Loans modelled
DPD snapshots
Base ECL
PD model AUC
Overview
A reproducible banking-risk analytics suite built around a synthetic securitised auto-loan portfolio. It combines a Streamlit dashboard, SQL insights, baseline probability-of-default modelling, IFRS 9-style expected-credit-loss analysis, delinquency roll rates, vintage curves and scenario-based stress testing.
Features
- Interactive portfolio overview with balance, ECL, loss and collection KPIs
- DPD staging, delinquency roll-rate analysis and vintage loss curves
- Baseline PD model with stratified out-of-fold evaluation and calibration diagnostics
- IFRS 9-style PD × LGD × EAD expected-credit-loss calculations
- Illustrative base, recession, COVID and 2008 stress scenarios
- SQL Insights page with filtered portfolio queries and CSV export
Challenges
- Keeping risk calculations reproducible across dashboard filters and scenario assumptions.
- Avoiding target leakage by separating origination-time features from delinquency outcomes.
- Presenting technical credit-risk concepts in a decision-ready dashboard experience.
Results
The platform turns a synthetic loan portfolio into a complete risk-analysis workflow: early-warning delinquency views, baseline PD diagnostics, portfolio-level ECL and scenario comparison. It is designed as a portfolio and learning project, not a production credit-decision or regulatory model.
Screenshots
Architecture
The end-to-end flow from raw data to a shipped result.
Ingest
Loan, DPD, loss & vintage tables
Model
PD, staging, ECL & roll rates
Stress
Scenario-weighted portfolio impact
Decide
Streamlit, SQL & Power BI insights