On testing for unit roots and the initial observation

Recent work has highlighted the crucial dependence of the power of existing unit root tests on the deviation of the initial observation from the deterministic component of the series. In practical applications, it is useful to have a test available with power more robust to the initial condition. In this paper, we propose such a procedure based on a data-dependent weighted average of the standard Dickey--Fuller and Elliott--Rothenberg--Stock tests. Simulation of the new test's power reveals very good performance across different magnitudes of the initial condition, and the procedure's value is further highlighted by application to U.S. producer price inflation. Copyright 2005 Royal Economic Society