How to fit an ARMAX model using statsmodels

Question

How do I use the statsmodels ARMA process to fit a difference equation of the form.

y[k] = - a1 * y[k-1] + b0 * u[k] + b1 * u[k-1] + c0 * e[k] + c1 * e[k-1]

I'm not shure how to set up the exog matrix. E.g.

import statsmodels.api as sm
# some stupid data

y = np.random.randn(100)
u = np.ones((100,2))
armax = sm.tsa.ARMA(y, order=(1, 1), exog=u).fit()

results in

ValueError: could not broadcast input array from shape (2) into shape (3)

It's probably easy to solve but I'm new to the field.

Thanks.

(I'm using statsmodels 0.6)

Answer 1

The issue here is that you're passing two constant columns, then telling fit to add another constant column with trend='nc'. Admittedly, we should fail more gracefully here, but you need to try something like

u = np.random.randn(100, 2)

Instead of the constant exog.