Stoch_simul not from Steady State

I would like to compute irfs starting from points away from the steady state. I thought that using an initval would have been enough to specify the point from which I want to compute the irf but it seems like the stoch_simul command always compute the irfs starting from the steady state. The user guide and the reference manual do not agree on this point. The former says that using initval is enough while the latter clearly states that the stoch_simul command computes policies (and therefore irfs) only around the steady state. What am I missing?

Thanks

The user guide is outdated. The manual should be correct. Note that for order=1 everything is linear. Hence, where you compute the IRFs is irrelevant and you can simply add the IRFs. For higher order approximations, you should use the function simult_.m
A user interface for choosing the starting values will only be implemented in a future version of Dynare.