Matched bootstrap procedure for INAR(1) processes

Abstract. Integer-valued autoregressive (INAR) processes is a relatively new field of time series analysis. Bootstrapping such data is not an evident problem, further assumptions are needed, or possibly the bootstrap samples will not form an INAR process. In this paper we present a new, non-parametric bootstrap method, based on the idea of block bootstrap. The procedure resamples blocks that perfectly match to the last element of the previous block. We present properties of this so called matched bootstrap approach and compare our method to other frequently used bootstrap procedures for INAR processes. We apply the methods to a natural disaster dataset.