the LGD is the result of a combination of different components (for example, secured and unsecured components), there is a risk that systematic deviations could be introduced to the estimation when combining these different components. In this case, the direction of divergences would not be random. To mitigate this risk, institutions should do the following.
- Provide evidence that the assumptions underlying the specification of the formula being used for the LGD estimation hold true to a sufficient extent in practice. Specifically, they should ensure that the different components are independent or, if dependency exits, that such dependency is adequately reflected in the LGD methodology.
- For defaults in the RDS which are closed or considered closed, compare the realised LGD at facility level with the estimates of LGD. Separate tests should be performed for the LGD applied to the performing portfolio and the LGD in-default. Analogous tests should be performed at component level.
- In the case of models based on components with underlying data covering time windows with different lengths and/or periods for each of the components, ensure that no bias is introduced in the LGD estimates at facility level with respect to the LRA. The analysis referred to in point (b) should be performed, at least, for the available common time period.
Recent Observations for Retail Exposures
For retail exposures institutions need not give equal importance to historical data if more recent data are a better predictor of loss rates. It is the ECB’s understanding that an institution may consider the more recent data to be a better predictor of loss rates and may give more importance to recent historical data if its methodology is in line with paragraphs 150 to 152 of the EBA GL on PD and LGD and if the following apply.
- There is a significant improvement in predictive power when using the more recent data with respect to the predictive power resulting from the use of an arithmetic average under paragraph 150 of the EBA GL on PD and LGD. This improvement would be evidenced by comparing the estimated LGDs for each grade with the average realised LGD covering as long a period as possible in accordance with Article 185(b) of the CRR.
- The oldest data are considered as non-representative as a result of specific policy or business changes in the bank, but not in order to reflect current trends in loss rates directly related to macroeconomic conditions.
- The weighting approach is used in a consistent manner over time and any change in the applied weights of historical data is appropriately justified.
In accordance with Article 179(1)(a) of the CRR, an institution’s own estimates must incorporate all relevant data and must be derived using both historical experience and empirical evidence. To comply with these requirements, when institutions use external or pooled data to complement their own loss or recovery experience, they should ensure that LRA LGDs derived from external or pooled data are also calculated separately from those based on internal data. In addition, the direction and magnitude of the differences between these averages should be properly analysed and documented when calibrating the model, including the adequacy of the MoC considered, and duly followed up in the review of estimates.
Variation in Lending Standards
Article 179(1)(d) of the CRR requires, among other things, that the population of exposures represented in the data used for estimation, the lending standards used when the data were generated and other relevant characteristics must be comparable with those of the institution's exposures and standards. Paragraph 164 of the EBA GL on PD and LGD further specifies that institutions should take into account not only the current characteristics of the portfolio but also, where relevant, any changes to the structure of the portfolio that are expected to happen in the foreseeable future. When institutions perform adjustments to their LGD estimates in order to comply with these requirements, it is the ECB’s understanding that the following principles should apply.
- The adjustment should be based on a comparison of the data used in risk quantification with the institution’s application portfolio. In many circumstances (for example where a type of product has been discontinued by the institution), the addition of these characteristics as risk drivers for LGD estimation is the most simple and effective way of dealing with issues of non-representativeness
- In the event of changes in lending or recovery policies, institutions should make only conservative adjustments until they are able to provide empirical evidence concerning the impact of the new policies. Such evidence should be based on the inclusion in the RDS of data from periods more recent than the change of policy
- All economic and market conditions experienced in the past and reflected in historical observations should be considered by institutions as part of foreseeable economic and market conditions (paragraph 147 of the EBA GL on PD and LGD). They are not, therefore, a reason to perform adjustments.
- ECB guide to internal models - Credit Risk, Sep 2018