A Simulation Study of Some Ridge Estimators

Abstract

Ridge regression, which defines a class of estimators indexed by a biasing parameter k, is an alternative to least squares estimation in the multiple linear regression model. Many algorithms for the biasing parameter have been proposed in the statistical literature. The present study identifies 10 promising algorithms and systematically evaluates and compares them using Monte Carlo methods. Three algorithms perform well overall. A two-parameter estimator offers potential improvement over one-parameter ridge estimators.

Key Words:

• Least squares
• Ridge regression
• Simulation
• Mean squared error