Dear Statalist

I would like to estimate an IV regression with a high-dimensional set of controls using -ivlasso- (by Achim Ahrens, Christian Hansen, and Mark Schaffer, part of the package -pdslasso-, available from SSC). I have a single endogenous variable, interacted with time-since-treatment dummies, and a single instrument, also interacted with time-since-treatment dummies. This is just-identified, so I would like to use the Lasso to select from among the available control variables only, not the instruments. However, ivlasso appears to think I am trying to select a subset of the instruments, so I sometimes end up with fewer instruments than endogenous variables. The problem appears to particularly affect the post-double-selection estimates, though I don't know if this is specific to the methodology.

My main question is: is there a way to ensure that my specified instruments are always included in the first-stage regression? If not, my follow-up question is, is there an econometric reason for this, e.g. is it invalid to use ivlasso methods with multiple low-dimensional instruments and endongenous variables?

To give an example of the syntax I am using and the output I am looking at, here is some sample code:

Code:
use https://statalasso.github.io/dta/AJR.dta, clear
ivlasso logpgp95 (lat_abst edes1975 avelf humid* steplow-oilres) (avexpr temp1-temp2=logem4 temp3-temp4), ///
    cluster(democ1)
Note that in this example, the problem goes away by removing the clustering, though this is not the case with my data.

Thanks for any help or suggestions