[R] Block Bootstrap Methods for Multivariate Time Series

[Andreas Klein]

Hello.

I have a problem with selecting the right block size, when I want to bootstrap multivariate non-iid time series.

Assume we have N time series each of length T and obtain for each time series an optimal block size l. So we get l1, l2,..., lN optimal block sizes.
When I want to apply a block bootstrap method (circular or stationary bootstrap) I have to draw blocks to sustain the serial dependency structure of each time series, before computing my statistic.

How long is the one block size for that multivariate bootstrap, which I have to draw? - Is it max(l1,...,lN)? Does anyone have any idea or a paper-link, where the topic is covered?

Thank you very much for your help.


Sincerely,
Andreas