EViews.com

Posted: **Sat Nov 03, 2012 3:59 am**

Hi everyone,

my workfile is set up as a dated panel structure. Let's say I have 3 crosssections for each of my two series (seriesA and seriesB) and I would like to estimate a Single Index Model (seriesA=constant + Beta*seriesB) for each crosssection. The only way I have found to do that is to change the sample for every crosssection. Therefore I would have to change the sample three times and following every change of the sample I would estimate the regression. For 3 crosssections that's no problem, but I will have between 100 and 200 in my real data set. So, is there any way that I could estimate a ls regression for each of my crosssections without changing the sample everytime?

Thanks a lot for help or suggestions.

Posted: **Sat Nov 03, 2012 5:41 am**

I think I found a workaround that is not exactly state of the art, but I get results. Here it is:

Code: Select all

    'create a panel with 3 ID series - xyz
    create a 1990 2020 3
    alpha id = "X"
    smpl if @crossid=2
    id = "Y"
    smpl if @crossid=3
    id = "Z"

    pagestruct @date(dateid) id

    'make some data series
    series seriesA=nrnd
    series seriesB = nrnd

' Define number of crosssections
scalar N=@max(crossid)

' Estimate regression for every crosssection
for !i=1 to N
	smpl if crossid=!i
	equation aaa!i.ls seriesA c seriesB
next

' Reset to full range
	smpl @all

' freeze equations
for !i=1 to N
	freeze(tab!i) aaa!i
next

' make parent table
	table cross_reg_results

' copy regression results of each crosssection regression in parent table
for !i=1 to N
	for !k=!i+3 to !i+3
		tab!i.copyrange b11 d11 cross_reg_results b!k
		tab!i.copyrange b12 d12 cross_reg_results e!k
	next
next

' Delete frozen tables
for !i=1 to N
	delete tab!i
next

' Delete estimated equations
for !i=1 to N
	delete aaa!i
next

' Include crosssection ID in parent table
for !i=4 to N+3
	cross_reg_results(!i,1)=!i-3
next

' Define headers of parent table
cross_reg_results(3,1)="Crosssection"
cross_reg_results(3,2)="Alpha"
cross_reg_results(3,3)="Std. Error"
cross_reg_results(3,4)="t-Statistic"
cross_reg_results(3,5)="Beta"
cross_reg_results(3,6)="Std. Error"
cross_reg_results(3,7)="t-Statistic"

' Define title of parent table
cross_reg_results(1,1)="Regression results of Single Index Model"
 
' show parent table
show cross_reg_results

It is really quite complicated for such a trivial problem, therefore I still hope that there is an easier solution already implemented in eViews. Maybe some of you have an idea how to solve this. Thanks a lot.

Posted: **Tue Nov 06, 2012 10:58 am**

For the 100 to 200 cross-section case, I'd just estimate the fully interacted model

Code: Select all

equation aaa.ls seriesA @expand(@crossid) seriesB*@expand(@crossid)

Alternately, you may reshape your data as a pool and use the pool tools.

EViews.com

ls estimation of each crosssection in panel data

ls estimation of each crosssection in panel data

Re: ls estimation of each crosssection in panel data

Re: ls estimation of each crosssection in panel data