Testing for common autocorrelation in data-rich environments

This paper proposes a strategy to detect the presence of common serial correlation in large-dimensional systems. We show that partial least squares can be used to consistently recover the common autocorrelation space. Moreover, a Monte Carlo study reveals that univariate autocorrelation tests on the factors obtained by partial least squares outperform traditional tests based on canonical correlation analysis. Some empirical applications are presented to illustrate concepts and methods.