Abstract: Traditional model selection criteria select a single model based on its global fit of the model. In contrast, a focused information criterion is tailored to a parameter of interest, based on which, it selects a model. We propose a focused information criterion and a plug-in averaging method for a general class of estimators in a unified theoretical framework, and investigate their asymptotic and finite-sample properties. The results from Monte Carlo simulations and an analysis of real data show that the proposed selection and averaging methods perform comparably with other methods.

Key words and phrases: Focused information criterion, model averaging, model selection.