Cross-section effects with period GLS weights not allowed?

[eng2010]

I tried to estimate a panel data model with fixed effects. The panel consists of 39 cross-sections and 32 time series. In the Panel Options, I chose Cross-section: Fixed. Then in the GLS Weights options, I chose Period SUR. Then I got an error message: Cross-section effects with period GLS weights not allowed? Please advise why there is an error message? Is this a numerical limitations in Eviews or a theoretical limitations in the first place?

Thanks,
Eric

[EViews Glenn]

It's a computational issue. Applying the period weights means that the simple projection matrices that we use to rid the estimates of the fixed effects no longer are valid.

You should be able to estimate the model in your case by adding

@expand(c, @crossid)

as a variable to the regression and dropping the cross-section fixed effect. This will do the brute-force fixed effects estimation.

[eng2010]

Thanks for your suggestions. But I would like to point out that there is a restriction on the maximum number of cross-section dummies. Once the number of cross-section dummies is greater than 14, there will be an error message saying that "Near singular matrix".

[EViews Glenn]

There are no restrictions on the number of dummies. That's a singularity. Do you have a full set of dummies and a constant?

[serkansans]

Hello,

I'm trying to replicate the study of Konya (2006) "Exports and growth: Granger causality analysis on OECD countries with a panel data approach," appearing in Economic Modelling 23 (2006) pp. 978–992. this model tests the null hypothesis within each cross-section within a SUR context. This means I think error terms of different cross sections show contemporaneous shocks. However, I could not imagine how to do it in eviews 7. I have run the CD test and it shows me to run SUR. Could you please show mo how to do it? I am confused witt how to do it in eviews. Thank you for your help in advance.

[EViews Glenn]

Without looking at the paper, I can say that in EViews, contemporaneous correlation SUR is handled using the Cross-section SUR option.

[heidi]

Dear Eviews-Users?

I wanted to do cross-section fixed together with period sur and tried to enter the command specified by Glenn.
Unfortunately it says insufficient number of observations.
I have panel-data for 15 companies and for 5 years and 14 regressors.

When I excluded all regressors, but one I still got the error message.

Thank you very much for your help.

[EViews Gareth]

We'd need to see the workfile.

[heidi]

Here is the workfile

I entered
ls C_gri c _01_indb _02_dual _03_meet _04_sizeb _05_Sizebsq _06_fem _08_inv _09_size _11_Profit _12_liqui _13_lever _14_beta "wgt =persur" @expand (c, @crossid)

[EViews Glenn]

Your command syntax is wrong. You want the period SUR to be an option on the ls command.

ls(wgt=persur) C_gri c _01_indb _02_dual _03_meet _04_sizeb _05_Sizebsq _06_fem _08_inv _09_size _11_Profit _12_liqui _13_lever _14_beta @expand (c, @crossid)

That said, I think you are still going to have problems. Based on your description of the problem (without looking at the workfile) if you are doing period SUR, you should have at least as many time series observations as cross-sections (which you don't have) for the relevant covariance matrix to be non-singular. It's akin to having fewer observations than number of regressors in a standard regression problem. Note that the period SUR estimator is a feasible GLS estimator requires asymptotics in the time series dimension.

[heidi]

ok, thank you. it makes sense now.

How do I then determine the right GLS weight methods? Or when do I know to leave it at no weights.

I cannot find any further instructions on this in the handbook.

[EViews Glenn]

That's really an econometric question. I will note that you've got relatively few observations to be doing any kind of GLS.

[heidi]

What possibility do I have then to handle existing serial correlation as indicated by the Durbin-Watson statistics? If I just include an AR(1), how can I then figure out if autocorrelation still exists, since what I learned from the discussion board, the typical autocorrelation do not exist for panel data?

[stungaro]

Hi, I am in a situation similar to the one described at the beginning of this thread.
After not being allowed to use cross-section fixed effects with period GLS weights I tried to "brute force" the fixed effects estimation as suggested by Glenn, but now I have another error message: "Near singular matrix in calculation of SUR weights."
I cannot look for correlations, as they do appear only when EViews calculates SUR weights, and not before.
Have you got any idea on what I could do? Thank you so much
Stefano

[stungaro]

The same type of error ("Near singular matrix in calculation of SUR weights.") appears when I keep the same specification (without fixed effects) but I run the regression only in a sub-period.
Thank you so much for your support
Stefano