Composable risk layers starting from SIMM

Banks run risk and collateral via SIMM for the large part of their business conducted via ISDA CSAs.

SIMM isn’t used for ETD, but it is the most widely adopted starting point for collateral and clearing in the bilateral world. If you want institutional participants in a new clearing model, you start where they already are. Low friction wins.

SIMM’s applicability beyond bilateral swap portfolios requires add-ons.

Additional components, sensitivities, corrections for nonlinearity, regime behaviour, tail risk.

We call these layers. As you add more layers, the model converges toward full VaR. Some layers are universal, some only apply to individual asset classes. The architecture is modular by design.

One of the lessons of crypto, maybe even the goal of web3, is composability.

So why are risk systems in clearing houses not composable? They can be.

Start with SIMM for standardisation and low-friction onboarding. Layer asset-class-specific corrections on top. The base is the same for every participant, every product. The layers adapt to the risk profile.

This is a better model. Five levels of estimation, each correcting the one below, activating per portfolio only when needed. SIMM handles the majority of books on its own. The upper layers close the gap where it matters.

That composability is what makes universal clearing possible. Crypto options, perps, interest rate swaps, commodities, carbon.

Same engine, same base, different layers.

All delivered at high speed.

-Proposer-verifier

-MoE analogues

-IO-aware fused kernels

-Cache-centric risk calculations

-Quantisation of risk (sparse Hessians, covariance matrices, scenario dictionaries)

-Fourier Neural Operators

No MC in sight outside the testing systems.

Genuine revolution based on AI engineering techniques, but only those with path determinism.

If you are still reading, well done!

This first appeared on LinkedIn on 26 March 2026. If you want to comment or discuss, that’s the place.