On the Number of Bootstrap Simulations Required to Construct a Confidence Interval

Abstract : The purpose of this document is to make two points about the effect of the number of bootstrap simulations, B, on percentile-t bootstrap confidence intervals. The first point concerns coverage probability; the second, distance of the simulated critical point form the true critical point derived with B=infinity. In both cases the author has in mind applications to smooth statistics, such as the Studentized mean of a sample drawn from a continuous distribution. He indicates the change that have to be made if the distribution of the statistic is not smooth. Additional keywords: Exponentive functions.