On bootstrapping kernel spectral estimates (Q1192961)

This paper considers the problem of determining the statistical characteristics (such as probability distribution and confidence limits) of a kernel spectral density estimator, by using the bootstrap approach. A simple and natural bootstrapping scheme based on resampling the data periodogram ordinates (appropriately normalized) is introduced. The validity of this bootstrapping scheme is established under mild conditions, and also illustrated by means of a Monte-Carlo simulation study. Applications of the proposed bootstrapping scheme to the problem of determining confidence intervals for and selecting the local bandwidth in kernel spectral estimation are also discussed.