xtivreg2 - causal interpretation

[Renira Corinne Angeles]

Dear users,
I'm unsure about the interpretation of the results of my xtivreg2 etsimation (Stata 18). I want to estimate x1 on y with controls - the main interest of variable is x1. I have been adviced to use dynamic panel estimation due to endogeneity of l.y.

The command is specified as the following:

 xtivreg2 y x1 x2 x3 y_trend y_trend2 (l.y=l2.z l3.z ), fe robust small first  endog (l.y)

I am wondering how to interpret the causality here. I get how to interpret the specific IV test - inthis case showing strong instruments and detection of endogeneity of l.y. Further, the effect of x1 (main interest) is still significant when using xtivreg2.
Is it correct that the only causal effect I can be sure of here (based on the results of good instruments and endogeneity) is the lagged y since (l.y=l2.z l3.z ) - which is good since I have handled the endogeneity of l.y. Then hat part is solved. But, as mentioned, the main interest is the effect of changes of x1. Can I say that x1 is also causal in this model? I am guessing no - and that I have to do an own IV-specification for x1 as well (or perhaps in the same specification of l.y). Is that right?

I hope this was understable.

Many thanks for you input, and apologies in advanced if this is not a typical Statalist question (not very technical).

Kind regards,
Renira

[Carlo Lazzaro]

Renira:
as you surmised, you cannot say that -×1- is causal. In addition, once you have successfully dealt with the endogeneity, you cannot establish a causal relationship with the regressand, unless your intrument resists the exclusion restriction.

[George Ford]

xtabond since you're lagging y?

[Renira Corinne Angeles]

Thanks the very helpful input, Carlo. According to the tests, the instruments are "good/strong"(?) or "correctly included" (according to thresholds given in the literature).

George: I will also run my models with xtbond2 - although, I have N=31 and t=49. I think it is much easier to understand and interpret xtivreg2 to be honest. For instance, with my panel, there will be many instruments. But I understand I can use the collapse command.