Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Re: Two way Fixed Effects & Autocorrelation& Heteroskedasticity


From   "Rodrigo Alfaro" <ralfaro76@hotmail.com>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: Re: Two way Fixed Effects & Autocorrelation& Heteroskedasticity
Date   Thu, 30 Mar 2006 15:27:54 -0500

Dear Max

I have some comments about your question:

(1) If you think that the error term has autocorrelation and heteroskedasticity I am not sure if Hausman test is still valid. My personal approach to this problem is not to use Hausman test and try to solve the possible correlation between unobservable and explanatory variables. Random Effects assumes that there is not correlation.

(2) Suppose that you decided to use Fixed Effects (FE), the fact of the error term has autocorrelation could be the result of misspecification in your model. In other word, maybe you model need a lag dependent variable in the right hand side. In that case FE is biased and you have to choose between try to correct the bias (Hahn-Kuersteiner, econometrica 2001 or Kiviet, journal of econometrics 1995)... here http://ideas.repec.org/c/boc/bocode/s450101.html you can find the work of Bruno based on Kiviet paper or use a Instrumental Variable approach (Anderson-Chiao, econometrica 1980? or Arellano-Bond, review of economic studies 1991) commands -ivreg- and -xtabond-.

(3) Finally, maybe you know that there is not lag dependent variable in your model but you still have errors with autocorrelation and heteroskedasticity. Well, the command -xtregar- solves the problem for the case of autocorrelation of order 1 in your model, but this solution is Prais or Cochrane solution for time series, which means that you estimate the model as you have the lag dependen variable in the right hand side, plus the lags of the exogenous variables in the context of a nonlinear constrained model. Details are available in autocorrelaton chapter of any textbook. In conclusion, you will have the lag dependent variable...

(4) A way to solve the problem without a lag dependent variable is using a standard error correction that takes care of the behavior of the error term but does not change the Least Square estimation of the parameters. You can use -newey- with manually fixed effect, but you have adjust the degree of freedom (I wrote a command for that I can send you if you are interesting).

Rodrigo.


----- Original Message ----- From: <kosak@gmx.de>
To: <statalist@hsphsun2.harvard.edu>
Sent: Thursday, March 30, 2006 11:11 AM
Subject: st: Two way Fixed Effects & Autocorrelation& Heteroskedasticity



My problem was already discussed before,but i didn`t found an appropriate
solution to it.-
I have a big panel data set (1500 observations over 20 years).I want to
estimate a model with time and group specific fixed effects (Hausman Test
performed). The Model suffers from autocorrelation and
heteroskedasticity.The problem is that the cluster option is not possible
for xtregar. Can i use xtgls or areg?As i understood xtgls estimates a
random effet model.
Any suggestions?
Max

--
"Feel free" - 10 GB Mailbox, 100 FreeSMS/Monat ...
Jetzt GMX TopMail testen: http://www.gmx.net/de/go/topmail
*
*   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/



© Copyright 1996–2020 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index