IV regression syntax (different sample periods)

[HyunSahn Ko]

Hello,

I am trying to run IV regressions in STATA. I use xtivreg or ivregress with year and industry-fixed effects.

My question is that the results are very sensitive to the choice of different sample periods (e.g., whether I start my sample in 2011 or 2012 matters a lot).
I believe that this should not be the case (I have available observations from 2010 to 2021, about 15,000 observations, well-distributed among different years).

Here is the syntax I use.

xtivreg ch_abs log_size btm i.sic2 i.FYEAR (m_b_ch_ab=m_b_focal_1), fe vce(cluster GVKEY)

m_b_ch_ab is the endogenous variable and m_b_focal_1 is the instrument.

Please let me know why I find the results very sensitive to the choice of the sample period.

Best,

HyunSahn

[Carlo Lazzaro]

HyunSahn:
1) it might be a matter of missing values;
2) without further details, your -timevar- should stretch over 2010-2021 (with 2010 omitted being the reference category). Therefore, I do not know whether is legal to decide to start from 2011 or 2012 (and on the grounds of what).

[Joro Kolev]

OP, you seem to think that you are talking to the Jedi knights council... this is only Statalist, and to answer questions such as "I expect this but I get that" without seeing any data or running the regressions oneself, the one answering has to be well-versed in using the force.

As I commented on another thread you started, you have to show how you xtset your data when you are showing the xt command you are executing.

But assuming that you xtset on GVKEY, what is unusual in your regression is that you include industry fixed effect on the top of GVKEY fixed effect. This is very brutal as your identification will come only from firms that change industries, which happens almost never.

Try running your regression without the industry fixed effect and see whether they are still unstable.