Back to Models
RecoveryIQ™ - Loss Severity Prediction Engine
Predicts expected recovery rates and loss severity at time of default, enabling optimized collection strategies.
The Challenge
Not all defaults are equal. Predicting recovery rates determines optimal collection strategies, accurate loss reserves, and capital requirements.
What It Does
Predicts expected recovery rates and loss severity at time of default, enabling optimized collection strategies and accurate financial provisioning.
Model Metadata
- Model Architecture: Two-stage Random Forest (recovery/no-recovery, then severity)
- Training Data Size: 2M+ defaulted loans, 10+ years of recovery data
- Features: 200+ loan, collateral, borrower, and economic features
- Update Frequency: Semi-annual retraining
- Inference Speed: <100ms per account
- Output Range: 0-100% recovery rate, $ loss amount
Business Outcomes
- 92% accuracy in recovery rate predictions
- $5M+ improvement in reserve accuracy (reduced over-provisioning)
- 25% improvement in collection ROI through strategy optimization
- Accurate CECL provisioning
- Optimized legal action decisions (sue vs. settle)
Training Approach
Two-Stage ModelingFirst model predicts probability of any recovery. Second model predicts recovery amount conditional on recovery occurring. This handles bimodal distribution of recoveries (many zeros, continuous positive values).
Survival AnalysisCox proportional hazards models for time-to-recovery predictions, integrated with amount predictions.
Segment-Specific ModelsSeparate models for secured vs. unsecured, different collateral types, various loan products to capture unique recovery dynamics.
Data Sources
Loan Characteristics
- Original loan terms (amount, rate, maturity)
- Current balance and accrued interest
- Lien position and security agreements
- Loan-to-value ratios
Collateral Data
- Property valuations (AVM and appraisals)
- Vehicle values (KBB, NADA for auto loans)
- Collateral condition and depreciation
- Insurance and title status
Borrower Information
- Assets and income at origination and default
- Employment status and stability
- Other obligations and liens
- State of residence (legal environment)
Default Context
- Reason for default (job loss, medical, other)
- Time in default before charge-off
- Borrower cooperation level
- Bankruptcy filing status
Economic Factors
- Local economic conditions at recovery time
- Real estate market trends (for secured loans)
- Legal environment (foreclosure timelines, deficiency judgments)
Recovery History
- Internal collection outcomes
- Third-party collection agency performance
- Legal action results and costs
- Settlement negotiation patterns
Third-Party Services
- Black Knight (property valuations and foreclosure timelines)
- S&P Global Market Intelligence (economic forecasts)
- LexisNexis Asset Locator (skip tracing, asset discovery)
Regulatory Application
Critical for CECL (Current Expected Credit Loss) compliance, Basel III capital calculations, and financial statement accuracy. Auditor-approved methodology.