A Focused Information Criterion for Locally Misspecified Autoregressive Models

Mardi | 2016-06-14
Sully05 16h00-17h20

Jean-Pierre URBAIN – Jan LOHMEYER – Franz PALM – Hanno REUVERS

This paper investigates the Focused Information Criterion (Claeskens and Hjort (2003)) in autoregressive models with local misspecification. A main advantage of the FIC is that model selection and/or model averaging is focused on the quantity of interest instead of aiming at some measure of global model fit. The quantity of interest is allowed to be any (su fficiently regular) function of the parameters. We derive the asymptotic properties and inspired by the work of Hansen (2005) we apply our estimation method to impulse responses both for the univariate and multivariate cases. Monte Carlo simulations show that our Focused Information Criterion performs comparably to both AIC and BIC selection procedures as well as model averaging procedures based on their smoothed counterparts.Impulse responses, Model selection, Model uncertainty.