Pooled time-series regression-how to modify reg coefs?

[johnsirius]

Hi Eviews experts and users,

I've got a question about modifying the regression coefficients based on a pooled least squares regression. My data structure consists of 18 country equity indexes and 1 world index. All series are within the same time interval.

The following is my model:
Y? = -(1- c(2)) * [X?*(-1)- c(1)] + c(3) * AR(12) + e
where c1 is the intercept, c2 and c3 are the regression coefs, e is the error term

What I'm trying to do is to make c2 and c3 common coefs, e to be common as well but at the same time let c1 to be cross-sectional specific.

This topic relates to one of my previous posts not long ago http://forums.eviews.com/viewtopic.php?f=4&t=3916, in the previous post I was estimating the model using a simple country-by-country approach. I was able to modify the regression estimation quite easily by typing the modified equation (such as above model) into the equation specification section. But when I switched to pooled regression after realising the ignored cross-sectional correlations, I couldn't modify the equation specification anymore as pool estimation window dosen't support any customizations. (see attached)

By modification or customization, I mean estimate a non-standard econometric model (e.g above model), not a standard Y=c1+c2*x+c3*z+e model where the regression coefs are always next to the regressor.

Thanks in advance!!! Hope I explained my question clear enough.

John
(Eveiews 6)

[EViews Gareth]

As you point out, pooled regression does not let you estimate non-linear regressions, so you're out of luck.

You could, however, estimate it in a panel workfile. Although there is nothing built in to estimate c(1) as cross-country specific, you could use a dummy variable approach to manually estimate it.

[johnsirius]

As you point out, pooled regression does not let you estimate non-linear regressions, so you're out of luck.

You could, however, estimate it in a panel workfile. Although there is nothing built in to estimate c(1) as cross-country specific, you could use a dummy variable approach to manually estimate it.

Thanks very much for the reply!!! Although I didn't know how to do it exactly, I did have a go. What I've done is I stacked the data onto a new worksheet, indeed I was able to manually enter the equation specification. However, pannel regression assumed all reg coefs to be "common" over the entire cross-section. I then was trying to create a dummy variable for c(1), this is where I got stuck again. I thought you can only create dummy variables for regressors not for reg coefs?? How do you actually create a dummy so that c(1) can be cross-country specific?

I tried

Code: Select all

series dum_at=1 series dum_au=2 ... series dum_uk=18 series dum_us=0

but this only creates a dummy for each cross-section, is this the correct way of doing it? where do u put the dummy in the eq specification then? Assuming the eq is DXIT=-(1-C(2))*(XIT(-1)-C(1))+C(3)*MM12

much appreciated!!

[EViews Gareth]

Just multiply any coefficient you want to vary by the dummy.

[johnsirius]

Just multiply any coefficient you want to vary by the dummy.

I'm sorry to brother you again Gareth, but I still can't figure out how to do it (very new to panel regression only learned it after you said pool reg does not support). How to exactly create 18 or 17 dummies for C(1)? Is it @expand or @recode?

I tried: dxit=-(1-c(2))*(xit(-1)-c(1)*@expand(@crossid))+c(3)*mm12 but Eviews returned with " @expand is not a Genr or series expression function"

Thanks a lot

[johnsirius]

I think I've got it working now, the eq specs looks like this：

DXIT=-(1-C(19))*(XIT(-1)-C(1)*(@CROSSID=1) + C(2)*(@CROSSID=2) + C(3)*(@CROSSID=3) + C(4)*(@CROSSID=4) + C(5)*(@CROSSID=5) + C(6)*(@CROSSID=6) + C(7)*(@CROSSID=7) + C(8)*(@CROSSID=8) + C(9)*(@CROSSID=9) + C(10)*(@CROSSID=10) + C(11)*(@CROSSID=11) + C(12)*(@CROSSID=12) + C(13)*(@CROSSID=13) + C(14)*(@CROSSID=14) + C(15)*(@CROSSID=15) + C(16)*(@CROSSID=16) + C(17)*(@CROSSID=17) + C(18)*(@CROSSID=18)) +C(20)*MM12

Is this correct? Thx!!!

[EViews Gareth]

Looks good.