Bandwidth Selection, Prewhitening, and the Power of the Phillips-Perron Test

Abstract
This study examines several important practical issues concerning nonparametric estimation of the innovation variance for the Phillips-Perron (PP) test. A Monte Carlo study is conducted to evaluate the potential effects of kernel choice, databased bandwidth selection, and prewhitening on the power property of the PP test in finite samples. The Monte Carlo results are instructive. Although the kernel choice is found to make little difference, data-based bandwidth selection and prewhitening can lead to power gains for the PP test. The combined use of both the Andrews (1991,Ecpnometrica59, 817–858) data-based bandwidth selection procedure and the Andrews and Monahan (1992,Econometrica60, 953–966) prewhitening procedure performs particularly well. With the combined use of these two procedures, the PPtest displays relatively good power in comparison with the augmented Dickey-Fuller test.