Acerca del problema de la multicolinealidad en la estimación del modelo lineal

Abstract

This paper summarizes various aspects related with the existence of Multicollinearity in the standard linear regression model. The core of the paper focuses on the causes for multicollinearity, its effects on the appraisal of least squares estimates, and diagnostic procedures. Details about these techniques, based on eigenvalues, singular values and condition numbers of the matrix of observations, are provided, stressing its relation with the variance of the estimates. The last section of the paper presents an application with data of Argentina.