Question on Seemingly Unrelated Regression

Questions related to panel (pooled cross-section time series) data.
cyang40
Posts: 9
Joined: Tue Apr 26, 2016 3:30 pm

Question on Seemingly Unrelated Regression

Unread post by cyang40 »

Dear Tom,

I am trying to run the seemingly unrelated regression command using preg(method=sur) and I keep getting the error message "## REG12. SIGMA Is Singular/Not PSD At Row 43. Too Many Equations for Data Set Size?" I have a panel data with 51 economies from 1971 to 2011. The sample is not big, so I have no idea why the system is reporting the error message when I try SUR. The code is provided below. Please advise. Thank you very much.

Code: Select all

open data open_fiscal_51.txt
cal(panelobs=42,a) 1970:1
data(format=free,org=columns) 1//1970:1 51//2011:1 year Y G opent total_a total_l gdp country

*Countries are in following order:

dec vect[strings] labels(51)
input labels
Australia
Austria
Belgium
Bolivia
Brazil
Cameroon
Canada
Chile
Colombia
Congo, Rep.
Costa Rica
Denmark
Dominican Republic
Ecuador
El Salvador
Finland
France
Gabon
Gambia, The
Germany
Greece
Guatemala
Honduras
Iceland
India
Indonesia
Ireland
Italy
Japan
Kenya
Malaysia
Mexico
Morocco
Netherlands
New Zealand
Norway
Pakistan
Peru
Philippines
Rwanda
Senegal
South Africa
Spain
Sweden
Switzerland
Thailand
Togo
United Kingdom
United States
Uruguay
Zambia
*

set growth = (Y - Y(t-1)) / Y(t-1)
set govt = (G - G(t-1)) / Y(t-1)

preg(method=sur) govt / sg
# constant growth{1} govt{1}

Sincerely,

Michael Yang
TomDoan
Posts: 7814
Joined: Wed Nov 01, 2006 4:36 pm

Re: Question on Seemingly Unrelated Regression

Unread post by TomDoan »

That's the whole problem. You have 51 equations (since each individual has a separate equation) and only 42 time periods. You can't estimate a full rank sigma matrix under those circumstances. You need to use some other estimation method that doesn't require as many "parameters".
Post Reply