Understanding Loss given default (LGD) in Risk Analysis

1. Why LGD matters for the target audience

LGD is one of the three pillars of the ECL calculation (PD × LGD × EAD). For institutions and companies preparing IFRS 9-compliant models, LGD drives the magnitude of lifetime or 12-month loss allowances, materially affecting reported provisions, capital management and profitability. Small LGD shifts can create large provision swings — for a portfolio with EAD = 100m and PD = 2%, a 10 percentage point change in LGD (e.g., 30% to 40%) increases ECL by 2m (100m × 0.02 × 0.10).

Beyond balance-sheet impact, LGD estimation affects pricing, portfolio strategy and stress testing. Boards, Audit and Risk Committees rely on clear LGD evidence in Risk Committee Reports to accept model outcomes and remedial actions.

2. Core concept: definition, components, clear examples

What is Loss given default (LGD)?

LGD is the percentage of exposure that is not expected to be recovered following a borrower default, after accounting for recoveries, collateral, costs and the time value of money (discounting). Under IFRS 9, LGD should reflect current and forward‑looking information and be consistent with how default is defined in PD.

Key components of LGD

• Gross default loss: initial shortfall between outstanding exposure and immediate recoverable value.
• Recoveries: cash inflows from collateral enforcement, guarantees, secondary sales and cure rates.
• Costs of recovery: legal fees, enforcement costs and collection expenses.
• Timing and discounting: present value adjustments for recovery timing (e.g., recoveries received over 3 years should be discounted to default date).
• Segmentation: LGD varies by product, collateral type, borrower segment and jurisdiction — e.g., collateralized mortgage vs unsecured consumer loan.

Illustrative numeric example

Example: Unsecured corporate loan of EAD = 1,000,000. Estimated recoveries over 3 years: year1 = 100,000, year2 = 50,000, year3 = 30,000. Recovery costs = 20,000. Discount rate = 5% per annum.

1. Nominal recoveries = 180,000. Present value of recoveries ≈ 100,000 + 50,000/1.05 + 30,000/1.05^2 ≈ 100,000 + 47,619 + 27,210 = 174,829.
2. Net recoveries after costs = 174,829 – 20,000 = 154,829.
3. LGD = (EAD – net recoveries) / EAD = (1,000,000 – 154,829)/1,000,000 = 0.845 ≈ 84.5%.

This LGD feeds into ECL: if PD = 10% for lifetime horizon, ECL = 1,000,000 × 0.10 × 0.845 ≈ 84,500.

3. Practical use cases and scenarios

3.1 Three‑Stage Classification and LGD application

In IFRS 9, staging determines whether you use 12-month or lifetime LGD assessments. For Stage 1 exposures (no significant increase in credit risk) ECL uses 12-month PD but LGD should reflect the loss given default within the 12‑month horizon (including recoveries expected within that timeframe). For Stage 2 and Stage 3, LGD must be lifetime and consider longer recovery patterns and cure probabilities. Be explicit in documentation how LGD horizons map to the three-stage classification.

3.2 Historical Data and Calibration

Use several years of default and recovery histories to estimate unconditional recovery rates and cure behavior by segment. Calibrate forward-looking adjustments separately: start with a historical baseline LGD and then apply macroeconomic overlays or scenario weights. When historical default volumes are low, consider pooled segments or external data sources and document judgmental overlays clearly.

3.3 Incorporating collateral and guarantees

Collateral valuation policies significantly affect LGD. For mortgages, haircuts, expected forced-sale discounts and time-to-sale are critical inputs. For complex guarantees, map legal enforceability and expected recovery timing. Document procedures for valuation refresh frequency and foreclosure timelines in your model validation pack.

3.4 Machine learning and LGD estimation

Where granular data exist, consider advanced approaches such as survival models or tree-based regressions to model recovery timing and amounts. If you explore automated techniques, integrate robust governance: maintain explainability, run out-of-time tests and compare against benchmarked parametric alternatives. For examples and guidance on automated approaches, see resources on machine learning for LGD.

3.5 Non‑financial corporates and special considerations

Non-financial corporates often have cyclical recovery rates tied to commodity prices, foreign exchange exposures and legal environments. Tailor segmentation and stress scenarios, and coordinate with credit officers familiar with sector-specific recovery pathways. See our guidance for broader treatment of these borrowers in ECL for non‑financial corporates.

4. Impact on decisions, performance and accounting

Accounting impact on profitability

Higher LGD increases ECL provisions, reducing profit before tax. For example, a bank with average performing exposure of 500m and blended PD of 1.5%: a 5 percentage point LGD increase adds 500m × 0.015 × 0.05 = 375k to provisions. Over multiple quarters, this cascades into capital planning, dividend capacity and pricing. Stress testing LGD sensitivities helps management quantify these effects and report them to the board.

