A periodogram-based test for weak stationarity and consistency between sections in time series

In one approach to spectral estimation, a sample record is broken into a number of disjoint sections, or data is collected over a number of discrete trials. Spectral parameters are formed by averaging periodograms across these discrete sections or trials. A key assumption in this approach is that of weak stationarity. This paper describes a simple test that checks if periodogram ordinates are consistent across sections as a means of assessing weak stationarity. The test is called the Periodogram Coefficient of Variation (PCOV) test, and is a frequency domain test based on a technique of spectral analysis. Application of the test is illustrated to both simulated and experimental data (EMG, physiological tremor, EEG). An additional role for the test as a useful tool in exploratory analysis of time series is highlighted