Alternative Estimation Techniques for Mitigating Multicollinearity in Macroeconomic Models of Nigerian Economic Growth

Abstract

This Paper studies alternative estimation techniques for mitigating multicollinearity in macroeconomic models of Nigerian economic growth. Using Real Gross Domestic Product (RGDP) as the dependent variable and age dependency ratio, foreign direct investment, inflation rate, population growth rate, exchange rate, crude oil exports, and unemployment rate as explanatory variables, the study compares the performance of Ordinary Least Squares (OLS), Ridge Regression, LASSO, Principal Component Regression (PCR), and Partial Least Squares Regression (PLSR). Descriptive statistics and diagnostic tests revealed severe multicollinearity among predictors, with Variance Inflation Factors ranging from 5.889 to 135.135 and tolerance values as low as 0.007. The OLS model achieved a high R² of 0.976 and adjusted R^2 of 0.952 but produced unstable coefficient estimates due to strong intercorrelations among variables. Ridge Regression delivered the best overall performance with R^2 = 0.980, adjusted R^2 = 0.969, and RMSE = 1248.13, indicating superior predictive accuracy and coefficient stability. LASSO achieved R^2= 0.974 and RMSE = 1325.88 while reducing model complexity by retaining only four significant predictors. PCR effectively reduced multicollinearity, lowering VIF values to 4.345 and achieving R^2= 0.952. PLSR recorded R^2= 0.9566 and RMSE = 1386.61. Ridge Regression emerged as the most effective technique for modeling Nigerian economic growth under severe multicollinearity conditions.