Bayesian predictors for an AR(1) error model

Abstract

It is shown that the "mixtures predictor" suggested by Latif and King (1993) as a predictor for an AR(1) error model is the same as a conventional Bayesian predictor. Viewing alternative predictors within the Bayesian framework, those that are both conditional and unconditional on the AR(1) parameter ρ, indicates how choice of a predictor depends (perhaps implicitly) on assumptions about prior information and loss functions.