Wrong results after a monetary shock

[jpfeifer]

Have a look at

Log linearized and non-linearity in one model

Ideally, you call the log-linearized variables something like N_hat and define N_hat=log(N)-log(STEADY_STATE(N)) That makes sure the mean is correctly taken care of. For perfect foresight, mixing linear and nonlinear equations is problematic, because you will care about nonlinearities but only preserve them in some part of the code.

You need to distinguish the linear and nonlinear variables and provide initial values for the steady states of the nonlinear ones.