Finite sample improvements in statistical inference with I(1) processes

Robinson and Marinucci (1998) investigated the asymptotic behaviour of a na rrow-band semiparametric procedure termed Frequency Domain Least Squares (F DLS) in the broad context of fractional cointegration analysis. Here we res trict discussion to the standard case when the data are I(1) and the cointe grating errors are I(0), proving that modifications of the Fully Modified O rdinary Least Squares (FM-OLS) procedure of Phillips and Hansen (1990) whic h use the FDLS idea have the same asymptotically desirable properties as FM -OLS, and, on the basis of a Monte Carlo study, find evidence that they hav e superior finite-sample properties. The new procedures are also shown to c ompare satisfactorily with parametric estimates. Copyright (C) 2001 John Wi ley & Sons, Ltd.