period specific dummy

[katrin]

Hi,

I have to implement a dummy for period effects that are independet from the underlying fundamentals in the model. Therefore I have to reestimate a model that was originally estimated with stata. They implemented this time dummy by the command
i.quarter
-> "unary operator to specify indicators"
(more info on i.quarter, see stata help for fvvarlist, command i.varname)

if I use period specific effects in the "estimation box", the results are not the same, unfortunately :(
i found @expand(@obsid) somewhere in the forum but this generates an error message

Does someone have an idea how to define such a variable?

THANKS!

[EViews Gareth]

http://www.eviews.com/Learning/dummies.html

[katrin]

Hi,

thanks for your reply. Unfortunately that's not what I really need.
Perhaps I explained it wrong. I need a time-varying coefficient. This "dummy" should take values "by itself". I've got a regression and there are large unexplained parts. This variable should keep the time independent movements of the underlying fundamentals. Do you know/ understand, what I am searching for? I have a regression, 10 countrys, 47 quarters. I've got country-fixed effects in my regression and I am using White period standard errors & covariance.
And now, I "know" that there is a time independent effect. And I would like to keep this time independent effect in this new "dummy".
Do you know what I want and how I could implement such a time-varying coefficient?

THANKS

[startz]

A dummy is a variable, not a coefficient. And a dummy only has the values zero and one.

Do you want a different dummy variable for each quarter?

[katrin]

Thanks for your reply.

I need a variable which one captures time-varying effects. The results should be different values for each period. If there are large variations from the underlying fundamentals, then this variable should be relative high. If these variations independent from time are only small, it should result a small value for the variable. Therefore I think, I do not need a dummy. I thought I need one because someone told me, I would need one and I haven't done that much with Eviews (it's a long time ago...)

Do you understand what I want to do? If so, do you have an idea how I can do this?

THANKS

[startz]

It sounds like you want to add period fixed-effects to the model. That's the convenient way in EViews of adding a separate dummy variable for each time period. The coefficients on the fixed effects allow for arbitrary time differences

[katrin]

Hi,

I think you're right :)
The problem, if I add period fixed effects by using the "estimation window" and then choose period fixed effects, the results are "good", but different from the values I should get. I've tried to use @expand(@obsid) - same problem.
The resulting fixed effects are not that far away from the "right" ones. The most striking one is quarter 1 -> my value ist about 0.5, the "right" one is zero...

The further problem is, I have to perform the Pesaran CD test for both, the regression with and without "time dummy". I got the code in the forum. For the regression without "time effects" I get the right result, p-value = 0.000
If I perform the same by using the above period fixed effects, the p-value is about 0.000359 -> but this one should be about 1,996 - I should not be able to reject cross sectional independence

Do you have another idea how to implement such perid effects?

THANKS :)

[startz]

Hi,

I think you're right :)
The problem, if I add period fixed effects by using the "estimation window" and then choose period fixed effects, the results are "good", but different from the values I should get. I've tried to use @expand(@obsid) - same problem.
The resulting fixed effects are not that far away from the "right" ones. The most striking one is quarter 1 -> my value ist about 0.5, the "right" one is zero...

The further problem is, I have to perform the Pesaran CD test for both, the regression with and without "time dummy". I got the code in the forum. For the regression without "time effects" I get the right result, p-value = 0.000
If I perform the same by using the above period fixed effects, the p-value is about 0.000359 -> but this one should be about 1,996 - I should not be able to reject cross sectional independence

Do you have another idea how to implement such perid effects?

THANKS :)

I don't understand how you know what results you "should" get.

I think you are also a little confused about p-values, as a p-value o 1,996 is impossible.

[katrin]

Hi,

p-value>1 is possible because of the two-sided p-values for non-symmetric distributions

I know the results because I have the results from an estimation with Stata. I have to "re-estimate" them and after having done so, I have to do some further thing - implement new variables, add some quarters, add some countries... Therefore I know the "right" results...

[startz]

Hi,

p-value>1 is possible because of the two-sided p-values for non-symmetric distributions

A p-value is a probability. It can't be outside 0 to 1. Something's wrong with the calculation.

I know the results because I have the results from an estimation with Stata. I have to "re-estimate" them and after having done so, I have to do some further thing - implement new variables, add some quarters, add some countries... Therefore I know the "right" results...

EViews and Stata generally give the same results to a few decimal places, so you're probably doing something different in the two programs. If you post your Stata command and your EViews command, perhaps someone here can spot the difference.

[katrin]

Hi,

you're right... I didn't thought about the high value of this p-value because it was just "given"...
I'll take a look and then we'll see

Stata code for the regression:
encode country, generate (iid)
i(iid)
iis iid
...
xi: reg interest_ exe_ growth_ debt_ i.quarter i.country, cluster (iid) fe

I have to do the same in eviews. Same Data are in Eviews. For the above regression without i.quarter I get the same results. But I can't implement these time component - it should be independet from the underlying fundamentals (exe, growth,debt). If I implement period fixed effects in the "estimation window", the results are near, but too far away.

Does someone have an idea?

THANKS :)

[EViews Glenn]

It would help if you post your data.

[katrin]

Hi
thanks for your reply!

What do you need? the excel file, stata-do file? Can I send you the data with pm?

[EViews Glenn]

I'd post the EViews file at the least with the specification that you tried to estimate. If you have the Stata file for your example as well, that would also help.

[katrin]

Hi Glenn,

here is my workfile. I'll send you a pm with the coefficients that should result.

Thanks :)