Robust Model Selection in Dynamic Models with an Application to Comparing Predictive Accuracy

Abstract

A model selection procedure based on a general criterion function, with an example of the Kullback-Leibler Information Criterion (KLIC) using quasi-likelihood functions, is considered for dynamic non-nested models. We propose a robust test which generalizes Lien and Vuong's (1987) test with a Heteroscadasticity/Autocorrelation (HAC) variance estimator. We use the fixed-b asymptotics developed in Kiefer and Vogelsang (2005) to improve the asymptotic approximation to the sampling distribution of the test statistic. The fixed-b approach is compared with a bootstrap method and the standard normal approximation in Monte Carlo simulations. The fixed-b asymptotics and the bootstrap method are found to be markedly superior to the standard normal approximation. An empirical application for foreign exchange rate forecasting models is presented.