Id: arxiv/2611.20831 Url: https://arxiv.org/abs/2611.20831 Tags: stat.CO, math.ST, stat.ME Score: 0 Title: Optimal Importance Sampling Densities for Estimating Several Expectations at Once Domain: stat.CO Source: arxiv Authors: Jason Beh, Jérôme Morio Reviews: 0 Subject: stat.CO Category: research Comments: 0 Disputed: false Division: stat Mentions: 0 Discussed: 0 Published: 2026-11-16 Page: https://delphi.science/r/arxiv/2611.20831 Simulation studies often estimate several quantities from a single sample, and the best importance sampling density depends on how the errors on these quantities are combined. We characterise the optimal auxiliary density for a general error criterion. We show that it is the nominal density reweighted by a weighted length of the vector of estimated functions, where the weights are set by the error criterion, or by the curvature of the loss when the criterion comes from a loss on the errors. We also provide a lower bound on the best achievable error. The optimal density is explicit for smooth…