Unit Root Tests: Results from some recent tests applied to select Indian macroeconomic variables

Abstract

Results from newly developed unit roots tests of ERS (1996), PN (1996), NP (2001) and LM (1994) are compared against their traditional counterparts (ADF, PP and KPSS) on select Indian macroeconomic data. Results from ERS, PN and NP are broadly in agreement. However, using the general to specific criterion of Hall (1994) and the Modified Information Criterion (MIC) of NP for lag length selection, it is found that different lag length can lead to different results. Furthermore, results from using these criteria are also sensitive to the 'maximum' lag length. Both KPSS and its modified version, LM, are found to be prohibitively sensitive to the lag length used. Since as of now no theoretical criterion exists for lag length selection for tests which test the null of stationarity, their use should be avoided, even for the purpose of so-called ‘confirmation’. Another important finding is that frequency of the data and span covered by the sample size plays an important role and whenever feasible, tests must be conducted with as many different frequencies as the availability of data permits. It is not only a large sample size that is important, but also the span covered, an issue raised long ago by Campbell and Perron (1991).