In this paper, we consider an extension of the covariance structure to an AR(g) process with random effects via Bayesian and maximum-likelihood (ML) approaches. We study both parameter estimation and prediction of future values. Numerical results are illustrated with real and simulated data.