Hi,
I am estimating a model with 4 equations, each of them Tobit for a Time Use Survey (US edition).
The code is the following:

cmp setup
#delimit ;
cmp (ccbasic = yoemean $x i.year $region)
(cceduca = yoemean $x i.year $region)
(ohccsec= yoemean $x i.year $region)
(twork = yoemean $x i.year $region)
if nomiss==1 & daytype==1 & edulvl>9 [aweight = myweight1mean], robust cluster(year)
ind("cond(ccbasic>0, $cmp_cont, $cmp_left)"
"cond(cceduca>0, $cmp_cont, $cmp_left)"
"cond(ohccsec>0, $cmp_cont, $cmp_left)"
"cond(twork>0, $cmp_cont, $cmp_left)");
#delimit cr

I get the model estimated without any problem for most data configurations (e.g. different days of the week, different key controls) except this one, where I get "Starting values not feasible. Numerical overflow".
My question is how to introduce starting values in this kind of model using D.Roodman's CMP procedure.
Thanks in advance for any help!