Credit Risk Assessment & Customer Segmentation

Consumer lenders face a persistent trade-off between growth and default risk. In practice, credit decisions often over-index on credit scores, implicitly assuming strong credit history compensates for weak affordability.

Why it mattered: Misclassifying risk leads to higher defaults, inefficient capital allocation, and poor risk-adjusted returns. The business required a transparent, regulator-friendly framework that distinguishes true affordability risk from credit quality and can be operationalized across underwriting, pricing, and monitoring.

Is the customer financially able to repay the loan, and how should the portfolio be segmented to improve risk-adjusted returns without constraining growth?

• Only pre-origination variables could be used
• Default outcomes are structurally imbalanced
• Decisions must remain explainable for regulatory review
• Portfolio impact matters more than pure predictive accuracy

Hypothesis: Affordability (DTI) is the primary driver of default risk and should act as a hard approval gate, while credit score should be used for differentiation only among affordable borrowers.

1. Filtered to completed-outcome loans to avoid default bias
2. Engineered affordability, credit quality, and repayment history signals
3. Quantified default risk by DTI, FICO, and their interaction
4. Designed a hierarchical, rule-based risk segmentation aligned with credit policy logic
5. Validated monotonic risk separation and complete portfolio coverage
6. Operationalized insights through an executive Power BI dashboard

Methods: Exploratory default rate analysis, DTI and FICO banding, Interaction analysis (DTI × FICO), Rule-based risk segmentation, Portfolio exposure and concentration analysis

Metrics: Default Rate, Debt-to-Income Ratio (DTI), FICO Score, Loan Exposure, Expected Loss Proxy

Data: 1.88M completed LendingClub loans (2007–2015), filtered to pre-origination variables only.

• Borrowers with DTI > 40% exhibit ~3× higher default rates regardless of credit score.
• Credit score meaningfully differentiates risk only among affordable borrowers (DTI < 30%).
• Risk is concentrated rather than evenly distributed across the portfolio.
• The Medium Risk segment represents ~46% of lending volume and is the core performance driver.
• Pricing increases with risk but does not fully compensate for elevated default rates.

• Introduced a transparent, regulator-friendly risk segmentation framework
• Enabled selective intervention instead of blunt portfolio tightening
• Aligned underwriting, pricing, and monitoring around a single risk logic

• ~8–10% reduction in portfolio default rate via DTI > 40% cutoff
• $80–125M estimated annual financial improvement

• Used rule-based segmentation over black-box ML for interpretability
• Optimized for policy clarity rather than maximum predictive accuracy
• Focused on completed loans to ensure unbiased default estimation

• Historical data may not reflect current macroeconomic conditions
• Causal impact requires controlled experimentation
• Limited borrower behavioral and asset-level data

Delivered a deployable, economically defensible credit risk framework supported by an executive dashboard and a clear policy playbook.

• Run A/B tests on DTI thresholds to validate causal impact
• Integrate segmentation into approval and pricing workflows
• Extend monitoring with early-warning signals and macro indicators