Why Segmentation Matters
A "one-size-fits-all" model rarely works for IFRS 9. Different loan products have fundamentally
different risk drivers, behavioral patterns, and data availability. Segmentation ensures that
loans are grouped into homogeneous portfolios where the risk characteristics are consistent.
Key Principle:
Segmentation should be granular enough to capture distinct risk profiles but large enough to be statistically significant.
Retail: Residential Mortgages
Mortgages are typically low-frequency, low-severity default products due to the strong collateral backing.
Modeling Nuances
- PD Driver: LTV (Loan-to-Value) at origination, borrower income stability, and regional unemployment rates.
- LGD Driver: Heavily dependent on property value (HPI indexation), foreclosure costs, and time-to-sale. LGD is often very low (< 10%).
- EAD: Simple amortization schedule. Prepayment risk is a key factor affecting lifetime EAD.
- Staging: Often driven by payment arrears (DPD). Forbearance is a common Stage 2 trigger.
Retail: Credit Cards & Revolving
Unsecured revolving facilities characterized by high volume, low balance, and rapid behavioral changes.
Modeling Nuances
- PD Driver: Behavioral scores (FICO/b-score), utilization rate, and payment history. Very sensitive to short-term economic shocks.
- LGD Driver: Unsecured nature means low recovery rates (often just sale of debt to collection agencies). LGD typically > 60%.
- EAD: Complex due to undrawn commitments. Credit Conversion Factors (CCF) are critical to estimate how much of the limit will be drawn before default.
- Lifetime: Since there is no maturity date, a "behavioral lifetime" (e.g., 3-5 years) must be estimated.
Wholesale: Corporate Banking
Low volume, high value exposures. Often assessed individually rather than collectively.
Modeling Nuances
- PD Driver: Financial statement analysis (leverage, liquidity, profitability), industry sector risk, and qualitative management assessment. Often mapped to external agency ratings.
- LGD Driver: Specific to the seniority of the debt and the value of specific pledged assets (machinery, inventory, receivables).
- Data Challenges: Low default counts make statistical modeling difficult. Often relies on "Shadow Rating" models or external benchmarks.
Wholesale: SME (Small/Medium Ent.)
A hybrid between retail and corporate. The risk is often tied to the business owner's personal credit.
Modeling Nuances
- PD Driver: Blend of business financials (if available) and owner's personal credit bureau data. Account turnover is a strong behavioral signal.
- Segmentation: Often split into "Small SME" (modeled like Retail) and "Large SME" (modeled like Corporate).
- Sector Risk: Highly sensitive to sector-specific downturns (e.g., Hospitality, Retail).
Key Differences Summary
| Product |
Key PD Driver |
LGD Characteristic |
EAD Challenge |
| Mortgage |
LTV, Affordability |
Property Valuation |
Prepayment Risk |
| Credit Card |
Utilization, Behavior |
Low/Unsecured |
CCF (Undrawn) & Lifetime |
| Corporate |
Financial Ratios |
Seniority/Asset Specific |
Low Volume Data |
| SME |
Turnover, Owner Credit |
Mixed Collateral |
Data Quality |