Exploring new estimators in ridge regression: Addressing multicollinearity in economic and petroleum product data analysis

Abstract

Multicollinearity remains a major challenge in regression analysis, leading to unreliable parameter estimates and reduced predictive accuracy. Existing preprocessing methods, such as K1 to K9, attempt to mitigate this issue but are not universally effective. This study proposes three novel ridge regression estimators that address multicollinearity without requiring additional preprocessing. We evaluate these estimators through extensive simulations and real-world datasets spanning multiple sectors. Results show that our approach consistently reduces mean squared error (MSE) and outperforms traditional methods, making it a reliable tool for improving predictive accuracy in economic forecasting and other data-driven fields. Our findings reveal that these new estimators reduce MSE in 136 out of 160 simulation cases and deliver superior performance across multiple datasets, including car consumption, South Africa’s economy, Pakistan’s socio-economic indicators, and Saudi Arabian petroleum product prices. These results highlight the reliability of our estimators in addressing multicollinearity and enhancing predictive accuracy, particularly in economic forecasting and other predictive analytics domains.