Wild Bootstrap for NL regression

[Daniela Vidart]

Hi,

I am currently using the nl command to run a non-linear regression. I want to cluster the standard errors that come out from this procedure, and due to the small number of clusters, I want to use a wild bootstrap or the scoretest procedure proposed by Kline and Santos. However, I am having a very hard time knowing how to do this. The nl command allows for clustering and bootstrap but not for a wild bootstrap, and the post estimation command boottest does not seem to work with nl.

Any guidance regarding this would be greatly appreciated.

Thank you!

[Steve Samuels]

Welcome to Statalist, Daniela!

Be sure to read the FAQ, especially FAQ 12. In the FAQ, you are asked to give full references, no just short versions like "Kline and Santos", and you are also asked to give the source of contributed commands like boottest (e.g. "by David Roodman, available at SSC")

Douglas Miller's web page links to bs_sample.do a do file that, among other things, performs a wild bootstrap. Try modifying this file to use nl instead of regress. The weights implemented in the do file are Rademacher weights. According to the help file for boottest, you can improve on these.

[Steve Samuels]

Correction: name of Miller's do file is bs_example.do.

[Daniela Vidart]

This is very helpful. Thanks a lot!