Effects of Multicollinearity on the Estimation of Macroeconomic Variables by Using Data from Sudan

Abstract

The problem of multicollinearity in the assessments of coefficients is well established. However, it is rarely researched in the estimations of macroeconomic variables and economic performance of developing countries. Predicatively, it has impacts on the estimations of coefficients that should be used in economic decisions, strategic planning and if researchers are more industrious estimations of monetary supplies and demands. All such parameters are very basic and essential in economic plannings and their applications should be done not only in research but in ground applications of the specialized authorities, e.g., Ministries of Finance, Central Banks, Pricing Units, etc. However, that is rarely done and when carried out, the estimators can be inefficient, misestimated or have statistic problems such as multicollinearity, autoregressions or homoscedacity. All the former lead to misinterpretarions of the statistical and econometric results and if used in any of above-mentioned parameters can be erroneous. Wrong and misinterpreted coefficients, reversed signals or incorrect significance levels are expected. Thus using such coefficients is useless and lead to bad decisions. Truly, macroeconomic variables usually have correlations with each with variable degrees. However, the latter shows the erroneous estimations only when tested. The current text book focuses on building-up models for macroeconomic variables of Sudan with the possibilities that they have the multicollinearity problem. In the analysis the Ordinary Least Squares method is used besides the Ridge Regression to estimate the equations. The analysis revealed that there are multicollinearity when the production and consumption functions were estimated. However, investment, liquidity functions and monetary equilibrium did not have that problem. We recommend here that the utilization of OLS in the estimation can be beneficial and reliable.