Dear all,
I have a question that is both technical (Stata) and theoretical. Based on the help file for npregress, and some hands-on replication, it seems that when we request bootstrap standard errors, the default is to use the original sample Cross-validation Bandwidth. We can, however, also request for the bandwidth to be estimated for each individual bootstrap sample.
Does anyone know about the consequences of recalculating the bandwidth for each bootstrap in terms of the standard errors and confidence intervals?.
The main draw back is the computing time, but i could not find anything regarding other consequences.
Thank you
Fernando