Spectral density estimators at frequency zero for nonstationarity tests in arma models

Abstract

In order to apply unit root and cointegration tests to ARMA models it is often required an estimate of the spectral density function at frequency zero. Commonly used are nonparametric estimators and the auto regressive spectral density estimator. It is well known, however, that these estimators can provoke important size and power problems, especially in presence of moving average components. This article proposes estimators based on the estimation of an ARMA model. A Monte Carlo experiment shows that the proposed procedures yields tests with better properties than competing procedures.