In that case, the hyperparameters only concern the variance-covariance matrix of the b coefficients and the precision τ. As is well-known, Bayesian methods are sensitive to misspecification of the distributions of the priors. [...] Conversely, if the base prior is not consistent with the data, the weight λ̂θ,g → 0 and the ML-II posterior density of θ is then close 0 to the posterior q̂∗ (θ|g0) and to the empirical Bayes estimator θ̂EB (b|g0). [...] The DGP of the Monte Carlo simulation study For the random effects (RE), the Chamberlain (1982)-type fixed effects (FE) world and the Hausman and Taylor (1981) (HT) worlds, we use the same DGP as that of Baltagi et al. [...] For each set-up, we vary the size of the sample and the length of the panel. [...] For the coefficient of the lagged dependent variable, ρ, the RMSE is the lowest for the 2S bootstrap when N = 100 and T = 10, but this RMSE is the lowest for the QMLE when N = 200 and T = 30, although the differences are small.