Bootstrap estimation of biasing parameter in ridge regression with heteroscedastic errors

Résumé fourni par la source

Linear regression models often face challenges from heteroscedasticity in error terms and multicollinearity among predictors, which can significantly impair estimation accuracy. The ordinary least squares estimator loses efficiency under pronounced heteroscedasticity, while multicollinearity may affect ridge regression. To address these issues, a novel methodology combining bootstrap techniques with ridge regression has been developed to substantially enhance estimation precision. Monte Carlo simulations demonstrate that this approach delivers markedly superior results compared to traditional estimators. The method’s robustness is evaluated across diverse error term distributions, including normal, Student’s t, and F distributions. Three established techniques for selecting biasing parameters are strategically incorporated under specific conditions. The findings show that this bootstrap ridge regression approach excels in real-world applications, such as environmental studies, where multicollinearity and heteroscedasticity are prevalent.