Abstract

This talk considers making the usual importance sampling estimator more robust via winsorization. The threshold level at which to winsorize is determined by a concrete version of the Balancing Principle, which may be of independent interest. The procedure adaptively chooses a threshold level among a pre-defined set by roughly balancing the bias and variance of the estimator when winsorized at different levels. As a consequence, it provides a principled way to perform winsorization, with finite-sample optimality guarantees. Empirically, the estimator outperforms the usual importance sampling estimator in high-variance settings, and matches the performance of the usual estimator when the variance of the importance sampling weights is low.