Any future plans for supporting AAD for risk sensitivities of vanilla swaps, etc.?

[yrtf]

I wanted to ask about Adjoint Algorithmic Differentiation support, particularly for risk sensitivity calculations going forward. I have been experimenting with computing IR deltas for vanilla swaps and more complex scripted products. The current sensitivity framework in ORE relies primarily on bump-and-reval, which works well but comes with a known computational cost. Is AAD support intended only for the ScriptedTrade / AMCValuation engine, or is there a roadmap to extend it to native trade types (Swap, Swaption, etc.)?

[pcaspers]

AAD starts to leak into traditional trade types, see e.g. the MC CG Swaption engine here. For us, this is motivated by performance (and to a lesser extent accuracy), Cross Currency Swaptions and Cross Currency Flexi Swaps are a recent example for this. We have no examples of FD engines or even Vanilla engines yet, but there is no fundamental reason why we wouldn't consider building CG based engines if the need arises.

[NatMatLogica]

Adding a separate integration example to Peter’s answer: I’m with MatLogica, and we’ve published an ORE/AADC worked example covering FX spot/forward valuation and sensitivities.

That doesn’t establish support for your vanilla IR workload. If this is still something you’re exploring, which pricing engine are you using for the swaps? That would help establish whether the same approach is relevant.