Moment Approximation for Least Squares Estimators in Dynamic Regression Models with a Unit Root

Asymptotic expansions are employed in a dynamic regression model with a unit root inorder to find approximations for the bias, the variance and for the mean squared error of theleast-squares estimator of all coefficients. It is found that in this particular context suchexpansions exist only when the autoregressive model contains at least one non-redundant exogenousexplanatory variable and that local to zero asymptotic approaches are here without avail.Surprisingly the large sample and small disturbance asymptotic techniques give closely relatedresults, which is not the case in stable dynamic regression models. The expressions for momentapproximations are specialized to the random walk with (trend in) drift model and their accuracyis examined in Monte Carlo experiments.