A Python-based credit risk stress testing project implementing PD estimation, Basel Vasicek stress mapping, and facility-level Unexpected Loss (UL) analysis with visualization.
This project performs end-to-end credit risk analysis using obligor-level Probability of Default (PD) and facility-level Exposure at Default (EAD) data.
Main components include:
- PIT PD aggregation by rating
- TTC PD construction
- Basel 99.9% stressed PD (Vasicek single-factor model)
- Facility-level Unexpected Loss (UL) computation
- Rating-level EAD aggregation
- Stress visualization (time series, histograms, 3D surfaces)
Credit-Risk-Stress-Testing/
│
├── src/ # Core modeling & stress testing logic
├── docs/ # Sample output visualizations
├── data/ # Input datasets (ignored in repo)
├── outputs/ # Generated outputs (ignored in repo)
├── requirements.txt # Python dependencies
└── README.md
The project follows a multi-stage processing pipeline:
- Obligor-level PD aggregation
- TTC PD construction
- Basel Vasicek stress transformation
- Facility-level UL computation
- Rating-level aggregation
- Visualization layer
Obligor Panel
→ Merge with Facility Data
→ Stress Mapping (Vasicek)
→ UL Calculation
→ Rating Aggregation
→ Plot Generation
src/— Core computation & modeling logicdocs/— Sample visualization outputsoutputs/— Auto-generated results (ignored in repo)data/— Input datasets (excluded from version control)
Time-averaged PD per obligor: PD_TTC = average(PD_t)
PD_stress = Φ((Φ⁻¹(PD_TTC) + √ρ · Φ⁻¹(0.999)) / √(1−ρ))
Where:
- ρ = asset correlation
- 0.999 = Basel confidence level
- Φ = standard normal CDF
UL_i = EAD_i × LGD_DT × (PD_stress − PD_TTC)
LGD^DT is estimated from default events during stress periods (GFC / COVID).
- Python
- pandas
- numpy
- scipy
- matplotlib
git clone https://github.com/YuyangZ1/Credit-Risk-Stress-Testing.git
cd Credit-Risk-Stress-Testing
pip install -r requirements.txt
python3 src/main.py
