Efficient importance sampling for copula models

Abstract

In this paper, we propose an efficient importance sampling algorithm for rare event simulation under copula models. In the algorithm, the derived optimal probability measure is based on the criterion of minimizing the variance of the importance sampling estimator within a parametric exponential tilting family. Since the copula model is defined by its marginals and a copula function, and its moment-generating function is difficult to derive, we apply the transform likelihood ratio method to first identify an alternative exponential tilting family, after which we obtain simple and explicit expressions of equations. Then, the optimal alternative probability measure can be calculated under this transformed exponential tilting family. The proposed importance sampling framework is quite general and can be implemented for many classes of copula models, including some traditional parametric copula families and a class of semiparametric copulas called regular vine copulas, from which sampling is feasible. The theoretical results of the logarithmic efficiency and bounded relative error are proved for some commonly-used copula models under the case of simple rare events. Monte Carlo experiments are conducted, in which we study the relative efficiency of the crude Monte Carlo estimator with respect to the proposed importance-sampling-based estimators, such that substantial variance reductions are obtained in comparison to the standard Monte Carlo estimators.