Comparison of estimators efficiency for linear regressions with joint presence of autocorrelation and multicollinearity

Abstract

This paper proposes a new estimator called Two stage K-L estimator by combining these two estimators previously proposed by Prais Winsten (1958) and Kibra with Lukman (2020) for autocorrelation and multicollinearity respectively and to derived the necessary and sufficient condition for its superiority over other competing estimators. Simulation study was used to ascertain the dominance of this new estimator using the finite sample properties of estimators in terms of the estimated mean squared error. The study findings shows that under severe autocorrelation and collinearity condition, the proposed Two stage K-L estimator appears to be having a similar performance with RMLE and MLE. Also, under severe autocorrelation and moderate collinearity condition, regardless of the sample size, the proposed Two stage K-L estimator is seen to outperform all other estimators and lastly, the Two stage K-L estimator appears to have an improved performance as the large sample sizes. The study recommends that when autocorrelation and multicollinearity level is at moderate to severe, the proposed Two stage K-L estimator will perform better regardless of the size of the data, and the degree of autocorrelation and multicollinearity should be considered while estimating parameters and thus applying an efficient estimator to avoid erroneous inferences.