topic Is there a .fit & .predict equivalent in SAS, specially when using proc autoreg in SAS Data Science

Is there a .fit & .predict equivalent in SAS, specially when using proc autoreg

arpit971 — Sun, 07 Mar 2021 20:37:40 GMT

I'm trying to design a multivariate time series model using proc autoreg. However, being new to SAS and having gone through the documentation of SAS, I'm having a hard time understanding if there's something similar to .fit & .predict step, like in python.

For example, if I want my model to fit on 500 rows of data and I want it to predict for the next 'n' time steps, how is that to be achieved, using proc autoreg.

Re: Is there a .fit & .predict equivalent in SAS, specially when using proc autoreg

Reeza — Mon, 08 Mar 2021 03:24:25 GMT

Does this help?

https://documentation.sas.com/?cdcId=pgmsascdc&cdcVersion=9.4_3.4&docsetId=etsug&docsetTarget=etsug_autoreg_gettingstarted06.htm&locale=en

For non time series, there's PROC SCORE and PLM which are analogous to predict (fit would be the PROC itself and the model is stored to an item store). Time series data is always a bit different though and I think it varies a bit by PROC. The method above is an old, tested and true but also just works.

@arpit971 wrote:

I'm trying to design a multivariate time series model using proc autoreg. However, being new to SAS and having gone through the documentation of SAS, I'm having a hard time understanding if there's something similar to .fit & .predict step, like in python.

For example, if I want my model to fit on 500 rows of data and I want it to predict for the next 'n' time steps, how is that to be achieved, using proc autoreg.

Re: Is there a .fit & .predict equivalent in SAS, specially when using proc autoreg

arpit971 — Mon, 08 Mar 2021 05:52:21 GMT

Thank you. This certainly helps a lot.

I was kind of doing the same thing but at the same time wondering if there's a better way to do this other than assigning null values to the dependent variable in the test period.

I was wondering if you could also help and redirect me to something similar to a recursive training a model in SAS; if there are some examples I could refer to. The goal is to further analyze how the model forecasts could be improved for a relatively shorter 'n' time step prediction.