Detecting Stationarity of GDP: A Test of Unit Root Tests

Abstract

Despite extensive research of research on unit roots, consensus on several important issues and implications has not emerged to date (Libanio, 2005). There are many series which were being investigated for existence of unit root and for these series, there is conflict between the researcher regarding the existence of unit root.  For a given data series it is generally not possible to decide which of unit root tests would be the best suited. The Monte Carlo experiments prove that the performance of unit root tests depends on the type of data generating process (DGP), but for the real data we do not know the true DGP. Hence, we cannot decide which of the tests would perform best for a series. The bootstrap approach of Rudebusch (1993) offers an alternative to measure the performance of unit root test for any real time series with unknown DGP. Rudebusch (1993)’s approach is extended to measure and compare the performance of unit root tests for annual real GDP series of various countries. Our results show that unit root tests have very low ability to discriminate between best fitting trend stationary and difference stationary models for GDP series of most of the countries and that Phillips Perron test is superior to its rivals including Dickey-Fuller, DF-GLS and Ng-Perron tests. The results also support existence of unit root in real GDP series.