Test to compare two population logspectra (Q880895)

A time series is observed for each sampled subject. It is assumed that the log-periodogram of the time series at frequency \(\omega_j\) can be represented as \(\mu_l(\omega_j)+Z_{li}(\omega_j)+e_{lij}\), where \(\mu_l\) is the population log-spectrum of the \(l\)-th population, \(Z_{li}\) is a specific component of the spectrum of the \(i\)-th subject in the \(l\)-th population, and \(e_{lij}\) are i.i.d. errors. The problem is to test the hypothesis \(\mu_1=\mu_2\). The authors propose a bootstrap-based test and prove that the bootstrap principle holds in Mallows' metric.