Permuted weighted area estimators

Calvin and Nakayama previously introduced permuting as a way of improving existing standardized time series methods. The basic idea is to split a simulated sample path into non-overlapping segments, permute the segments to construct a new sample path, and apply a standardized time series scaling function to the new path. Averaging over all permuted paths yields the permuted estimator. This paper discusses applying permutations to the weighted area estimator of Goldsman and Schruben. Empirical results seem to indicate that compared to not permuting, permuting can reduce the length and variability of the resulting confidence interval half widths but with additional computational overhead and some degradation in coverage; however, the decrease in coverage is not as bad as with batching.