This paper proposes a strategy to detect the presence of common serial correlation in high-dimensional systems. We show by simulations that univariate autocorrelation tests on the factors obtained by partial least squares outperform traditional tests based on canonical correlations.