Unit Roots in Macroeconomic Time Series: Some Critical Issues

This paper suggests that the relevant question concerning unit root' in the U.S. real GNP time series pertains to the relative importance of difference-stationary and trend-stationary components. Various analytical approaches indicate than an accurate answer is not obtainable with existing data. The paper next considers whether trending series should be differences prior to use in regression analysis and suggests it may not matter greatly if autocorrelated residuals are avoided. Finally, the paper argues that the absence of cointegration among variables does not imply the absence of any practically useful long-run relationship.