Finite sample comparisons of the distributions of the ols and gls estimators in regression with an integrated regsorad correlated errors

We compare the finite sample distributional properties of the OLS and GT,S mtinialors 11 a rcgrassior wilh arl inl,cgrd,ctl rcgrtssor ant1 corrctifical errors of the form of AR(1) and MA(1) processes. The approach is one of first deriving the joint characteristic function of the quadratic forms in the clefiriit,on of t,hc est,irrial,ors and then rurrierically inverting these 1.0 find the distributions. When the characteristic functions are intractable, Monte Carlo integration is employed. We find substantial differences in the finite jarriplc ditributiorls of OLS ant1 C:LS dthough lsynlptotically thee distributions are equivalent.