Second-order asymptotic expansion approximations to thejoint distributions of dynamic forecast errors andof static forecast errors in the stationary Gaussianpure AR(1) model are derived. The approximation tothe dynamic forecast errors distribution can beexpressed as a multivariate normal distribution withmodified mean vector and covariance matrix, thusgeneralizing the results of Phillips [12]. However,the approximation to the static forecast errorsdistribution includes skewness and kurtosis terms.Thus the class of multivariate normal distributionsdoes not provide as good approximations (in terms oferror convergence rates) to the distributions of thestatic forecast errors as to the distributions ofthe dynamic forecast errors. These results cast somedoubt on the appropriateness of model validationprocedures, such as Chow tests, which use the staticforecast errors and implicitly assume that thesehave a distribution which is well approximated by amultivariate normal.