<![CDATA[This study develops Bootstrap-Stability Adaptive Ridge Regression (BSA-Ridge), a methodological extension of classical ridge regression for multicollinear regression data. Classical ridge regression controls coefficient variance by adding a uniform quadratic penalty, but it does not distinguish predictors whose coefficients are empirically unstable from predictors whose coefficients are relatively stable. The proposed method estimates coefficient instability through bootstrap resampling and converts the bootstrap variance into predictor-specific penalty weights. The empirical illustration uses the Longley benchmark dataset, a public dataset widely used to examine numerical instability and multicollinearity in least-squares regression. The results show that BSA-Ridge produces interpretable adaptive shrinkage and competitive predictive performance relative to ordinary least squares and classical ridge regression. The contribution of this article is a transparent, equation-based, and reproducible extension of ridge regression that can be further evaluated through simulation and high-dimensional applications.]]>
Copyrights © 2026
