LGD Estimation: Advanced Methodologies

1. LGD Components You Need to Model Explicitly

• Exposure at Default (EAD) net of eligible guarantees.
• Recovery cash flow curve separated into collateral, legal, and negotiated payments.
• Cost structure including legal expenses, servicing fees, and collateral liquidation discounts.
• Timing assumptions converted into discount factors via effective interest rate.

Institutions should track realized LGD per segment to calibrate model priors and prove representativeness of historical workouts.

2. Data Foundations and Workout Taxonomies

Reliable LGD requires granular treatment statuses (judicial recovery, voluntary restructuring, collateral sale, write-off) and timestamps for each action. Map data lineage from collection systems, custodians, and collateral management platforms to guarantee auditability.

Segregate secured versus unsecured exposures, differentiate collateral types, and capture macro indicators (housing price index, used car prices) to support downturn adjustments.

3. Modeling Approaches and Hybrid Stacks

1. Workout LGD models simulate expected recoveries and costs over time using survival curves, often preferred for wholesale portfolios with limited observations.
2. Direct regression models predict LGD percentages using elastic net or gradient boosting with borrower, product, and collateral features.
3. Two-stage frameworks model cure probability first and LGD conditional on cure or non-cure scenarios separately.
4. Bayesian models shrink estimates toward portfolio averages, stabilizing SME or project finance segments.

Champion/challenger comparisons should include predictive accuracy, interpretability, and implementation complexity criteria.

4. Downturn LGD and Stress Overlays

Basel and local supervisors require stressed LGD measures reflecting adverse macroeconomic periods. Recommended practices include:

• Identify historical downturn windows relevant for each collateral type.
• Apply haircut factors derived from regression of recovery rates versus macro indicators.
• Use scenario analysis (mild, moderate, severe) aligned with ICAAP or stress testing routines.
• Document how overlays revert back when macro conditions normalize.

5. Timing of Recoveries and Discounting Choices

LGD is sensitive not only to how much you recover but when you recover it. Build recovery timing curves using Kaplan-Meier estimators or parametric distributions, and discount cash flows using effective interest rate or funding costs depending on the accounting standard applied.

For legal recoveries with high variance, stress the tail of the distribution or cap recovery periods to avoid optimistic LGD.