Wild bootstrap-based ridge estimators under multicollinearity and heteroscedasticity

Multicollinearity and heteroscedasticity frequently co-occur in many practical applications of linear regression undermining the efficiency of the ordinary least squares estimator. Although standard ridge estimators mitigate the effects of multicollinearity but do not account for heteroscedastic error variances, which are not integrated into their estimation frameworks. We have proposed a wild bootstrap framework in ridge parameter estimation to simultaneously deal with multicollinear features and heteroscedastic errors. The wild bootstrap preserves heteroscedastic residual structure while improving inference accuracy in finite samples. The proposed wild bootstrap-based ridge estimators are evaluated via extensive Monte Carlo simulations and real-world data, showing significant improvements over conventional approaches under ill-conditioned heteroscedastic systems.