topic Re: Proc varmax / panel data in SAS Procedures

Proc varmax / panel data

emeric — Tue, 17 May 2016 11:18:18 GMT

Dear friends,

I have a database with GDP and energy consumption by country (100 countries). This is yearly data from 1900 to 2015. I would like to perform a global varmax model for the whole scope.

Would you have any idea for doing that ?

Best regards,

Emeric

Re: Proc varmax / panel data

gergely_batho — Fri, 22 Jul 2016 10:47:25 GMT

Sound like an interesssting study.

At your fist attempt you probably will use only a couple of countries.

First you will need to put your data into the following format:

Year GDP_Country1 Energy_Country1 GDP_Country2 Energy_Country2

1990 10 6 20 8

1991 12 7 22 10

...

And then... it depends. What is the goal?

Do you want to treat GDP variables as explanatory? Or all variables are Y-s?

Probably there will be high correlation beween variables... but as this is a very long time period, those correlations can change over time. To assess the degree of cointegration will be a challege...

Maybe you could try PROC SSM (more flexible handling changing correlations).

Re: Proc varmax / panel data

emeric — Mon, 25 Jul 2016 08:23:34 GMT

Hello,

Thanks for your answer. Actually, the model I'd like to perform, in one consistent step, is a two-fixed effect for country and time and a model var ecm on energy consumption and GDP.

What I want to assess is the long run relationship between energy consumption and GDP and Granger causalities.

It can be easily perfomed in Eviews and I am surprised it couldn't be computed in SAS (probably due to my lack of knowledge).

I hope it clarifies. I gonna have a look in the SSM procedure.

Kind regards

Emeric

Re: Proc varmax / panel data

gergely_batho — Mon, 25 Jul 2016 10:41:15 GMT

Two-way fixed effects models are available with PROC PANEL.

http://support.sas.com/documentation/cdl/en/etsug/68148/HTML/default/viewer.htm#etsug_panel_details09.htm

But I don't know how to incorporate two-way effects, ecm and causality test in one step (is it possible?).

Maybe VARMAX with country_dummy variables... But then it is just one-way.

The closest example (panel data) in PROC SSM documentation:

http://support.sas.com/documentation/cdl/en/etsug/68148/HTML/default/viewer.htm#etsug_ssm_examples11.htm