Private Credit
Deep Generative Models for Private Credit SPV Analytics
A hierarchical framework for modeling Special Purpose Vehicles (SPVs) that securitize loan portfolios across multiple asset classes.
Key Features
| Component |
Architecture |
Purpose |
| Macro VAE |
Conditional LSTM-VAE |
Generate correlated macro scenarios |
| Transition Transformer |
Transformer Encoder |
Cohort-level state dynamics |
| Loan Trajectory Model |
AR Decoder + Diffusion |
Individual loan path generation |
| Portfolio Aggregator |
Differentiable Waterfall |
Loss distribution & tranche returns |
Quick Start
1
| pip install privatecredit
|
1
2
3
4
5
6
7
8
9
10
11
| from privatecredit.data import LoanTapeGenerator, MacroScenarioGenerator
from privatecredit.models import MacroVAE, PortfolioAggregator
# Generate synthetic portfolio
generator = LoanTapeGenerator(n_loans=10000)
loans, panel = generator.generate()
# Generate macro scenarios
macro_gen = MacroScenarioGenerator()
baseline = macro_gen.generate_scenario('baseline')
adverse = macro_gen.generate_scenario('adverse')
|
Architecture Overview
1
2
3
4
5
6
7
8
9
10
11
| Level 1: MACRO SCENARIO GENERATOR (VAE)
|-- GDP, Unemployment, Credit Spreads
v
Level 2: TRANSITION TRANSFORMER
|-- Cohort-level transition matrices
v
Level 3: LOAN TRAJECTORY MODEL
|-- Individual loan state & payment sequences
v
Level 4: PORTFOLIO AGGREGATOR
|-- Waterfall, VaR, CVaR, Tranche Returns
|
Asset Classes Supported
- Corporate Loans: SME, mid-market, large corporate
- Consumer Credit: Personal loans, auto loans
- Real Estate: Commercial and residential mortgages
- Trade Receivables: Invoice financing, factoring
Use Cases
| Application |
Output |
| Pricing |
Fair value of SPV tranches |
| Risk Management |
VaR, CVaR, capital allocation |
| Regulatory |
IFRS 9 ECL, Basel IRB, Solvency II |
| Investment |
Risk-adjusted returns by tranche |
Documentation
Citation
1
2
3
4
5
6
| @software{privatecredit2026,
title = {Private Credit: Deep Generative Models for SPV Analytics},
author = {Digital Finance Research},
year = {2026},
url = {https://github.com/Digital-AI-Finance/private-credit}
}
|
License
MIT License - see LICENSE
View on GitHub