Operational and strategic impact

• Pricing and origination: LGD expectations should feed into loan pricing tools to ensure expected returns compensate for anticipated losses.
• Portfolio management: identify high-LGD segments for de‑risking or stricter collateral requirements.
• Regulatory and capital planning: LGD assumptions inform internal capital adequacy assessments and regulatory conversations.

Reporting to stakeholders

Risk Committee Reports should include segmented LGD trends, sensitivity testing results, recent recoveries vs expectations, and the accounting impact on provisions. A concise table that contrasts baseline LGD, stressed LGD, and ECL impact by portfolio helps non-technical board members understand the drivers.

5. Common mistakes and how to avoid them

• Mistake: Using raw historical recovery rates without discounting or cost adjustments.
  Avoid by: Present-valuing expected recoveries, deducting realistic recovery costs, and documenting assumptions.
• Mistake: Ignoring timing differences between PD and LGD horizons (12-month vs lifetime).
  Avoid by: Mapping LGD horizons to each IFRS 9 stage and applying consistent segmentation.
• Mistake: Overfitting models on limited default data.
  Avoid by: Pooling segments, using parsimonious models, and applying conservative overlays when data are thin.
• Mistake: Not performing Sensitivity Testing or scenario analysis.
  Avoid by: Running parameter shocks (±10–30% LGD), macro scenario overlays, and presenting impacts in Risk Committee Reports.
• Mistake: Poor documentation and weak model governance.
  Avoid by: Preparing model validation packs with back‑testing, out‑of‑time tests, and a formal model change log.

6. Practical, actionable tips and checklists

Estimated LGD model checklist

• Define default, cure and recovery events consistently with PD and collections data.
• Segment by product, collateral and jurisdiction; justify any pooling decisions.
• Build baseline LGD from historical recoveries; document data windows and exclusions.
• Include recovery costs and discounting to default date; state discount rate used (e.g., effective interest rate or market rate).
• Apply forward‑looking adjustments linked to explicit macroeconomic scenarios; provide weighting rationale.
• Perform Sensitivity Testing: shock LGD inputs by at least ±10% and run at least three macro scenarios.
• Integrate Model Validation: independent reviewer, benchmark models, and back‑testing against realised recoveries.
• Prepare Risk Committee Reports: charted trends, material changes, and quantified ECL impacts.

Template: quick LGD sensitivity test

For a portfolio: EAD = 200m, PD = 1.2%, baseline LGD = 35%.

• Baseline ECL = 200m × 0.012 × 0.35 = 840,000.
• LGD +10% = 38.5% → ECL = 200m × 0.012 × 0.385 = 924,000 (increase 84k).
• LGD -10% = 31.5% → ECL = 200m × 0.012 × 0.315 = 756,000 (decrease 84k).

Include these sensitivity rows in the quarterly Risk Committee Reports to show P&L volatility under plausible LGD shifts.

KPIs / success metrics

• Model accuracy: RMSE or MAE of predicted vs realised recovery amounts over rolling 3–5 year windows.
• Calibration drift: percentage change in LGD relative to historical baseline (target ±5% annually unless explained by scenario).
• Provision sensitivity: ECL change per 1 percentage point LGD shift (GBP/EUR/USD per pp).
• Time-to-recovery median: measured in months/years by segment (target improvement with process initiatives).
• Governance: time between model update requests and implemented changes (target <90 days).
• Disclosure completeness: proportion of required IFRS 9 LGD disclosure fields populated and approved (target 100%).

FAQ

Reference pillar article

This article is part of a content cluster supporting our pillar guide: The Ultimate Guide: The basic equation for calculating ECL – explanation of PD, LGD, and EAD, how the formula is applied in practice, and a simple illustrative example. For a broad overview of the ECL concept see Expected credit losses (ECL.

Model Validation and governance (Model Validation & Sensitivity Testing)

Robust validation involves independent review, back-testing against realised recoveries, stability tests and benchmarking. Key steps:

1. Data quality checks: completeness, correct default labelling, and consistent exposure measures (EAD).
2. Performance testing: out‑of‑time tests, decile lift charts for recovery probability models, and error metrics for amount predictions.
3. Sensitivity Testing: systematic shocks to LGD inputs, scenario overlays, and stress cases aligned with ICAAP/ILAAP exercises.
4. Documentation: model specification, assumptions, limitations, and change management logs for the Model Validation file.

Present validation findings and remedial actions in the Risk Committee Reports quarterly to maintain board oversight and audit readiness.