A New Family of Estimators for Random Walk Problems

Abstract

In this paper the weighted uniform sampling technique introduced by Powell & Swann (1966) is extended to random walk problems and generalized. The family so defined includes, as special cases, not only weighted uniform estimators, but traditional unbiased importance sampling estimators and an interesting new biased importance sampling estimator. Model problem analysis shows that biased importance sampling offers the promise of uniformly low mean square error over a large spectrum of problem types.