Multicollinearity mitigation and unbiased estimations: an application of restricted least squares

Abstract

This work presents an application of Restricted Least Squares (RLS) that achieves the mitigation of severe multicollinearity problems along with obtaining unbiased estimates. Both targets are restricted to the verification of the null hypothesis of the restrictions on the coefficients of the RLS model, which has been obtained with the application of two recent methodologies that mitigate multicollinearity problems: residualization and raise regression. The reliability of the proposed methodology is shown through Monte Carlo simulations, which illustrate that it is independent of the degree and type of multicollinearity. Finally, a particular empirical application is implemented, and it displays the possible combination of constraints depending on the case study. In conclusion, this application introduces a technique that can be used in all fields and particular studies where problematic relationships among explanatory variables emerge and the traditional methodologies to overcome the issue are biased techniques.