A Guide to Accelerated Sensitivities for SIMM

The ISDA Standardized Initial Margin Model (SIMM) will become effective from September 2016 for computing the initial margin (IM) requirements of OTC deals. It follows a sensitivity-based approach (SBA), i.e. it involves computing sensitivities of the portfolio to the market risk factors (delta and vega Greeks) on a daily basis. These sensitivities allow estimating the P&L with a 10-day horizon efficiently, and subsequently the IM. This regulatory change also affects calculations of the Initial Margin Valuation Adjustment (MVA) where SIMM needs to be applied dynamically at many time-steps and scenarios within a Monte-Carlo simulation. Since collateral also needs to be calculated for counterparty credit risk, SIMM also affects CVA/DVA and regulatory capital simulations.

When traditional bump-and-revalue techniques (finite differences) are applied to calculate the sensitivities, i.e. the full valuation is re-run for each of the many market risk factors, the computational complexity is huge. However, ISDA's definition of the delta and vega sensitivities allows for infinitesimal shifts and is thus compatible with the Adjoint Algorithmic Differentiation (AAD) approach.

This white paper details how AAD can be applied for computing the SIMM sensitivities with a drastically lower computational complexity than bumping (finite differences). It further explains how SIMM can be calculated dynamically within MVA simulations – with high accuracy and speed. The performance benefits are shown on an example implementation and guidance for a practical implementation is given.

‍