Efficient SIMM-MVA Calculations for Callable Exotics

Computing Standardized Initial Margin Model Margin Valuation Adjustment (SIMM-MVA) requires the simulation of future sensitivities, but these are expensive to compute for callable products. This paper introduces a method which avoids nested calls to the pricing function, similar to the use of least-squares Monte Carlo (LSMC) for producing future exposures. The method algorithmically differentiates continuation values and underlying swap values with respect to model parameters. These are transformed into sensitivities over market quantities via Jacobians. An illustrative numerical example is provided for a Bermudan swaption demonstrating a massive acceleration w.r.t. brute-force calculations