@marcfbellemare Most methods I've encountered break up the treatment into semiparametric bins to sidestep the issue (re: bunching).

I thought there would be more in the Panel GMM space, but most of the methods I find are the workhorses with which you are familiar, just converging faster.