Variable Lag Augmentation in Regression Models with Possibly Integrated Regressors : Some Experimental Results

Abstract

This paper is concerned with the Wald test statistic of general restrictions in dynamic regression models with possiobly integrated regressors. We try to improve the size and power of the Wald statistic through the extended lag augmentation (LA) in the regression model and the bias correction of the instrumental variable ( IV) estimator. It has been known that the extended lag augmentation is generally, but not always, useful in increasing the finite sample power of the Wald statistic. In this papper we propose a new approach, called the variable lag augmentation approach, which selects an appropriate lag length. The finite sample experiments show that the proposed approach produces higher power of the test than the conventional LA estimator.