Investigate Rising Loan Delinquency with Stable Approval Rates

# Investigate Rising Loan Delinquency with Stable Approval Rates A new loan cohort shows a higher rate of loans that are at least 30 days past due after three months, while the overall approval rate has changed little. Investigate the deterioration. Later, you learn that the approved loan mix shifted toward borrowers with lower FICO scores. A new merchant also prompts consideration of tighter underwriting policy. ### Clarifying Questions to Ask - Is delinquency measured by loan count or balance, and are cohorts compared at the same loan age with equivalent reporting completeness? - Which application, approval, score, loan-term and merchant information is available, and how is the policy change defined? ### Part 1 — Diagnose the cohort change Explain how stable approval rates can coexist with worse delinquency and how you would separate a composition change from deterioration within comparable borrower groups. #### What This Part Should Cover - Comparable vintage definitions and checks for reporting or servicing changes. - Decomposition by credit risk, merchant and loan characteristics, including changes in the applicant pool. ### Part 2 — Assess the underwriting model Given the shift toward lower-FICO borrowers, determine whether the existing underwriting model remains useful or needs replacement. #### What This Part Should Cover - Discrimination, calibration and observed outcomes on mature comparable cohorts. - The distinction between model failure, a policy threshold change and a riskier population with accurately predicted risk. ### Part 3 — Evaluate tighter merchant underwriting Describe how you would decide whether to tighten policy for the new merchant. #### What This Part Should Cover - Incremental loss reduction versus lost profitable approvals, with uncertainty and delayed outcomes. - A valid comparison strategy and constraints on learning about applicants who are declined. ### What a Strong Answer Covers Connect the diagnosis to an action: keep, recalibrate or replace the model, or change policy only where the evidence supports it. Avoid treating an unchanged approval rate as proof of unchanged risk. ### Follow-up Questions - What would you conclude if observed delinquency rose exactly as much as the model predicted? - How would selective observation of repayment outcomes limit an offline policy comparison?