Department of Pure and Applied Mathematics, Ladoke Akintola University of Technology, P.M.B. 4000, Ogbomoso, Oyo State, Nigeria
Ayinde, K., Department of Pure and Applied Mathematics, Ladoke Akintola University of Technology, P.M.B. 4000, Ogbomoso, Oyo State, Nigeria
In linear model with autocorrelated error terms, regressors are not only assumed fixed (non - stochastic) in repeated sampling but also uncorrelated with the error terms. These assumptions are not always tenable especially in business, economics and social sciences. Therefore in this paper, we examined the performances of some estimators of linear model namely; ordinary least square(OLS) and four feasible generalized least estimators which are Cochrane Orcut (CORC), Hidreth - Lu (HILU), Maximum Likelihood (ML), Maximum Likelihood Grid (MLGD) when normally distributed stochastic regressors exhibit various degrees of correlation with the autocorrelated error terms through Monte - Carlo studies. At various levels of autocorrelation (ρ) and correlation between stochastic regressor and autocorrelated error terms (λ), 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 except when λ →1 the best estimator is either ML or MLGD or both; and to a very lesser extent CORC and HILU when autocorrelation level is low (ρ = 0.4) and high (ρ = 0.8). When λ →1, the OLS estimator is best except when the sample size is moderate (n=40) and large (n=80). Furthermore, when the autocorrelation level is very high (ρ = 0.9) or tends to unity (ρ → 1) and λ≤0.75 the HILU and the CORC, in that order, are superior to the other estimators. However, when λ>0.75, the HILU, ML and to a lesser extent, CORC are best. © EuroJournals Publishing, Inc. 2008.
