Negative R2 in SUR

[bastou]

Hi everyone,
First I apologize for my english that is not my native language.

I'm trying to run a SUR estimation on a system of this form (I simplify) :
Cost = c(1) + c(2)*p1+c(3)p2+c(4)*p1p2 (i create a new variable named p1p2 in my datas that is equal to p1*p2)
a1 = c(2)+c(4)*p1
a2= c(3) + c(4)*p2

For doing that first i import my datas.
Then I create a new system and write my equations
then I estimate using SUR estimation

=> Eviews gives me the estimation of all parameters of the system, and it looks like what I expected. This is on the first page, but then, on the following pages, Eviews gives the estimation of each equation by OLS ...and for some of them it gives an ajusted R quared of -300 (i mean "minus 300).

Is it normal with a SUR or am I wrong in my estimation procedure??

Please help me these neg R2 make me crazy!!

[startz]

A negative R2 is possible in SUR. It is not possible in OLS if you have a constant term in the equation. A negative adjusted R2 is possible, but -300 seems very surprising.

PS: Your English is pretty darn good!

[bastou]

Thank you for your answer! I have some strange new elements:

While my SUR estimations gave me negative R² i tried to estimate my system with OLS => negative R² appeared again (-1000 this time!!) that's why i wonder if there is not a problem in the way i estimate. Let me describe it step by step:
- I have panel datas (15 years, 18 countries)
- I import them on eviews => on my worfile there all my variables (so each variable is of dimansion [15*18,1]
(I try an OLS estimation, everything is fine, eviews recognize that my estimation method is panel least squares)
- I go on Object/new Object/ system
- a white window appears where i can write my equation
- my equation are like : cost = c(1)*p1+c(2)*p2+c(3)*(p1p2)^0.5 (i created myself the variable the variable (p1p2)^0.5 before importing datas)
demand1 = c(1)+c(3)*(p2/p1)^0.5 (i also created this variable before importing)
demand2 = same derivate with p2
- when everything is written i choose Estimate/SUR/ok
I go the results with negative R² (my initial and actual problem..) but i also notice that is seems that eviews doesnt take into account that these are panel datas :
System: UNTITLED
Estimation Method: Seemingly Unrelated Regression
Date: 06/30/16 Time: 15:34
Sample: 1995 2007
Included observations: 234
Total system (balanced) observations 2808

The term "panel" is not included in the estimation method => this is my first pist of error

-then I do exactly the same way and choose OLS this time => same problems :
- negative R²
- "Panel" seems not taken into account (System: UNTITLED
Estimation Method: Least Squares
Date: 06/30/16 Time: 14:51
Sample: 1995 2007
Included observations: 234
Total system (balanced) observations 2808

Regarding at these two stranges things (negative R² with OLS on a system, and "panel" not in the estimation method) I wonder if the problems is not due to the non panel dimension of my regression (but how to ask eviews to make SUR regression with panel as my datas are in the panels format), and if i didn't with a step in my procedure that leads to a negative R² even with OLS?

I hope this is enought clear and will not discourage you

Anyone who sees my error is more than welcome!!

[EViews Gareth]

EViews does not offer panel system estimation

[startz]

The cost equation doesn't have a constant term. This may explain the negative R2.

[bastou]

Thanks for your answers!
Indeed if i add a constant term in the cost function R² become nealy positive (but still negative (-0.4)).
Are you sure that eviews does not offer panel system specification? this could explain everything! Do you know if R or stata (:/) can do that?

[startz]

Note you are restricting the coefficients across equations. That can also explain the negative R2.

[bastou]

I didn't know that restricting coeff across equations could lead to negative R². Thanks! i'm going to look further on that