Further Investigation of the Uncertain Unit Root in Gnp

This paper adopts a different approach to the study of the persistence of U.S. GNP. First, this paper uses a more powerful version of the ADF test developed by Elliot, Rothenberg and Stock (1992). Second, we also examine the results from a unit root test that has trend stationarity as the null (Leybourne and McCabe, 1994). Third, simulated critical values generated from plausible trend stationary and difference stationary models for GNP data are used in order to minimize the possible biases induced by nuisance parameters in finite samples. The ability of these two tests to discriminate against plausible alternatives is evaluated using alternative-specific rejection frequencies. Fourth, to evaluate the implication of extending the time span of the data on the ability to make clear inferences regarding the presence of unit roots, we examine both post-war quarterly data and a longer annual series spanning the period 1869 to 1986. For quarterly data, these two unit root tests do not provide a definite conclusion regarding the existence of a unit root in GNP data. In contrast, when analyzing annual data over the 1869-1986 period, we obtain very sharp results: The unit root null is rejected, while the trend stationary null is not. Moreover, the alternative-specific power for the trend stationary null test is fairly high. Hence, with a longer span of data, one can obtain strong evidence of trend stationarity in per capita GNP. We discuss possible explanations for this result, including shifts in the data generating process as well as changes in the method by which the earlier GNP data is constructed