Bootstrap-based Budget Allocation for Nested Simulation

Simulation budget allocation is at the heart of a nested (also referred to as two level) simulation approach to estimating functionals of a conditional expectation. In this paper, we propose a sample-driven budget allocation rule under a unified nested simulation framework that allows for different forms of functionals. The proposed method employs bootstrap sampling to guide an effective choice of outer-and inner-level sample sizes. Furthermore, we establish a central limit theorem for nested simulation estimators, and incorporate the sample-driven allocation rule into the construction of asymptotically valid confidence intervals (CIs). Effectiveness of the sample-driven allocation rule and validity of the constructed CIs are confirmed by numerical experiments.