groups in a panel

[mariposa]

Dear all,

for my current project I want to analyze groups of countries. As on the time series level I have much too less observations for the estimations I want to make (f.ex. rolling estimations of 5 year windows with annual data), I want to structure my workfile as a panel (by the way I have 50 countries, 27 years).

Now I am interested in analyzing different subgroups of the countries (my cross-section-identifiers). I am aware of that this is not the idea of a panel - but still, that is what I want to do
A pool probably would be easier, but as far as I read the forum, with 50 countries it is better to use a panel.

One idea is to make several pages with different panel (containing different subsets of countries), but I hope, that there is a more elegant way to change the composition of cross-section-identifieres in my panel!!

Can anybody help me with this??

Best regards, Mariposa

[EViews Gareth]

Just put everything in one page, then whenever you want to do something, just set the sample to be only the cross-sections you want.

[mariposa]

Thanks, Gareth! I already left my office, where I have my EViews-Version, so I can only try on Monday... Actually, it is the first time I use a panel workfile structure. So I guess the programming code to set the sample is similar to the time series case: sth. like

Code: Select all

smpl _usa _deu _fra

? But I will have a look at the Reference Manual Monday.

Thanks again and have a nice weekend
(I know in the US you have still some hours to work...)

Best regards, Mariposa

[EViews Gareth]

no, something like:

Code: Select all

smpl if crossid = "_usa" or crossid = "_deu" or crossid = "_fra"


[mariposa]

Thanks

[mariposa]

Hi again,

I tried it, but unfortunately it does not work (so in the meantime I worked with a pool, but I think I would be really better off with a panel structured workfile). Can you give me a hint?!

I tried it also with the workfile of one of your posts (http://forums.eviews.com/viewtopic.php?f=7&t=74#p168 ):
http://www.eviews.com/Forum_support/pool2panel/poolg7.wf1, that I changed to a panel structure with:

Code: Select all

pagestack unemp?

Now, with the suggested

Code: Select all

smpl if crossid = "_can" or crossid = "_fra"

I get the error message "Error in Sample: CROSSID is not defined" ; thus as I think that the crosssection identifiers are stored in id01, I tried

Code: Select all

smpl if id01 = "_can" or id01 = "_fra"

But then I get the message "WARNING: Sample contains no observations."

What do I do wrong? What is the correct way to define the sample?

[EViews Gareth]

It's case sensitive. In that particular example, it should be:

Code: Select all

smpl if id01 = "_CAN" or id01 = "_FRA"


[mariposa]

oh great! thank's a lot!

[mariposa]

Hi again,

as I want to make rolling estimations for different "subgroups" within my panel, I want to programm a loop. Therefore my string variable should support quotes within quotes (with some "hierarchical order"), what it apparently does not do. Is there any solution??

Here comes a summary of what I want to do:

Code: Select all

%kl1=" id01="_BRA" or id01="_CHN" or id01="_CZE" "
%kl2=" id01="_BEL" or id01="_DNK" or id01="_FIN" or id01="_FRA" "

for %condition {%kl1} {%kl2}

' set window size
!window = 5

' get size of workfile
!length = @obsrange

' compute stats for each window
for !i = 1  to  !length-!window+1

   smpl @first+!i-1 @first+!i+!window-2 if {%condition}
   'now the estimations etc.

next 'refering to condition


The problem is in the first part, where I define the conditions. As the part of if id01="..." already contains quotes I did not find out how to sum it up.

[EViews Gareth]

Make a sample object first, then use that sample object inside the loop. Something like:

Code: Select all

sample smpl1 if id01="_BRA" or id01="_CHN" or id01="_CZE"
sample smpl2 if id01="_BEL" or id01="_DNK" or id01="_FIN" or id01="_FRA"

for %condition smpl1 smpl2

smpl @first+!i-1 @first+!i+!window-2 if {%condition}


[mariposa]

Great, it works! Many thanks, Gareth!!

I think, I already said it several times, but as it is true, once again: I really appreciate the reliable help in this forum!! Without your aid, I suppose that I already would have despaired several times of my EViews programms

[acsbarnabas]

Dear All,
I am a first time user here, so I just hope that my query finds the right person. I have a panel database with 500 cross sections. I have 60 variables I would like to create a correlation matrix on each cross section. However I do not want to create the correlation matrix 500 times by making a sample for each cross section. Do you have an idea how I could create 500 correlation matrices for the 500 cross sections in one or a few step?
I am really grateful for your reply,
brgds,
Barnabas

[startz]

You're talking about 500 tables, each with about 1800 unique entries. That's close to a million numbers you want to generate. Out of curiousity, what are you going to do with a million numbers when you get them?

[acsbarnabas]

Hi Startz:)
No, it is in one table, ordered in a Panel structure, if it makes the difference. Actually I am examining the relationship between stock and macro fundamentals, and by generating the correlation tables, I would like to spot high correlation "areas", which I could model later on.

In relation to this. Do I have to set sg, in order to run regression on panel data?

Thank you.
brgds,
Barni