Two default sampling algorithms for continuous parameters have been added to the procedure: Hamiltonian Monte Carlo (HMC) and No-U-Turn Sampler (NUTS). You can select them as replacements for the normal- or t-distribution-based random-walk Metropolis algorithm to draw posterior samples.
PROC MCMC supports lagging and leading variables. This enables the procedure to fit dynamic linear models, state space models, autoregressive models, or other models that have a conditionally dependent structure on either the random-effects parameters or the response variable.
PROC MCMC adds an ordinary differential equation (ODE) solver and a general integration function, which enable the procedure to fit models that contain differential equations (for example, PK models) or models that require integration (for example, marginal likelihood models).
The PREDDIST statement makes predictions from marginal random-effects models. For example, you can make predictions for new observations that do not have group membership information in a random-effects model.