The notes I'm reading define the IRF as $\text{IRF}(h)=E_t(X_{t+h})-E_{t-1}(X_{t+h}|\text{shock}_t)$. All the examples they show are linear, i.e. ARMA. However, they say that for nonlinear the formula gets harder. But they don't show explicitly how to do it. So,how does one compute the IRF for non-linear models?
Any help would be appreciated.