Loan Portfolio Risk & Collection Analytics

Overview

The framework studies early warning indicators by benchmarking defaulting borrowers against five healthy peers with similar origination profiles. It compares trajectories from eleven months before default through the default month, aiming to distinguish leading behavioral changes from coincident delinquency signals.

Methodology

• Filter healthy comparison borrowers to those active through the defaulting borrower's matching month on book.
• Standardize origination credit score, debt-to-income, revolving utilization and loan amount; select five nearest healthy peers.
• Compare six monthly measures against the median trajectory of each matched peer group and run matching sensitivity diagnostics.

Results & Findings

• The README reports 6,580 defaulters, 32,900 matched-control rows, and 61,808 borrower-month trajectory records.
• At T-3, the median credit-score velocity gap is -3 points and the revolving-utilization velocity gap is +0.006 (+0.6%).
• At T-0, the reported delinquency gap reaches 90 DPD and payment-ratio gap is -1. These are observational findings, not causal estimates.

Project Structure

• data/: origination and monthly performance inputs, matched controls, trajectory tables and matching diagnostics.
• scripts/: ingestion checks, matched-control construction, validation, sensitivity analysis and trajectory reporting.

Getting Started

The README documents Python 3.10 or later. Run the scripts sequentially from the repository root after installing requirements.