Efficiency in linear model with AR (1) and correlated error-regressor

Abstract

In this study, we conduct several Monte-Carlo experiments to examine the sensitivity of the estimators relative to OLS using the Variance and RMSE criteria, in the presence of first order autocorrelated error terms which are also correlated with geometric regressor. We examine the of efficiency to ı, _, as well as, its asymptotic behaviour, N, when the above two assumptions are violated. We observe that CORC and HILU give similar result, same for ML and MLGRID. OLS is more efficient than CORC and HILU while ML and MLGRID dominate OLS. In the scenarios, efficiency does increase with increase in autocorrelation level, only ML and MLGRID at _ = 0.05 show that efficiency increases with increase in autocorrelation level. All estimators show that efficiency reducesas significant level increases only when the autocorrelation value and samplesize are small (_ = 0.4, N = 20). There is more efficiency gain when N and _are large at all significant correlation levels. Asymptotically, the efficiency of FGLS estimators increase with increasing autocorrelation but it is in different to the correlation levels. The asymptotic ranking is CORC and HILU followed by MLGRID and ML.