Construction of Stationarity Tests with Less Size Distortions

Abstract

We propose a (trend) stationarity test with a good finite sample size even when a process is (trend) stationary with strong persistence; this is useful for distinguishing between a (trend) stationary process with strong persistence and a unit root process. It could be considered as a modified version of Leybourne and McCabe’s test (1994, LMC), but with a dierent correction method for serial correlation. A Monte Carlo simulation reveals that in terms of empirical size, our test is closer to the nominal one than the original LMC test and is more powerful than the LMC test with size-adjusted critical values.