A global tier one bank successfully moved from simply monitoring their CVA to actively hedging its risks, and thus reducing volatility in prices and capital. For this, they needed to calculate CVA sensitivities to a large number of risk factors – typically several hundreds. Using finite differences (bumping) for that task is prohibitively expensive. After a thourough assesment of alternative solutions, they turned to Xcelerit to adapt their implementation to use Adjoint Algorithmic Differentiation (AAD).

AAD allows the client to calculate all CVA sensitivites at up to two orders of magnitude faster than bumping. Xcelerit helped the client to overcome the implementation challenges, such as picking the right software tools, coping with the memory footprint, and adapting the existing code-base with minimum changes.