Research
Academic foundations and methodology.
Overview
Private Credit implements a hierarchical deep generative framework for modeling structured credit portfolios. The approach addresses key challenges in credit risk modeling:
- Complex correlation structures across macro, cohort, and loan levels
- Non-stationary dynamics driven by macroeconomic conditions
- Path-dependent outcomes (state transitions, prepayment, default)
- Fat-tailed loss distributions requiring accurate tail estimation
Key Contributions
1. Hierarchical Generation
Unlike flat approaches that model loans independently, our framework captures:
| Level | Component | Captures |
|---|---|---|
| Macro | Conditional VAE | Systematic risk, scenario conditioning |
| Cohort | Transformer | Vintage effects, asset class dynamics |
| Loan | AR + Diffusion | Idiosyncratic risk, path dependence |
| Portfolio | Differentiable Waterfall | Aggregation, tranche structuring |
2. Conditional Scenario Generation
The Macro VAE enables:
- Scenario interpolation: Generate paths between baseline and stress
- Conditional generation: Fix endpoints, generate consistent paths
- Reverse stress testing: Find scenarios producing target losses
3. End-to-End Differentiability
The entire pipeline is differentiable, enabling:
- Joint training with portfolio-level objectives
- Gradient-based calibration to historical data
- Sensitivity analysis via automatic differentiation
Methodology
Full methodology documentation
Publications
Coming soon
References
Research Team
- Digital Finance Research Group
- FHGR - University of Applied Sciences of the Grisons