topic TAR Models in SAS Forecasting and Econometrics

TAR Models

anais — Mon, 23 Jul 2012 13:44:43 GMT

Hey,

I would like to estimate a threshold autoregressive model with SAS software, but I do not know if it is possible to estimate it?

Is it possible to estimate it with the procedure MODEL?

Can someone help me to find some codes to estimate this model?

I know that the estimation of this particular class of models is possible with the software R (using the procedure SETAR).

Thanks,

Anaïs

Re: TAR Models

ets_kps — Mon, 23 Jul 2012 16:55:16 GMT

Hello Anais,

It seems PROC MODEL can be used to estimate a Threshold Autoregressive model in SAS. In fact, in our examples we show how to estimate a switching model in PROC MODEL. This can be found here. http://support.sas.com/documentation/cdl/en/etsug/63939/HTML/default/viewer.htm#etsug_model_sect096.htm

From the developer of the PROC, "...estimation of the switching parameter is problematic. That’s why we approximate
the discrete switch with a smoother (first order continuous) transition between models in the example. For TAR models where more than TWO regimes are considered, this could be onerous to implement in MODEL."

I hope this helps-Ken

Re: TAR Models

anais — Tue, 24 Jul 2012 08:17:43 GMT

Thanks,

I read the example that you gave me.

I modified it a little bit in order to adapt it to my model. I changed the "threshold variable", instead of choosing an exogeneous variable, I chose to put a lag of my endogeneous variable, in order to replicate a SETAR Model.

I tried to estimate it but, it seems that there is a problem of convergence in the estimaors. I don't know how to solve it?

Do I have to change the level of convergence, in order to have "algorithm converged"?

Thanks,

Anaïs

Re: TAR Models

anais — Tue, 24 Jul 2012 09:22:27 GMT

I also tried to use an exogeneous variable (as the "threshold variable"), but it seems that there is a problem of singularity and SAS did not want to estimate my model.

I don't understand, I wrote exactly the same codes as in the example...