Improved model selection criterion

Publisher Website
Google Scholar
Abstract
In this article, we propose a new model selection criterion with a penalty function based on the penalty functions of the Bayesian information criterion (BIC) of Schwartz (1978) and Theil’s (1961) R2 criterion. We compare the performance of this new criterion with some commonly used model selection criteria in the context of selecting regressors for the linear regression model. We show that this new criterion is strongly consistent, very competitive and usually selects more parsimonious models compared to Akaike’s (1973) information criterion (AIC), R2 and the generalized cross validation (GCV) criteria but is generally less parsimonious than BIC.