Bootstrapping the Box–Pierce Q test: A robust test of uncorrelatedness

Abstract

This paper describes a test of the null hypothesis that the first K autocorrelations of a covariance stationary time series are zero in the presence of statistical dependence. The test is based on the Box–Pierce Q statistic with bootstrap-based P-values. The bootstrap is implemented using a double blocks-of-blocks procedure with prewhitening. The finite sample performance of the bootstrap Q test is investigated by simulation. In our experiments, the performance is satisfactory for samples of n=500. At this sample size, the differences between the empirical and nominal rejection probabilities are essentially eliminated.