Power of the augmented dickey-fuller test with information-based lag selection

Abstract

This study examines the issue on whether lag selection using standard information criteria can help improve the power of the Augmented Dickey-Fuller (ADF) test. Unlike previous studies, the analysis explicitly corrects for the potential sensitivity of critical values – on which measurements of test power are based – to the sample size and lag selection. Empirical power curves are estimated under different lag selection methods. Monte Carlo evidence supports that lag selection based on usual information criteria can lead to power gains for the ADF test for autoregressive processes, but not for processes with moving-average errors.