I’m working with dynare and the IRF obtained does not return to the steady state (even with a large number of periods). I have check the example in the RBC model, but the IRFs in this model doesn’t either.

Actually why should they? In a VAR for instans, when working with stationary time series, the response of your endogenous variables to a shocks has to tend to 0.

Can you be a bit more specific? What’s your model and how did you solve it? Maybe you can post the mod file. Note that sometimes you might get a unit root (or close to) in the system due to parametre choice, even if the exogenous process is stationary. Moreover, there are some specifications that are not consistent with a balanced-growth path. One example is a utility function of the GHH form that eliminates wealth effects on labour supply.

Thanks for your answer. I already found the solution. In fact Dynare calculates the IRF as y_t - ys, so the impulse response function goes to 0 in the graphic which means the same that saying that y_t return to the steady state.