Application of the Dickey-Fuller test to the Nelson and Plosser (1982) Data

Abstract

In this letter, Nelson and Plosser's (1982) (NP, hereafter) time series has been retested for unit roots using the Dickey and Fuller (1979) (DF, hereafter) test. There have been two potential misspecification sources identified that may have biased NP's inference towards non rejection of the unit root null hypothesis for all but one series in their data set. These are incorrect lag order choice and (conditional) heteroskedasticity. Furthermore, appropriate testing strategy has been undertaken to alleviate these problems and identify which NP time series have unit roots. In contrast to previous classical (non Bayesian) evidence, only a few time series of NP appear to have unit roots.