Optimal Importance Sampling Parameter Search for Lévy Processes via Stochastic Approximation

The author proposes stochastic approximation methods of finding the optimal measure change by the exponential tilting for Lévy processes in Monte Carlo importance sampling variance reduction. In accordance with the structure of the underlying Lévy measure, either a constrained or unconstrained algorithm of the stochastic approximation is chosen. For both cases, the almost sure convergence to a unique stationary point is proved. Numerical examples are presented to illustrate the effectiveness of our method.