Maximum likelihood vs. minimum sum-of-squares estimation of the autocorrelated error model
A Monte Carlo comparison is made of two iterative methods of estimating the linear regression model with first-order autocorrelated errors. Both methods use T transformed observations: T-1 generalized first differences plus the differentially weighted first observation. They differ in that Beach and MacKinnon uses a maximum likelihood estimate of the autocorrelation coefficient rho, while Prais and Winsten uses a sum-of-squares minimizing estimate. On balance, with small samples of either trended or untrended variables, Prais and Winsten is slightly preferable to Beach and MacKinnon.