An alternative approach to the seasonal integration test

Abstract

The aim of this paper is to propose an alternative approach for detecting a non-constant seasonal pattern in a time-series. It differs from the Canova and Hansen test in taking account of the structure of the autocorrelation of the process in a parametric way. A Monte Carlo experiment was carried out for comparing the two approaches. The statistics calculated according to the alternative one highlighted a better ability in detecting a stationary process at a particular frequency, even with a coefficient near the border of the non-stationarity region, not implying any reduction in the power of the test.