EViews.com

Your results indicate extreme serial correlation. It appears that your prices are nonstationary, which is not unusual.

Could you please advise me a solution for reducing serial correlation?

Your results indicate extreme serial correlation. It appears that your prices are nonstationary, which is not unusual.

And could you please explain me why you said it was extreme serial correlation from those results? Even the Durbin-watson stats results above were all around 1.7, or because I include AR in the specification. I mean from which result you can conclude that. Please explain me because I want to understand how to interpret the results of SUR (with AR).
Many thanks!

Your results indicate extreme serial correlation. It appears that your prices are nonstationary, which is not unusual.

And could you please explain me why you said it was extreme serial correlation from those results? Even the Durbin-watson stats results above were all around 1.7, or because I include AR in the specification. I mean from which result you can conclude that. Please explain me because I want to understand how to interpret the results of SUR (with AR).
Many thanks!

The use of the AR term makes c(2) the estimate of the first-order serial correlation coefficient.
Out of curiousity, what textbook are you using?

I am sorry that I read many different textbooks on econometrics and also explanations form the internet, but not only a certain book. Anyway for sure, the specification I used with AR term as above is right or not, and how to interpret the results of c(2) in the results.

I am sorry that I read many different textbooks on econometrics and also explanations form the internet, but not only a certain book. Anyway for sure, the specification I used with AR term as above is right or not, and how to interpret the results of c(2) in the results.

The interpretation is that the error terms in the equations have the form
u_t = c(2)*u_(t-1) + epsilon_t

Since you have c(2) around 1, the effect of a shock is essentially permanent. The is a standard result for an asset price.

Thannk you for your prompt reply! As you know the results indicate serial correlation, do you have any suggestions for me to solve this problem? It seems that using AR term is not a good solution, since the coeffecients in the analysis are to much different with results of an estimation withour AR term.

I am sorry that I read many different textbooks on econometrics and also explanations form the internet, but not only a certain book. Anyway for sure, the specification I used with AR term as above is right or not, and how to interpret the results of c(2) in the results.

The interpretation is that the error terms in the equations have the form
u_t = c(2)*u_(t-1) + epsilon_t

Since you have c(2) around 1, the effect of a shock is essentially permanent. The is a standard result for an asset price.

I have a question that as you mentioned above, is c(2) the "p" in the fomula "x_t = px_t−1 + e_t" I met in many textbooks?

Thannk you for your prompt reply! As you know the results indicate serial correlation, do you have any suggestions for me to solve this problem? It seems that using AR term is not a good solution, since the coeffecients in the analysis are to much different with results of an estimation withour AR term.

Unfortunately, this may suggest that the coefficients without the serial correlation correction are wrong.

I am sorry that I read many different textbooks on econometrics and also explanations form the internet, but not only a certain book. Anyway for sure, the specification I used with AR term as above is right or not, and how to interpret the results of c(2) in the results.

The interpretation is that the error terms in the equations have the form
u_t = c(2)*u_(t-1) + epsilon_t

Since you have c(2) around 1, the effect of a shock is essentially permanent. The is a standard result for an asset price.

I have a question that as you mentioned above, is c(2) the "p" in the fomula "x_t = px_t−1 + e_t" I met in many textbooks?

That's right!

Thannk you for your prompt reply! As you know the results indicate serial correlation, do you have any suggestions for me to solve this problem? It seems that using AR term is not a good solution, since the coeffecients in the analysis are to much different with results of an estimation withour AR term.

Unfortunately, this may suggest that the coefficients without the serial correlation correction are wrong.

Thank you. Thus you mean the coefficients got from the estimation with AR term are right, dont you? And results of durbin-watson stat which are around 1.7 in that analysis can indicate no serial correlation, can't they? Can I use that analysis for my research? I am really worried about this analysis due to presence of serial correlation, that is why I put you so many quetions about that. And please apologize me for poor knowledge of this area.

Thannk you for your prompt reply! As you know the results indicate serial correlation, do you have any suggestions for me to solve this problem? It seems that using AR term is not a good solution, since the coeffecients in the analysis are to much different with results of an estimation withour AR term.

Unfortunately, this may suggest that the coefficients without the serial correlation correction are wrong.

Thank you. Thus you mean the coefficients got from the estimation with AR term are right, dont you? And results of durbin-watson stat which are around 1.7 in that analysis can indicate no serial correlation, can't they? Can I use that analysis for my research? I am really worried about this analysis due to presence of serial correlation, that is why I put you so many quetions about that. And please apologize me for poor knowledge of this area.

You've hit on something that is easy to get confused about in EViews. The Durbin-Watson statistics from the regressions including the AR(1) term are a check for further serial correlation. So unless you were getting satisfactory Durbin-Watson's before including the AR term, which is unlikely, you're sort of stuck with the problem.

Thank you very much, Startz, for you kind and considerate helps!

Dear all,

I recently stumbled upon the same issue - negative Rsquared in SURE. I followed the advice here, and tested individual regressions for miss-specification using Ramsey`s test, and yes, they were not well specified. I did some changes, which have not altered my model interpretation and single equations were correctly specified. So I tried the SUR again but was not lucky :)
My first question is: Is it a problem Having negative R-squared in sure?
My second question is: I am new to SUR I have red the eviews manual but have not gone to far. How do I interpret the results?
(by the way, it is a cross-sectional analysis)
(offtopic: the results posted earlier in this thread were certainly not from a stationary time series.....)

Thank you for any help,
D

Coefficient Std. Error t-Statistic Prob.

C(1) -109.4957 15.51494 -7.057437 0.0000
C(2) 6.323659 0.495765 12.75534 0.0000
C(3) 106.8206 19.41212 5.502779 0.0000
C(4) 77.63038 20.46586 3.793165 0.0003
C(5) -7.236468 15.38872 -0.470245 0.6395
C(6) -146.3217 14.43684 -10.13529 0.0000

Determinant residual covariance 3983580.


Equation: FER15_24 = C(1) + C(2)*BR + C(3)*DR + C(4)*HD
Observations: 27
R-squared -0.685209 Mean dependent var 76.84444
Adjusted R-squared -0.905018 S.D. dependent var 23.53378
S.E. of regression 32.48189 Sum squared resid 24266.68
Durbin-Watson stat 1.901243

Equation: FER25_34 = C(5) + C(2)*BR + C(3)*DR + C(4)*HD
Observations: 27
R-squared 0.445545 Mean dependent var 179.1037
Adjusted R-squared 0.373225 S.D. dependent var 39.45731
S.E. of regression 31.23801 Sum squared resid 22443.70
Durbin-Watson stat 2.177375

Equation: FER35_49 = C(6) + C(2)*BR + C(3)*DR + C(4)*HD
Observations: 27
R-squared 0.831042 Mean dependent var 40.01852
Adjusted R-squared 0.809005 S.D. dependent var 15.72081
S.E. of regression 6.870470 Sum squared resid 1085.677
Durbin-Watson stat 1.915539