PROC SSM Estimation of Time-varying parameters for Time-series forecasting

pap — Fri, 12 Oct 2018 16:16:57 GMT

Hi – I want to run a time-series linear regression model with time-varying betas using the SSM approach (see attached for equations). The dependent variable is stationary (through differencing). I looked into SAS PROC SSM but I am not clear of the right syntax to use. Given the expertise in this community, I thought you may have a quick answer to my question. I would greatly appreciate your help. I am using SAS v 9.4 on Windows 10.

Regards, -Pappe

Re: PROC SSM Estimation of Time-varying parameters for Time-series forecasting

rselukar — Sun, 14 Oct 2018 19:01:50 GMT

You can easily do this in SSM. However you must lag your C variables prior to using them in the SSM procedure (in a DATA step). Even though SSM procedure permits DATA step statements, the LAG and DIF operations are not permitted. Suppose X = Lag( C ) and there are three X variables X1 X2 X3 (so beta is three dimensional). Then you can use the following syntax:

Proc ssm data=test;

State beta(3) type=rw cov(g); /* this assumes epsilon_t is three dimensional noise with full covariance */

Comp regEffect = (X1 X2 X3)*beta;

Irregular eta;

Model y = regEffect eta;

Run;

I have shown syntax for univariate Y. Your Y can be multivariate. You will need to read the SSM syntax and DOC examples for these more involved cases. For univariate Y you can also use the RANDOMREG statement in PROC UCM, which might be even easier.

See SSM and UCM procedure docs at http://support.sas.com/documentation/onlinedoc/ets/indexproc.html#ets143

topic PROC SSM Estimation of Time-varying parameters for Time-series forecasting in SAS Forecasting and Econometrics

PROC SSM Estimation of Time-varying parameters for Time-series forecasting

Re: PROC SSM Estimation of Time-varying parameters for Time-series forecasting