A STUDY OF UNIT ROOT TESTING

Abstract

Unit root testing is important in time series applications, in particular in economic and financial time series. However, testing for the presence of a unit root is neither easy nor straightforward. The existence of a number of unit root tests in the literature and the sensitivity of these tests to the deterministic and stochastic parts of a time series complicate unit root testing. In view of these considerations some recommendations for practitioners on unit root testing would be desirable. The objective of this study is to make some recommendations on unit root testing based on existing tests, graphical aids and testing strategies. The unit root tests considered here are the Dickey-Fuller, Sargan-Bhargava, Said-Dickey, and Phillips-Perron tests. In addition to the usual graphical aids such as time plot, autocorrelation function, partial autocorrelation function and inverse autocorrelation function, we propose the use of a new graphical procedure, due to Cressie (1988), to determine the order of differencing. The testing strategies are the Dickey, Bell and Miller (1986) and Perron (1988a) strategies. In the exploratory stage we propose the use of graphical aids to detect nonstationarity as well as determine an appropriate model for the stochastic part of the series. To test for the presence of a unit root we recommend the testing strategy of Dickey, Bell and Miller. Three real time series were analyzed using the recommendations suggested.