Formulate regression TSLS

[Student123]

-

[startz]

Unless something very special is going on, you can't use one endogenous variable to explain another.

[Student123]

Thank you for your input, but it seems that it is done like this.

So could you help me out defining the Instrument Variables, even though you do not think it's appropriate?

[startz]

By definition, TSLS uses the same instruments for each variable. There's no way around this in EViews.

[Student123]

By definition, TSLS uses the same instruments for each variable. There's no way around this in EViews.

Uhm that does not make it easier.
I am on vacation, so that's why I am responding this slow.

In Pathan (2009) they do a 3SLS with the regressions from the first post.
Is it possible to do a 3SLS with the regressions from the first post in EVIEWS? How? What instruments do I need to define?

Another question: what if I use the predicted values of the regressions 2,3 and 4 in regression 1.
This probably gives biased results, but can this be adjusted?

Many thanks for your responses startz! Hope you can help me out with these questions as well (or any other person).

Greetings

[Student123]

Please, anyone?
Still more than interested!

[Student123]

Please?
I am still more than interested!

[Herbo]

Hello Student123

To convince you the previous replies are valid, I am pasting parts of page 190 from Kit Baum's Econometrics book.
This should assure you that you cannot pick different variables as instruments for each endogenous variable.

The Stata commands are not particularly relevant here.


In a situation with multiple endogenous regressors such as
. ivreg y x2 (x3 x4 = za zb zc zd)
novice users of instrumental variables often ask, "How do I tell Stata that I want to use
za, zb as instruments for x3, and zc , zd as instruments fo! x4?" You cannot, but not
because of any limitation of Stata's ivreg command. The theory of ~2SLS estimation
does not allow such designations. All instruments-included and excluded-must be
used as regressors in all first-stage regressions. Here both x3 and x4 are regressed on z:
x2 za zb zc zd and a constant term to form the k matrix.


Herbo

[startz]

It is true that in two-stage least squares all the instruments are used for each variable. I see no reason in principle that one couldn't have an instrumental variable estimator that used different subsets. I don't know why you would want to. And I don't know of any software that will do it. But it's probably possible.

[Student123]

Ah guys, it is really appreciated that you try to answer my question!
But it is getting really confusing!

Please have a quick look at Pathan (2009). The link is http://faculty.ndhu.edu.tw/~sywang/a17.pdf.
On page 7 (or 1346) a robustness check is performed by using 3SLS. He is doing the exact same thing as I want, doesn't he? His instruments are not the same for each variable, are they?
I understand that this is 3SLS and not 2SLS as I asked. But can anyone tell me wether and how a 3SLS test as in Pathan (2009) can be done in Eviews??

Many thanks and even more in advance!

[startz]

If i understand what he did, and I'm not sure I do, his instruments are same for each variable. All the exogenous variables are used as instruments.