“…Existing approaches are based on Monte Carlobased techniques, like on finite-differences (bump and revalue), pathwise or likelihood ratio techniques, for which details can be found in [14] , chapter 7. Several extensions and improvements of these approaches have appeared, for example, based on adjoint formulations [15] , the ChF [16,17] , Malliavin calculus [18,19] , algorithmic differentiation [20,21] or combinations of these [22][23][24] . Intuitively, the ddCOS method follows a similar approach as the likelihood ratio method, i.e.…”