Addressing Multicollinearity in Economic Forecasting: A Comparative Study of Ridge and Bayesian Regression Models

Abstract

Multicollinearity poses a significant challenge in econometric modelling, particularly when analysing economic indicators that are inherently interrelated. This study investigated the efficacy of Ridge Regression (RR) and Bayesian Regression (BR) as alternatives to Ordinary Least Squares (OLS) in the presence of multicollinearity. Using both simulated datasets and Nigerian economic data spanning 1995 to 2016, including GDP, oil revenue, education, agriculture, manufacturing, trade, and construction, we compared the predictive and inferential performance of the three methods. Model performance was evaluated using Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and Mean Squared Prediction Error (MSPE). The findings revealed that while OLS exhibited superior model fit under AIC and BIC, BR outperformed both OLS and RR in predictive accuracy, as indicated by the lowest MSPE. These results underscore the importance of prior distribution assumptions in Bayesian estimation and highlight the need for careful model selection when handling multicollinearity in economic data.