Loss Given Default
Loss Given Default (LGD) captures the uncertainty about the actual loss that will be realized given a Credit Event. It is thus complementary to Recovery Risk, the possibility that in case of default the recovered amount may be less than expected. In economic terms, the LGD metric arises from the decomposition of Expected Loss into its constituents risks, namely the Probability of Default, Exposure at Default and Loss given Default.
Usage in Banking Regulation
LGD is one of the Risk Parameters used for establishing capital requirements for Credit Risk in the Basel II (and subsequent) regulatory frameworks. A conservatively estimated LGD measure (Downturn LGD) is used as input to the ASRF model.
Usage in IFRS 9 / CECL Accounting
Many models used in the context of IFRS 9 or CECL involve an expected credit loss decomposition that requires the estimation of LGD conditional on a range of possible future economic scenarios
Relationship with NPL valuation
Loss given default is related to but distinct from the assessment of a Non-Performing Loan. A key conceptual difference is that LGD is estimated ex-ante (before any credit default event), whereas NPL valuation or risk management is performed ex-post. The default event itself and subsequent events may contain information that must be incorporated into the NPL analysis. Further difference may emerge from the different regulatory or accounting requirements for performing versus non-performing exposures.
Issues and Challenges
- As with all quantitative estimates of risk, there is potential for significant Model Risk
- Certain type of credit exposures (Low Default Portfolios) may pose particular challenges to the estimation of LGD
- An LGD estimate is sensitive to the Default Definition which can substantially change the implied Cure Rate
- An LGD estimate is sensitive to the duration of the workout and the possibility of time censoring of observations
- LGD and Probability of Default may exhibit significant correlation, thus necessitating more complex joint dynamic models