Measured confounders when using ivsvar

[Abi Stevely]

Hi,

I am designing an instrumental variable svar analysis and my instrument is likely associated with the outcome through both endogenous and exogenous (seasonality) measured confounders.

I am not sure if I can control for this by including my endogenous confounder as a dependent variable and months of the year as exogenous dummies? I would control for measured confounders in a typical IV analysis, but I have seen some information suggesting that the instrument has to be exogenous in IVSVAR. I have not been able to figure it out myself from the command documentation.

There is also some risk that lags of my target shock variable predict the instrument. Will lags of the target shock variable be included in the equation for the instrument?

Thank you for your help.

[Abi Stevely]

I have an answer from the developer, so posting it here in case it is useful to anyone else.

Regarding my first question (on controlling for a measured confounder), the developer wrote: “the instrument has to be exogenous”.

With regards to my second question (on lags of the target shock variable), the developer wrote: “For the -ivsvar mdist- estimator, lags of the target shock variable can be included in the instrument equation with the undocumented option -zexog()-. For the -ivsvar gmm- estimator, only lags of the instrument itself are allowed in the instrument equation, via the -zlags()- option.”