Bayesian predictive densities based on latent information priors

► The notion of latent information priors is introduced. ► Parametric submodels of multinomial models are considered. ► Predictive densities are evaluated by the Kullback-Leibler divergence. ► Limits of Bayesian predictive densities form an essentially complete class. ► Minimax predictive densities are constructed by using latent information priors.