Comparative Study Of Estimators In Autocorrelated-Endogenized Linear Model

by Olutunji John Olaomi, Dahud Kehinde Shangodoyin.

Abstract: This study compares the estimators of linear model when the least square assumptions of independence of the error terms and the zero correlation between the regressor and the error terms are violated using a Monte Carlo experiment. OLS, OLSA, 2SLS and 2SLSA estimators were considered in 10,000 replications of the experiment on single equation model where the error terms are AR(1) autocorrelated and at the same time significantly correlated with the regressor(endogeneity). We consider autocorrelation levels 0.4, 0.8 and 0.9, endogeneity levels between the regressor and the error terms at 0.01, 0.02 and 0.05 each at the sample size (N) 20, 40 and 60. The estimators are adjudged using the RMSE criterion on the 108 scenarios. The result shows that 2SLS perform best when N<=40 at all autocorrelation and endogeneity levels. OLSA is the best estimator when N is large (N = 60) and autocorrelation level is <= 0.8, while 2SLSA performs best when N and autocorrelation level are large, at all endogeneity levels. All estimators perform worse as autocorrelation level increases while they perform better asymptotically and with increase in significant level.

Key Words: Monte-Carlo Experiment, Endogeneity, 2SLS, OLS, Autocorrelation

Authors:
Olutunji John Olaomi, olutunji.olaomi@mopipi.ub.bw
Dahud Kehinde Shangodoyin, shangodoyink@mopipi.ub.bw

Editor: Sapra, Sunil K., ssapra@exchange.calstatela.edu.

READING THE ARTICLE: You can read the article in portable document (.pdf) format (215981 bytes.)

NOTE: The content of this article is the intellectual property of the authors, who retains all rights to future publication.

This page has been accessed 1501 times since MAY 17, 2010.


Return to the InterStat Home Page.