A NOTE ON THE POWER OF BOOTSTRAP UNIT ROOT TESTS推荐.pdfVIP

Econometric Theory, 19, 2003, 32-48. Printedin the United States of America. DOI: 10.1017/S0266466603191025 A NOTEON THE POWEROF BOOTSTRAP UNITROOTTESTS ANDERS RYGH SWENSEN University of Oslo In this note we consider the asymptotic power functions of some bootstrapunit root tests underlocal alternativesand show that they are in fact the same as for ordinaryunit root tests. This is regardlessof whetherthe differencesof the obser- vations, i.e., the so-called restrictedresiduals,or the ordinaryleast squaresresid- uals are used to constructthe resampledobservations.We also consider models containing a constant and a linear trend and the DF-GLS tests proposed by El- liott, Rothenberg,and Stock (1996, Econometrica64, 813-836). A small Monte Carlo experimentis included. 1. INTRODUCTION The recent surveys on bootstrap methods in time series by Li and Maddala (1996) and Berkowitz and Kilian (2000) both contain special sections on how to deal with nonstationarydata and a discussion of bootstrapunit root tests. Bootstrapproceduresoffer an opportunityto take into account such factors as sample size, various specifications of the initial condition, and the distribution of the errors.They may thereforehave more accuratefinite-sample properties than procedures making use of asymptotic approximations,where such ele- ments typically do not enter. Tabulationsbased on Monte Carlo simulations will necessarily have to specify these factors, and the assumptions may not be well suitedin The always particularapplications. bootstrapprocedurespresent an interesting possibility making more extensive use of data and may there- fore represent an alternative and supplement to tabulations and asymptotic approximatio