ARMA param estimation in stan

hi! find the code and model in pystan over here: Jupyter Notebook Viewer

So i was wondering if higher order moving average models are difficult to predict in general or is that a stan thing. I’d appreciate any feedback, also i’m new to stan, i work on PyMC mostly, so let me know if i made any mistakes here

I am not sure I understand your inquiry - are you having problems with the model you shared? My best guess is that you compare the posterior means to the simulated values and see little agreement - if that’s the case, I think you also need to look at the posterior uncertainty (e.g. the posterior 90% credible interval) to see if there is a mismatch between the model and the simulator (if the uncertainty is low and estimates are away from the simulated values), or if simply the amount of data/type of data you simulated does not really give a lot of information about the coefficients (if the uncertainty is high and the simulated values are a posterior still plausible).

If there is a mismatch between the simulator and the model, then one needs to investigate whether the Stan model and the simulator you use actually have the same parametrization of the model. Only once you are sure the parametrization is identical and you are still unable to recover the simulated parameters would I start looking for bugs in the model.

Best of luck with your model!