Comparing OLS and Shrinkage estimators in the presence of multicollinearity: Evidence from Nigerian macroeconomic data

This study compares Ordinary Least Squares (OLS) and shrinkage estimators under multicollinearity using Nigerian macroeconomic data (2000–2023) from the World Development Indicators. Ridge, Lasso, and Elastic Net are evaluated based on coefficient stability and predictive performance. Results reveal multicollinearity among predictors, particularly exchange rate and capital formation variables. OLS achieves the best in-sample fit (MSE = 4.214; R² = 0.665) but produces unstable estimates. Shrinkage methods improve stability through regularization. Cross-validation results show that Elastic Net achieves the lowest prediction error (CV-MSE = 4.29), indicating superior generalization, while Lasso provides the most parsimonious model based on AIC and BIC. The findings demonstrate that although OLS performs well in-sample, shrinkage estimators offer more reliable and interpretable results in the presence of multicollinearity