{"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","summary":"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…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-11-16","reactions":{}}