Performance Of Some New Quantile-Based Two Parameter Ridge Estimators For Linear Regression Model: Simulation And Application

Abstract

In regression analysis, the efficiency of the ordinary least square (OLS) estimator decreases when the predictors become highly correlated leading to the problem of multicollinearity. In this study, new quantile- based two-parameter ridge (TPR) estimators are introduced to deal with the issue of multicollinearity in the linear regression model.  The study presents a novel class of modified two-parameter ridge estimators developed using eigenvalues of the correlation matrix of predictors. The performance of the proposed TPR estimators is examined using extensive simulations using the mean squared error (MSE) criteria. The findings revealed that the TPR estimators have better performance than the one-parameter ridge estimators. In addition, the suggested estimator has superior performance than the OLS and is considered a one-parameter and two-parameter ridge estimator. Next, the application of the new TPR estimators is shown in the Economic Survey data. The findings indicate that the suggested NQW2 estimator outperformed all the competing estimators.