combine instrumental variable model with panel data model

[Johnny YANG]

Hello everyone,
I'm conducting tests on the effects of companies' international expansion on the values of companies. I used two models, instrumental variable model and panel data model (fixed vs. random effects model). The results are not converging to each other. The IV model indicates that the effect of internationalization on firm value is negative, while the panel data (fixed effect) model indicates that the effect is positive. Thus, right now I'm considering to use xtivreg to combine the two models. But I don't know if it will make the interpretations of my results richer. And another question is that with xtivreg, stata doesn't test for the overidentification, weak instrument. So what should I do to solve this?
Thank you very much for helping me in advance!

[Joro Kolev]

If your instrument is time varying, it does not cost you anything to try -xtivreg- and to see how results come out.

You can look at the first stage for the strength of your instrument. With only one instrument you cannot test for overidentification.

[Johnny YANG]

If your instrument is time varying, it does not cost you anything to try -xtivreg- and to see how results come out.

You can look at the first stage for the strength of your instrument. With only one instrument you cannot test for overidentification.

Hi Joro, Thank you for replying! My instruments are time varying. But I used several instruments, so how should I test for the underidentification, and overidentification? As for weak instrument, I can just look at the coefficient significance, if I understand you correctly?

[Johnny YANG]

If your instrument is time varying, it does not cost you anything to try -xtivreg- and to see how results come out.

You can look at the first stage for the strength of your instrument. With only one instrument you cannot test for overidentification.

Actually, one of my key independent variables is not time-varying. In this case, what should I do?

[Joro Kolev]

Actually, one of my key independent variables is not time-varying. In this case, what should I do?

If your independent variable is not time varying, you cannot use fixed effects methods; both -xtivreg, fe- and -xtreg,fe- will wipe out the time constant variable.

[Joro Kolev]

Hi Joro, Thank you for replying! My instruments are time varying. But I used several instruments, so how should I test for the underidentification, and overidentification? As for weak instrument, I can just look at the coefficient significance, if I understand you correctly?

For overidentification test check out the user written -xtoverid-.

For strength of the instruments, look at the F/Wald statistic in the first stage for the joint significance of your instruments.

[Johnny YANG]

For overidentification test check out the user written -xtoverid-.

For strength of the instruments, look at the F/Wald statistic in the first stage for the joint significance of your instruments.

That's great! Thank you so much!

[Johnny YANG]

For overidentification test check out the user written -xtoverid-.

For strength of the instruments, look at the F/Wald statistic in the first stage for the joint significance of your instruments.

Hi, Joro, I have three following up question actually. Since I want to do a fe and re regression with xtivreg. I need also to test the overidentification IVs in the re model. Does the xtoverid still work the same for random effect model?
Second question is that I need to compare xtivreg, re with xtivreg, fe , in order to select a better estimation. Can I use hausman test still?