ON THE PERFORMANCE OF NEW WEIGHTED RIDGE ESTIMATORS FOR MITIGATING HETEROSCEDASTICITY PROBLEM ON A LINEAR REGRESSION MODEL WITH APPLICATION

Heteroscedasticity, a common violation of the homoscedasticity assumption in classical linear regression models, adversely affects parameter estimation and predictive accuracy, is a critical issue to be addressed. This study developed and evaluated new weighted ridge estimators for mitigating the effects of heteroscedasticity in linear regression models, particularly where there is little or no multicollinearity. The proposed estimators were derived by combining ridge regression and weighted least squares techniques. Monte Carlo simulations at varying levels of heteroscedasticity, error variances, and sample sizes, were used to assess their performance with the mean square error (MSE) showing that the proposed ORERW estimator performed best when the nature of heteroscedasticity is known, while OREW2 was superior when heteroscedasticity is unknown. Real-life data on passenger car mileage was used to validate the estimators with GREW3 and OREW3 outperforming the traditional methods.