Statistical tests for changes in the amplitude, frequency or phase of a sinusoidal variation
The problem considered is that of testing for small changes over time in the properties of a sinusoidal signal contaminated by noise. Two test statistics are proposed. The first is a frequency domain statistic based on the spectrum of the data, while the second is a time domain statistic due to Nyblom. Power studies are used to show that the Nyblom statistic is generally the better. The form of the Nyblom statistic appropriate for phase-change tests also performs well in testing for frequency changes.