I just found that the deterministic simulation using the command SIMUL is sensitive to the number of periods you choose. Shortly, the impulse responses are different if you choose periods=10 or periods=100.
Has anybody found something similar? If so, do you know what to do to fix this problem.
The deterministic simulation algorithm has initial values and end values. The simulated path has to obey both of them. End values should be reached in infinity. Infinity is typically not within 10 periods, in a persistent model not even within 100 periods. So, extend your periods!
An extra problem arises when model has (dampening) deterministic cycle. The cycle can - sometimes - be far away from end values at the end period. So, remember to plot your solutions.