xtdpdml convergence

[ely mattos]

Dears,

I am having trouble to estimate a model using xtdpdml. I am using a panel counting on 13 years and 400 observations - strongly balanced. Here is the command I'm using:

xtdpdml tx_ab_em L.desp_educ L.desp_saude L.desp_seg, pre(pop_perc_60) difficult technique(bhhh 50 nr 50 dfp 50 bfgs 20)

When I set trace on, I can see that the code return "convergence failed" after reporting the coefficients table. It seems that Stata is experiencing difficuties to run pos-estimation tests...

Some other times, it stops eve before the coefficients table and displays this:

        ---------------------------------------------------------------------------------------- end sem_parse_spec ---
      - local display_opts `s(display_opts)'
      = local display_opts 
      - mata: st_sem_fit()
cannot compute an improvement -- flat region encountered

Any clue about what's happening? Any tip?

Thanks!
Ely

[Richard Williams]

Try simplifying the model. Use fewer variables and/or fewer time periods. T=13 is a lot for xtdpdml. Section 5.2 of

https://www3.nd.edu/~rwilliam/dynamic/SJPaper.pdf

offers various tips for addressing convergence problems.

[ely mattos]

Thank you Professor Williams. I will try fewer time periods.