Predictive performance of linear regression models

Ozkale M. R.

STATISTICAL PAPERS, vol.56, no.2, pp.531-567, 2015 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 56 Issue: 2
  • Publication Date: 2015
  • Doi Number: 10.1007/s00362-014-0596-4
  • Journal Name: STATISTICAL PAPERS
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.531-567
  • Keywords: Ridge estimator, Liu estimator, Two parameter estimator, Linear restrictions, Cross-validation, Prediction, BIASING PARAMETER, BIASED-ESTIMATION, RIDGE REGRESSION, CRITERION, VALIDATION, ESTIMATORS, VARIABLES, SELECTION, SQUARES, ERROR
  • Çukurova University Affiliated: Yes

Abstract

In this paper, the cross-validation methods namely the , PRESS and GCV are presented under the multiple linear regression model when multicollinearity exists and additional information imposes restrictions among the parameters that should hold in exact terms. The selection of the biasing parameters are given so as to minimize the cross-validation methods. An example is given which illustrates the comprehensive predictive assessment of various estimators and shows the usefullness of computing. Besides, the performance of the estimators under several different conditions is examined via a simulation study. The results displayed that the biased estimator versions and the restricted form of the biased estimator versions of cross-validation methods give better predictive performance in the presence of multicollinearity.