PREDICTIVE PERFORMANCE OF AVERAGING-G (AV-G) PRIOR IN BAYESIAN MODEL AVERAGING

David F. Adiele; Onyedikachi O. John; Felix U. Ochuba; Vivian O. Adiele; Umezurike Chinaegbomkpa

David F. Adiele
Onyedikachi O. John
Felix U. Ochuba
Vivian O. Adiele
Umezurike Chinaegbomkpa

Keywords: Bayesian model averaging, av-g prior, parameter priors, model priors, model uncertainty, predictive performance, posterior inclusion probability

Abstract

Bayesian model averaging provides a good theoretical foundation for addressing model uncertainty. However, it faces challenges in the specification of priors. This study investigates the statistics ability of the av-g prior and benchmark prior in the light of different model priors as a palliative in such a situation. Thus, this paper considered growth data set with 41 independent variables and 72 observations. Using the benchmark prior and av-g prior in combination with model priors such as uniform model prior, binomial model prior and beta-binomial model prior, the results reveal that in most of the effective regressors, the av-g prior gives more posterior inclusion probabilities than the benchmark prior. Further investigation reveals that the predictive performance of benchmark prior and av-g prior competes favourably. This study shows that av-g prior is a credible non informative prior that can be applied in BMA when no substantial prior information is available.