Small-Sample Properties of Nonlinear Least Squares and Maximum Likelihood Estimators in the Context of Autocorrelated Errors

Abstract Rao and Griliches (1969) compared several methods of parameter estimation in models having autocorrelated errors. They concluded that nonlinear least squares estimators were not superior to two-stage linear estimators. This study partially replicates the Rao and Griliches Monte Carlo simulation and, in addition, examines the maximum likelihood estimator as a possible competitor. The simulation results are not consistent with those of Rao and Griliches; the small-sample efficiency of nonlinear and maximum likelihood estimators appears to be consistently high and thus reverses some of Rao and Griliches's conclusions.