Appendix A.2 of the paper about boottest describes an opportunity to speed up the program's construction of confidence sets for one-dimensional hypothesis tests after OLS (or after IV estimation when performing the Anderson-Rubin test). I have finally, mostly exploited that opportunity, which is about taking advantage of the knowledge that the computations, while sometimes complex, are to a great degree linear in the trial value for the parameter being tested.

The main benefit is for certain cases where boottest still took significant time. I think there are two: where there is multiway error clustering, and the subcluster bootstrap. These can involve large numbers of bootstrap or error clusters. The paper includes an example of applying the subcluster bootstrap to a treatment effects regression with few treated clusters. The slowest regression in that example now takes a minute on my machine rather than 10 minutes.

The surgery to the code was substantial, so let me know if you spot problems.

Install with
Code:
ssc install boottest, replace