Abstract

Bootstrap is a popular methodology for simulating input uncertainty. However,it can be computationally expensive when the number of samples is large. Wepropose a new approach called \textbf{Orthogonal Bootstrap} that reduces thenumber of required Monte Carlo replications. We decomposes the target beingsimulated into two parts: the \textit{non-orthogonal part} which has aclosed-form result known as Infinitesimal Jackknife and the \textit{orthogonalpart} which is easier to be simulated. We theoretically and numerically showthat Orthogonal Bootstrap significantly reduces the computational cost ofBootstrap while improving empirical accuracy and maintaining the same width ofthe constructed interval.

 

Quick Read (beta)