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 in order to find approximations for the bias, the variance and for the mean squared error of the least-squares estimator. For this purpose such expansions are shown to be useful only when the autoregressive model contains at least one non-redundant exogenous explanatory variable. It is found that large sample and small disturbance asymptotic techniques give closely related results in this model, which is not the case in stable dynamic regression models. The results are specialised to the random walk with drift model, where it is seen that the ratio of the standard deviation of the disturbance tot he drift term plays a crucial role. The random walk to the model with drift plus a linear trend is also examined. The accuracy of the approximations are checked in the context of these models making use of a set of Monte Carlo experiments to estimate the true moments.