Manually set model parameters in SARIMAX for prediction

101 views

Skip to first unread message

unread,

May 13, 2018, 7:29:00 AM5/13/18

to pystatsmodels

Hi,

using some articles form Chad Fulton's site as a guidance, I have estimated the model parameters for

a SARIMAX model. Now I want to get the posterior predictive forecasts for out-of-sample dates of my time series.

I have tried so by fitting another SARIMAX model of the same type via MLE and then manually adjusted

the model.params from the posterior predictive distributions of the estimated parameters.

However, the .predict() method does not seem to use the model.params for prediction but the relevant values

seem to be stored somewhere else, as the predictions are always the same for each draw from the posterior.

My question: Where are the parameters stored that are actually being used for SARIMXfit.predict()? Or

is there another intended way to perform predictions with a given set of model parameters?

Thanks for any help,

Sarem

unread,

May 13, 2018, 8:12:44 AM5/13/18

to pystatsmodels

All the statespace model need to create the underlying statespace representation (kalman filter) to make prediction.

see Chad's answer for how estimated params can be used with a new dataset, and the issues for the explanation

Josef

Reply all

Reply to author

Forward

0 new messages