why do I get negative Rsquared in SURE?

[marwa]

Dear all, I want to estimate a system of equations simaltaneously,I am running a the system using SURE (seemingly unrelated regression). I usually get negative R squared for at least one or two of the equations. Why would this happen? and if this is possible how can I comment on the model.
This also arise when I use 2SLS method.
thank you.
Marwa

[startz]

Dear all, I want to estimate a system of equations simaltaneously,I am running a the system using SURE (seemingly unrelated regression). I usually get negative R squared for at least one or two of the equations. Why would this happen? and if this is possible how can I comment on the model.
This also arise when I use 2SLS method.
thank you.
Marwa

It is possible to get a negative R squared in SURE, although it it unusual. It is quite likely a sign of misspecification. And you should check to be sure the equations in question have constant terms.

Negative R-square in 2sls is not unusual and not a problem.

[cuongnh]

Dear all,
I am having a problem of executing SUR in Eviews. I cannot write the right estimate equation for using SUR in Eviews software. I am almost new to SUR in this software, please help me!
Thank you in advance!

[EViews Gareth]

Dear all,
I am having a problem of executing SUR in Eviews. I cannot write the right estimate equation for using SUR in Eviews software. I am almost new to SUR in this software, please help me!
Thank you in advance!

I think you'll have to provide a little more information on what is going wrong...

[cuongnh]

Dear QMS Gareth,
Thank you for reply. I am almost new to using SUR in Eviews. I am trying to see the relationships between prices of crude oil and prices of grains including rice, corn, wheat and soybeans (each equation includes price of each grain as dependent variable and price of oil and some other dummies as independent variables). I would like to execute all those equations simultaineously, that is why I tried to use SUR. However, I could not make a good equation specification. I am not good at this, and what I have been trying is, example:
priceofrice c priceofoil
priceofcorn c priceofoil
priceofwheat c priceofoil
priceofsoybeans c priceofoil
c: constant (as in OLS estimation)
In the system window, I type exactly the same as above (in different lines), but eviews reported that "Index missing for coeffienct C in the "Priceofrice c priceofoil"
Please give me some suggestions. Thank you!

[EViews Gareth]

When you enter an equation into a System object in EViews you have to enter it by Expression, rather than by List.

Thus you'll need to enter them along the lines of:

Code: Select all

priceofrice = c(1)+c(2)*priceofoil priceofcorn = c(3)+c(4)*priceofoil priceofwhat = c(5)+c(6)*priceofoil priceofsoybeans = c(7)+c(8)*priceofoil

[cuongnh]

Thank you for your great supports. I have tried and it was successful.
However, could please help me to explain how c(1), c(2),..., c(8) were defined in the equation? Please excuse me if this is a stupid question. And, I would like to get Durbin-watson stat (general result) of the equation, can I get it somewhere, or I have to get each result from each equation?
So great for me! many thanks!

[EViews Gareth]

c(1)...c(8) are just the coefficients.

EViews only reports the DW statistic for the individual equations.

[startz]

Thank you for your great supports. I have tried and it was successful.
However, could please help me to explain how c(1), c(2),..., c(8) were defined in the equation? Please excuse me if this is a stupid question. And, I would like to get Durbin-watson stat (general result) of the equation, can I get it somewhere, or I have to get each result from each equation?
So great for me! many thanks!

Rather than asking for a Durbin-Watson, which is not usually estimated in a system, estimate a serial correlation coefficient common across equations.
Here's an example:

Code: Select all

y1 = c(1) +[ar(1)=c(2)] y2 = c(3) +[ar(1)=c(2)]

[cuongnh]

Thank you for your suggestions. I am trying with serial correlation as recommended.

[cuongnh]

I have tried to use AR in the equation to reduce serial correlation, but it seems that the coefficients are changed also. Please explain me more about this, and also help me to find a solution to presence of positive serial correlation in the time series regression. The Durbin-Watson stats in my analyses are aroung 0.25.
Many thanks!

[cuongnh]

Thank you for your great supports. I have tried and it was successful.
However, could please help me to explain how c(1), c(2),..., c(8) were defined in the equation? Please excuse me if this is a stupid question. And, I would like to get Durbin-watson stat (general result) of the equation, can I get it somewhere, or I have to get each result from each equation?
So great for me! many thanks!

Rather than asking for a Durbin-Watson, which is not usually estimated in a system, estimate a serial correlation coefficient common across equations.
Here's an example:

Code: Select all

y1 = c(1) +[ar(1)=c(2)] y2 = c(3) +[ar(1)=c(2)]

Dear Startz, could you please explain me more about how to use the y1 = c(1) + [ar(1)=c(2)]? Thank you very much!

[startz]

Thank you for your great supports. I have tried and it was successful.
However, could please help me to explain how c(1), c(2),..., c(8) were defined in the equation? Please excuse me if this is a stupid question. And, I would like to get Durbin-watson stat (general result) of the equation, can I get it somewhere, or I have to get each result from each equation?
So great for me! many thanks!

