st: to hijack the stata ml?

[Sun Yutao <yutao.sun.statalist@outlook.com>]

Hi all,

I’m wondering if there is a way to “hijack” the stata ml command, sorry for using this word but I have a very big panel probit with fixed effects and lots of cross-sections to maximize (I know the incidental parameter problem and let’s don’t talk about that right now)

What I’m thinking is a partitioned mle which I either have to set constraints on some of the parameter in every iteration (e.g. the leapfrog) or to use a different approach of updating the parameter vector (e.g. Chamberlain 1980)

Both can be done, of course, but with problems: For the first method, the number of constraints are 1999 in stata se 12. I can put constraints in a matrix though, but it can be a huge matrix and it’s time consuming to operate it. For the second method, I can use mata to bypass the stata ml routine but then I think I will lose a lot features from the stata ml command, which I don’t really like.

I'm wondering if there is a way to "hijack" (sorry for this ugly word again :) ) the stata ml so it computes the derivatives and hessian, and updates the parameter vector in my own way but does everything else conventionally.

It will also be appreciated if someone can provide me a detailed document (really detailed) on the stata ml ado package, because I tried to read it but it's really complicated.



Best regards,
Sun Yutao




*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/