Institutions should validate all LGD models according to the validation principles articulated in the MMS. Both qualitative and quantitative assessments should be conducted for an appropriate validation.
4.7.2
Institutions should ensure that segment-level LGD values represent the risk profile of their books. Statistical tests alone are not sufficient to conduct appropriate validation of LGD at segment level. Consequently, institutions should combine statistical tests, judgement and checks across comparable metrics to ensure that the calibration of these metrics are meaningful and accurate.
4.7.3
The scope of the validation should be comprehensive. If the validation is performed by a third party consultant, institutions should ensure that the validation effort is comprehensive in scope and substance.
4.7.4
The validation scope should include, at a minimum, the following components:
(i)
The data quality, comprehensiveness and collection process. This should cover the analysis of unusual features observed in historical data and their subsequent treatment for modelling.
(ii)
The definition of default. This should cover the treatment of technical defaults and restructured accounts.
(iii)
The methodology employed to compute historical LGD. This should cover in particular the possible outcomes as described earlier in this section. Partial replication of the historical LGD for a sample of facilities should be performed.
(iv)
The methodology employed to estimate TTC LGD and subsequent PIT adjustments. This should cover the model fit, functional form, robustness, properties and sensitivities.
(v)
The treatment of collateral. The treatment of LGD segmentation granularity. The quality of the model output in terms of economic and business meaning. This can rely on comparables based on data available outside of the institution.
(vi)
The existence of spurious accuracy and excessive generalisation. In particular, the validation process should report the excessive usage of average LGD employed across a large population of heterogeneous obligors.
(vii)
Back-testing of modelled LGD, estimated separately for defaulted and non-defaulted obligors.