Title: Estimating nested expectations without inner conditional sampling and application to value of information analysis

Abstract: Motivated by various computational applications, we investigate the problem of estimating nested expectations. Building upon recent work by the authors, we propose a novel Monte Carlo estimator for nested expectations, inspired by sparse grid quadrature, that does not require sampling from inner conditional distributions. Theoretical analysis establishes an upper bound on the mean squared error of our estimator under mild assumptions on the problem, demonstrating its efficiency for cases with low-dimensional outer variables. We illustrate the effectiveness of our estimator through its application to problems related to value of information analysis, with moderate dimensionality. Overall, our method presents a promising approach to efficiently estimate nested expectations in practical computational settings.