Importance sampling and active subspace in quasi-Monte Carlo

Quasi-Monte Carlo (QMC) is widely used in computational finance, and its efficiency strongly depends on the smoothness and effective dimension of the integrand. In this work, we propose a three-step approach, referred to as the IS\_AS\_PI method, which sequentially applies importance sampling (IS), the active subspace (AS) method and preintegration (PI). The proposed method overcomes the problem of empirical failure in constructing active subspaces when nonzero payoff gradients are rarely sampled. Here, empirical failure means that the estimated gradient information matrix is zero and thus provides no useful directions. We establish the orthogonal invariance of active subspaces and IS-active subspaces under suitable conditions. The method is applied to pricing Asian and basket options and estimating the Delta of Asian options under the Black--Scholes model. Numerical experiments show that the proposed method avoids this empirical failure in the deep out-of-the-money cases. Moreover, in all experiments, the proposed method with Laplace IS achieves the largest variance reduction factors, indicating the best performance, among the methods compared.