Alternative lag length selection criteria and the split-trend stationarity hypothesis

Abstract

This study asks whether evidence that key macroeconomic time series are stationary around broken trends is robust to using different criteria to determine the lag length in the ADF regressions. When lag lengths are determined using the Schwarz criterion or two different specific-to-general methods, tests for unit roots in several series in the Nelson-Plosser (1982) data and in US postwar real GNP find weaker evidence against the unit root hypothesis than either Perron (1989), who set the date for the break in the trend a priori or Zivot and Andrews (1992), who determined the break date endogenously.