Ulrich,
you helped me a lot - thank you very much. Your second hint to fix the
dummies did the job.
Kind regards to Berlin :) Niko
-----Urspr�ngliche Nachricht-----
Von: [email protected]
[mailto:[email protected]] Im Auftrag von Ulrich Kohler
Gesendet: Mittwoch, 14. September 2005 11:58
An: [email protected]
Betreff: Re: st: Out-of-sample prediction: Time dummy problem
Niko Wrede wrote:
> I estimate a model using an in-sample-dataset (e.g. 1989-1999)and the
> xtabond2 command with time dummies. Then I want to predict and
> calculate residuals using an out-of-sample-dataset (e.g. 2000-2004)
> for another time period.
>
> My problem: Using predict and the out-of-sample-dataset, STATA
> complains, that some variables are missing in the dataset, since it
> cannot find the time dummies.
>
> Question: How can I predict using only the estimates of the main model
> variables without the estimates for the time dummies?
>
> My idea was:
> 1. Estimate:
> Use insample.dta
> Xi: xtabond2 depvar var1 var2 i.year (endogvar, lag(2 .)) ivstyle(var1
> var2
> i.year) rob twostep
>
> 2. Predict:
> Use outsample.dta
> ... and then some kind of adjust or xpredict ... But it is not working
> so far and I am not sure, if this is the right way.
Niko,
you have generated dummy-variables for year on the fly by prefixing
-xtabond2-
with -xi:-. Therefore you need to reproduce this step also in the outsample:
. use insample
. xi i.year
. xtabond2 depvar var1 var2 I* ...
. drop _all
. use outsample
. xi i.year
. predict [whatever]
Note that you do not necessarily need to break the model estimation and the
generating of the time dummies for the insample-step. I only did this to
show
the logic.
Your question implies that you don't want to use the year dummies in the
prediction. I guess that this means to "fix" the year to the reference-year.
In this case you might want to -replace- all year dummies to zero before
prediction, i.e.
. foreach var of varlist I* {
. replace `var' = 0
. }
. predict [whatever]
I am not familiar to -xtabond2-, but I suspect that it will also generates
variables holding lags of the endogenous variables. If so, you will also
need
to generate these variables by hand in the outsample, i.e. something along
the line of
. gen varname = l2.varname
hope that helps
Uli
--
[email protected]
+49 (030) 25491-361
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
