Performances of some estimators of linear model with autocorrelated error terms when regressors are normally distributed

Abstract

Different specification of regressors and levels of autocorrelation are known to affect the performances of estimators of linear model with autocorrelated error terms. In this paper, we examined the performances of the ordinary least square (OLS) and four feasible generalized least estimators namely; Cochrane Orcut (CORC), Hidreth – Lu (HILU), Maximum Likelihood (ML), Maximum Likelihood Grid (MLGD) when regressors are normally distributed at various levels of autocorrelation and sample size through Monte – Carlo studies. The estimators are compared by examing the finite properties of estimators namely; sum of biases, sum of absolute biases, sum of variances and sum of the mean squared error of the estimated parameter of the model. Results show that when the autocorrelation level is small (ρ=0.4), the MLGD estimator is best except when the sample size is large (n=80) where the CORC estimator is best. When autocorrelation is high (ρ=0.8), the CORC estimator is best except when the sample size is small (n=80) where the ML estimator is best. When autocorrelation is very high (ρ=0.9), the HILU estimator is best except when the sample size is large where the CORC estimator is best. Furthermore, when the autocorrelation level tends to unity (ρ → 1), the HILU estimator is best in all the sample sizes.


IJONAS Vol. 3 (1) 2007: pp.22-28