About the limits of raise regression to reduce condition number when three explanatory variables are involved

Abstract

This manuscript shows that the raise regression can be considered as an appropriate methodology in order to reduce the approximate multicollinearity that naturally appears in problems of linear regression. When three explanatory variables are involved, its application reduces the condition number of the matrix associated to data set. Nevertheless, this procedure has a threshold: although the columns of X can be separated, it is proved that the condition number will never be less than a constant that can be easily worked out by using the elements of the initial matrix. Finally, the contribution is illustrated through an empirical example.