What's the difference among xtivreg2,xtivreg and xtabond2?

[Zhang Hui]

Hi, friends, what's the difference among xtivreg,xtivreg2 and xtabond2? Can I apply ivregress command to panel data?

[Amin Sofla]

For the xtivreg and xtivreg2, you might want to take a look at here .
xtabond2 is used for the dynamic panel estimator.

[Carlo Lazzaro]

Zhang:
as an aside to Amin's helpful reply, please note that xtivreg2 and xtabond2 are both user-written commands.
If your panel has a continuous dependent variable, take a look at -help xtivreg- for instrumental regression.

[Zhang Hui]

Thank you for Amin and Carlo's reply. I also want to know whether "ivregress"/ "ivreg2" can be applied for panel data?

[Carlo Lazzaro]

Zhang:
-ivregress- is not the first choice for instrumental-variable regression performed on panel data; see -xtivreg- instead.
That said, -ivregress- can be applies to pooled OLS (with standard errors clustered on -panelid-) performed on panel dataset; however, please note that rarely pooled OLS outperforms -xt- commands when it comes to panel data analysis.
The same holds for -ivreg2-; -search ivreg2- will give you all the details concerning this user-written programme.

[Zhang Hui]

Thank you for Carlo's reply and helpful advice. My command is

ivregress 2slsl indepvar depvar [endo = iv i.region i.year] , vce(robust)

Is there anything incorrect? Is it right to take time and area fixed effects into account in this way?

Thank you in advance.

[Carlo Lazzaro]

Zhang:
I wold only replace -vce(robust)- with -vce(cluster clusterid) in your code, as you're performing a pooled instrumental OLS.
However, one issue still remains in the background: are you sure that -regress- outperforms -xtreg- for your data?

[Zhang Hui]

Thank you for Carlo's reply and helpful comments. Right, “xtivreg" is better. But could I test the endogeneity of IV, weak IV and etc. after using the command “xtivreg"? Could you give me some information?

I just know the "xtoverid" can be applied. And enough test information could be implemented after using "ivregress", like "estat firststage".

Thank you a lot.

[Carlo Lazzaro]

Zhang:
I would go -xtoverid-; all the information you're probably interested in are reported in the related help file.

[Zhang Hui]

Thank you for Carlo's reply and help. But I found xtoverid can only test the over identification problem of IV.
How could I test the endogeneity of IV, weak IV after using “xtivreg"? Sorry. Thank you.

[Carlo Lazzaro]

Zhang:
have you already taken a look at -search xtivreg2- and related help file?

[Enza Testa]

Hi Carlo, to avoid endogeneity problem in my sample I use xtivreg2 choosing two IVs and xtabond2 using as IVs the first two lags for each endogenous variable (I have a growth regression and it is plausible having a dynamic setting). Unfortunately the lagged dependent variable is not significant while the regressors are significant. It makes any sense to use the dynamic setting in this case? The regressors are strongly significant but I'm wondering if this was enough to justify the use of AB. Thanks a lot!!!

[Carlo Lazzaro]

Enza:
take a look at https://blog.stata.com/2015/11/12/xtabond-cheat-sheet/
For the future, I would recommend you to start a new thread. Thanks.

[Enza Testa]

Thanks Carlo and sorry for replying to this post but sometimes I read about the proliferation of the posts without searching in the contents already available so I use this thread. Before I made a mistake, I used the second and the third lag. I read the article you mentioned and several ones related to the xtabond. I did all the tests and everything is ok exept for the significance of the lagged dependent variable. It is justifiable in my case using an AB estimator?

[Carlo Lazzaro]

Enza;
I would run -estat abond- and -estat sargan- to ahave a comprehensive picture.
See also http://econometricstutorial.com/2015...el-data-stata/