I noticed a timing inconsistency in the Making a Custom Statespace Model.ipynb notebook.
To be consistent with the general form in (1)-(2), the ar(3) model should be:
$$x(t) = \rho_1 x_{t-1} + \rho_2 x_{t-2} + \rho_3 x_{t-3} + \varepsilon_{t-1}$$
instead of
$$x(t) = \rho_1 x_{t-1} + \rho_2 x_{t-2} + \rho_3 x_{t-3} + \varepsilon_{t}$$
Alternatively, in (1) instead of $\varepsilon_{t}$ it should be $\varepsilon_{t+1}$. I realize that this doesn't matter in many cases, but in some models there is a big difference between the new state being driven by the same period innovation or the one from the last period. I apologise if this is explained elsewhere in the documentation, I have only looked at this notebook.
