Autoregressive models can be run in two ways: sequential or parallel. The former is faster. The later can lead to inaccurate predictions when the forces underlying the data generation tool change. This tradeoff is worth considering.