A new improved estimator for reducing the multicollinearity effects

Abstract

The least-squares (LS) is a known estimator for the estimation of the linear regression model parameter. The LS is inefficient in the happening of the significant correlations among the explanatory variables. Alternatively, we propose a new regression estimator for the purpose of the reduction of multicollinearity effects. The comparisons between the new regression estimator with each of the available regression estimators are performed theoretically. Then, a massive simulation with different factors is done. The main finding point of this study is that the new regression estimator is superior to other available regression estimators under some determined conditions using the mean squared error. Finally, a numerical example is also done to ensure the superiority of the new regression estimator.