 Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# st: Question about FE/RE and IV

 From "." To statalist@hsphsun2.harvard.edu Subject st: Question about FE/RE and IV Date Wed, 7 Sep 2011 10:45:53 +0200

```Dear Statalist,

I would like to ask you a question about doing fixed/random effects
estimation with instrumental variables. Thank you very much for your

In the FAQ I found many examples of how to do FE/RE with instrumental
variables. My situation is a little bit different, however, and I am
not sure that I am doing the right thing.

I have the following system of 2 structural equations:

GDP = R&D + Capital + Labour +...+ Materials
R&D = GDP + Subsidies +...+ Labour

where GDP and R&D are endogenous; capital, labour, materials and
subsidies are assumed to be exogenous. I have a panel of 22 companies
over a period of 3 years.
So far I have estimated both equation with different methods (pooled
OLS, fixed and random effects, 2sls). Now I would like to estimate
them with fe/re and instrumental variables.

The problem is that I do not have any ready to use instruments for GDP
and R&D. So I am considering to use the predicted values of both
variables and include them into xtreg, fe and xtreg, re. Could you
tell me please if the following steps are correct?

Step 1. In the first step I regress GDP on all exogenous variables
from the first and second equation:

reg GDP Capital Labour Materials Subsidies, robust

Step 2. Predict the value of GDP based on step 1.

Predict p_GDP

Step 3. The same procedure is repeated for R&D.

Step 4. I do fixed and random effects on both equations. In the first
equation I use the predicted value of R&D instead of R&D and in the
second equation I use the predicted value of GDP instead of GDP.

xtreg GDP p_R&D Capital Labour Materials, fe i(firm)
xtreg p_GDP Subsidies Labour, fe i(firm)

I estimate both equations with fixed and random effects and then
choose between them based on the Hausman test. Could you tell me
please if the above steps are correct?

Another way to do this, as suggested on your website, is to use the
“xtivreg” command. I am not sure however whether I can use this
command in case that i have the predicted value of GDP and R&D as an
instrument.

xtivreg GDP Capital Labour Materials (R&D=p_R&D), fe i(firm)

It is a kind of long message but I hope that it will be useful also