Cross-sectionally invariant IV for a panel data set

[Dmitrii Petrukhin]

Hi,

I am thinking of assembling a panel dataset with locations over time. If I were to use an Instrumental Variable, is it possible for my instrument to only vary over time but be constant for the cross-section over a year? Eg if my IV is something like the yearly GDP of a neighbouring country, and my cross section are the counties of my country of interest. So the IV variable for year t will be the same for all n regions.

Also please note that I have data for 10 years. So my IV will have 10 values, would this be an issue?

I think it should work(?) as it is conceptually similar to a Wald estimator (binary IV)

Are there any papers that discuss this?

Kind regards,
Dmitrii

[Sebastian Kripfganz]

If you want to include year dummies, which is normally recommended, then such a cross-sectionally invariant variable would be perfectly collinear with the set of year dummies.

[Dmitrii Petrukhin]

If you want to include year dummies, which is normally recommended, then such a cross-sectionally invariant variable would be perfectly collinear with the set of year dummies.

Hi Sebastian,

Yes, you are right I just realized this. I guess I wont be able to implement an IV in this specific case.