ISDA SIMM

Python implementation of ISDA Standard Initial Margin Model (ISDA SIMM) v2.3-2.6 for calculating initial margin of uncleared OTC derivatives, with AADC integration for automatic sensitivity computation and gradient-based margin optimization.

Based on the official ISDA SIMM Methodology.

Features

• SIMM Calculation: Full ISDA SIMM v2.6 aggregation (Delta, Vega, Curvature, BaseCorr)
• AADC Integration: 30-50x speedup via automatic adjoint differentiation
• Trade Allocation Optimization: Gradient-based netting set optimization (10-30% IM reduction)
• Stress Margin Analysis: 7 predefined stress scenarios
• Pre-Trade Analytics: Marginal IM, bilateral vs cleared comparison, counterparty routing
• What-If Analysis: Real-time margin impact for add/remove/hedge scenarios
• Multi-Asset: IR swaps, FX options, equity options, inflation swaps, cross-currency swaps

Getting Started

Prerequisites

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

For GPU benchmarks:

pip install -r requirements_gpu.txt

Basic SIMM Calculation

Place your CRIF file in the CRIF/ directory and run:

python -m model.simm_baseline --trades 100 --threads 4

AADC-Enabled Calculation

python -m model.simm_aadc --trades 1000 --threads 8

Portfolio Optimization

python -m model.simm_portfolio_aadc \
    --trades 100 --portfolios 5 --threads 8 \
    --optimize --method gradient_descent

Benchmark

python benchmark_simm.py --trades 1000 --threads 8

Project Structure

ISDA-SIMM/
├── model/                  # Trade models and SIMM implementations
│   ├── simm_baseline.py    # NumPy baseline (bump-and-revalue)
│   ├── simm_aadc.py        # AADC-enabled implementation
│   ├── simm_common.py      # Shared SIMM logic
│   ├── simm_portfolio_aadc.py      # Portfolio-level AADC calculations
│   ├── simm_allocation_optimizer.py # Trade allocation optimization
│   └── trade_types.py      # Trade definitions and pricing
├── src/                    # Core SIMM calculation engine
│   ├── agg_margins.py      # Top-level SIMM aggregation
│   ├── margin_risk_class.py # Risk class margin calculations
│   ├── agg_sensitivities.py # Sensitivity aggregation
│   └── wnc.py              # Weights and correlations loader
├── common/                 # Configuration, logging, utilities
├── benchmark/              # Benchmarking framework (CPU vs GPU)
├── tests/                  # Test suite
├── scripts/                # Benchmark and utility scripts
├── Weights_and_Corr/       # SIMM calibration parameters (v2.3-v2.7)
├── CRIF/                   # Input CRIF data
├── visualization/          # Interactive optimization demo
├── docs/                   # Technical documentation
├── data/                   # Output data (execution logs, CRIF samples)
└── benchmark_simm.py       # Main benchmark runner

Performance

Implementation	1000 Trades	Speedup
Baseline (NumPy)	~45s	1.0x
AADC Python	~2s	~20x

Results Example

SIMM Total	Product Class	Risk Class	Risk Measure	SIMM
16,111,268,937	RatesFX	Rates	Delta	759,108,218
			Curvature	1,947
		FX	Delta	11,926,343
	Credit	CreditQ	Delta	3,922,360,448
		CreditNonQ	Delta	11,472,297,989
	Equity	Equity	Delta	158,843,566
	Commodity	Commodity	Delta	171,187,064

Documentation

• Architecture & Technical Details
• SIMM Implementation with AADC (Practitioner's Guide)

License

If you intend to implement this for any commercial purpose, reach out to ISDA SIMM to obtain a proper license. See ISDA SIMM Licensing FAQ.