If you solve a model using the stochastic extended path, I understand that when the order is set to 0, then the model is simply solved using a perfect foresight algorithm. Then the IRFs would be identical to what you have in the deterministic case. (As http://www.dynare.org/phpBB3/viewtopic.php?f=1&t=4107)
However, with higher order, there is uncertainty for the number of periods specified, hence there can be an interaction of the shocks similar to when solving the model using a higher order perturbation method. Then, should we calculate GIRFs that are different from the perfect foresight IRFs? Does simult_ work after using extended_path?
Thanks for clarifying,