Recovery Modeling for Defaulted Exposures

1. The Recovery Process: From Default to Resolution

Recovery modeling estimates cash flows realizable from defaulted exposures through workout strategies—voluntary settlements, collateral liquidation, bankruptcy claims, or debt forgiveness. Accurate recovery projections drive IFRS 9 LGD parameters, NPL provisioning, and capital allocation for workout teams.

Unlike performing loan modeling (focused on default probability), recovery models handle post-default uncertainty: borrower cooperation, collateral condition, legal system efficiency, and macroeconomic factors affecting asset values and resolution timelines.

2. Recovery Channels and Strategies

• Voluntary settlement: Negotiate repayment plan or lump-sum discount with borrower—fastest, least costly option when borrower has capacity.
• Collateral foreclosure: Repossess and sell pledged assets (real estate, vehicles, inventory). Timeline: 6-36 months depending on jurisdiction and asset liquidity.
• Bankruptcy/insolvency proceedings: File claims in court-supervised liquidation or reorganization. Recovery rate depends on creditor priority (senior secured vs. unsecured) and asset pool adequacy.
• Third-party collection: Outsource to specialized agencies for commission (15-35% of recoveries)—economic for low-balance, high-volume portfolios.
• Portfolio sale: Sell NPL to distressed debt funds at discount (see NPL Valuation article)—monetizes immediately but forgoes potential upside.
• Debt forgiveness / write-off: Cease collection after cost-benefit analysis shows negative expected recovery (legal costs exceed recoverable amount).

3. Components of Recovery Rate Estimation

Loss Given Default (LGD): LGD = 1 - Recovery Rate = (EAD - Recoveries) / EAD

Recovery rate depends on:

• Collateral realization value (CRV): Market value at liquidation minus forced sale discount (10-40%), haircut for illiquidity, and legal costs.
• Cure / Restructuring proceeds: Present value of renegotiated payments (principal reduction, maturity extension, rate cut).
• Unsecured recovery: Borrower asset liquidation in bankruptcy, garnishments, voluntary payments—typically 0-20% of exposure.
• Time to resolution: Longer workout periods erode NPV of recoveries; discount cash flows at risk-adjusted rate (12-18%).
• Direct costs: Legal fees, court costs, property maintenance, appraisals, servicer fees—deduct from gross recovery.

4. Collateral-Based Recovery Models

Secured exposure formula:

Recovery = min(EAD, CRV × (1 - Haircut) - Legal Costs)

• Current market value (CMV): Updated collateral appraisal at default date (not origination).
• Forced sale discount: Auction/distressed sale prices 10-30% below retail—higher for specialized assets (industrial equipment, unique properties).
• Haircut factors:
  • Residential real estate: 15-25% (liquid markets), 30-40% (illiquid/rural).
  • Automobiles: 20-30% (depreciation + auction fees).
  • Equipment / inventory: 40-60% (obsolescence, condition deterioration).
• Legal / administrative costs: Foreclosure: $5K-$20K per property. Repossession: $500-$2K per vehicle.

LTV threshold effects: High LTV exposures (>80%) more likely to generate shortfalls; model recovery rate as function of LTV at default.

5. Unsecured Exposure Recovery Models

Without collateral, recovery depends on borrower solvency and collection intensity:

• Cure rates: Historical % of defaulted unsecured accounts returning to performing status—varies by DPD bucket and borrower segment.
• Bankruptcy recovery: Priority waterfall: senior secured → priority unsecured (taxes, wages) → general unsecured. Typical recovery 5-15% for general unsecured creditors.
• Roll rate models: Predict migration from early delinquency to charge-off; accounts reaching 180 DPD rarely recover (>90% loss rate).
• Income garnishment: Court-ordered wage deduction (up to 30% disposable income)—effective when borrower employed but requires ongoing monitoring.

6. Regression-Based LGD Models

Statistical models estimate LGD as function of risk drivers:

Dependent variable: Realized LGD from historical defaults (recovered amount / EAD at default).

Explanatory variables:

• Loan characteristics: LTV, seniority, collateral type, remaining term.
• Borrower attributes: Income, employment status, prior defaults, age.
• Macroeconomic conditions: GDP growth, unemployment rate, house price index at default date.
• Workout factors: Time to resolution, foreclosure vs. settlement, legal costs incurred.

Model forms:

• Linear regression: Direct LGD prediction (bounded 0-100% post-hoc).
• Beta regression: Natural fit for LGD as rate variable in (0,1) interval.
• Two-stage model: Binary classifier (will recover >0?), then regression on recovery amount given recovery.

7. Downturn LGD Adjustments

Basel IRB and IFRS 9 require downturn LGD reflecting stress conditions:

• Historical downturn identification: Periods with default rates ≥1.5x long-run average—typically recessions (2008-2009, 2020 pandemic).
• Collateral value stress: Housing prices decline 20-30% from peak during crises; apply haircuts to CMV in base LGD model.
• Time to resolution extension: Courts backlogged, buyers scarce—foreclosure timelines double, increasing holding costs and NPV discount.
• Cure rate compression: Fewer borrowers regain employment or refinance during downturns—reduce voluntary settlement assumptions.
• Regulatory add-ons: Basel conservatism margin or supervisory floor (e.g., LGD floor of 10% for senior secured, 25% for unsecured).

8. Workout Optimization and Resource Allocation

Maximize portfolio recoveries by prioritizing high-value cases:

• Expected recovery scoring: Rank defaulted accounts by predicted net recovery (gross recovery - costs). Focus legal resources on top quintile.
• Collection cost curves: Model marginal cost of additional collection effort vs. incremental recovery—cease when cost > benefit.
• Early resolution incentives: Offer settlement discounts (20-40% principal forgiveness) to borrowers willing to pay immediately—improves NPV despite haircut.
• Servicer performance monitoring: Track servicer resolution rates, timelines, cost per account—replace underperformers or renegotiate fees.

9. Integrating Recovery Models into IFRS 9

• Stage 3 ECL calculation: LGD models directly feed Stage 3 provision—defaulted exposures provisioned at EAD × LGD.
• Cure provisions: Accounts exiting Stage 3 (return to performing) require unwinding provisions—validate cure rates against model assumptions.
• Forward-looking adjustments: Overlay macroeconomic scenarios onto collateral values and recovery timelines—pessimistic scenarios increase LGD.
• Post-model adjustments (PMA): Management overlays for unprecedented events (court shutdowns during pandemic, regulatory forbearance programs).