Effects of model specification on tests for unit roots in macroeconomic data

Abstract

Tests for unit roots in autoregressive models (tests for stationarity) are popular in the macroeconomics literature. Monte Carlo experiments in Schwert (1987) show that unit root tests derived for pure autoregressive processes have different sampling distributions when the true process is a mixed autoregressive-integrated moving average (ARIMA) process. Tests suggested by Said and Dickey, 1984, Said and Dickey, 1985, Phillips (1987), Phillips and Perron (1986), and Dickey and Fuller, 1979, Dickey and Fuller, 1981 are applied to a variety of monthly and quarterly macroeconomic time series to illustrate the effects of ARIMA model specification on inferences about stationarity.