Rather than asking for a Durbin-Watson, which is not usually estimated in a system, estimate a serial correlation coefficient common across equations.
Here's an example:

Code: Select all

y1 = c(1) +[ar(1)=c(2)] y2 = c(3) +[ar(1)=c(2)]

Dear Startz, could you please explain me more about how to use the y1 = c(1) + [ar(1)=c(2)]? Thank you very much!

Paste the equations into your system and then estimate but SURE. Of course, you should use your real variable names and real equations.
The coefficient c(2) is common across all equations and is the estimated serial correlation coefficient. You can do the usual "t-test" as to whether it equals zero.

[cuongnh]

Dear Startz,
I tried with the specification like this, could you please check for me?
PRICE = C(1) + C(3)*POIL + C(4)*RTDRICQ + C(5)*SDVXB + C(6)*DTHS + [ar(1)=c(2)]
PCORN = C(7) + C(8)*POIL + C(9)*SDCRNQ + C(10)*DJAN1_415 + [ar(1)=c(2)]
PWHEAT = C(11) + C(12)*POIL + C(13)*RTDWHTQ + C(14)*SDJL07 + C(15)*SDA_S07 + C(16)*SDJ_MR08 + C(17)*DEXRSTR + [ar(1)=c(2)]
PSOYBEANS = C(18) + C(19)*POIL + C(20)*SDSBNSQ + [ar(1)=c(2)]
I include c(2) in all equations, and eviews gave result of c(2) also. But I really dont know if this specification is right as you recommended me or not.

[cuongnh]

This is the result of the estimation equation above:
System: SYS02
Estimation Method: Seemingly Unrelated Regression
Date: 12/15/08 Time: 11:51
Sample: 2 276
Included observations: 276
Total system (balanced) observations 1100
Iterate coefficients after one-step weighting matrix
Convergence achieved after: 1 weight matrix, 23 total coef iterations

Coefficient Std. Error t-Statistic Prob.

C(1) -20.91652 50.89161 -0.411001 0.6812
C(3) 0.044845 0.008481 5.287714 0.0000
C(4) -1.753421 0.424716 -4.128458 0.0000
C(5) -0.002309 0.002057 -1.122491 0.2619
C(6) 0.049021 0.214256 0.228797 0.8191
C(2) 1.001912 0.002940 340.7842 0.0000
C(7) 158.4557 386.0675 0.410435 0.6816
C(8) 1.908677 0.265701 7.183560 0.0000
C(9) -0.048923 0.067619 -0.723502 0.4695
C(10) -2.595167 5.614736 -0.462206 0.6440
C(11) -992.5368 3032.916 -0.327255 0.7435
C(12) 2.586863 0.627586 4.121928 0.0000
C(13) -105.5602 25.08193 -4.208615 0.0000
C(14) 0.053482 0.282135 0.189564 0.8497
C(15) -0.031662 0.195773 -0.161729 0.8715
C(16) -0.257889 0.153411 -1.681038 0.0930
C(17) 70.11487 10.62486 6.599132 0.0000
C(18) 374.5164 846.2281 0.442571 0.6582
C(19) 5.016987 0.616814 8.133718 0.0000
C(20) 0.083293 0.135045 0.616778 0.5375

Determinant residual covariance 1304522.

Equation: PRICE = C(1) + C(3)*POIL + C(4)*RTDRICQ + C(5)*SDVXB
+ C(6)*DTHS + [AR(1)=C(2)]
Observations: 275
R-squared 0.994523 Mean dependent var 13.80673
Adjusted R-squared 0.994422 S.D. dependent var 4.095366
S.E. of regression 0.305876 Sum squared resid 25.16767
Durbin-Watson stat 1.805684

Equation: PCORN = C(7) + C(8)*POIL + C(9)*SDCRNQ + C(10)
*DJAN1_415 + [AR(1)=C(2)]
Observations: 275
R-squared 0.993884 Mean dependent var 502.3973
Adjusted R-squared 0.993793 S.D. dependent var 124.4859
S.E. of regression 9.807635 Sum squared resid 25971.22
Durbin-Watson stat 1.643589

Equation: PWHEAT = C(11) + C(12)*POIL + C(13)*RTDWHTQ + C(14)
*SDJL07 + C(15)*SDA_S07 + C(16)*SDJ_MR08 + C(17)
*DEXRSTR + [AR(1)=C(2)]
Observations: 275
R-squared 0.968601 Mean dependent var 867.8218
Adjusted R-squared 0.967777 S.D. dependent var 128.5154
S.E. of regression 23.06939 Sum squared resid 142096.5
Durbin-Watson stat 1.925458

Equation: PSOYBEANS = C(18) + C(19)*POIL + C(20)*SDSBNSQ +
[AR(1)=C(2)]
Observations: 275
R-squared 0.990187 Mean dependent var 1180.436
Adjusted R-squared 0.990078 S.D. dependent var 228.3465
S.E. of regression 22.74494 Sum squared resid 140197.1
Durbin-Watson stat 1.813277