What's the best way to implement an autoregressive moving average of a time series with a variable lag and window size?

Are all values of c_mu strictly positive at the starting point? What does the output of model.debug() look like?