Dear Statelist

I am working with a panel dataset consisting of more than 100 potential predictors. In the absence of well-established theories as to which variables are a priori better determinants for the outcome variable, I would like to implement some statistical procedure to identify a smaller set of predictors (no more than, say, 15) as a starting point.

I read the manual for -nestreg- and found that it does not support -xtreg- commands. My questions are:

1) Can someone please provide some intuition as to why Wald or Likelihood ratio tests (as implemented by -nestreg-) do not work for panel commands?

2) If there are alternatives I can explore.

Thanks