RE: st: eclass program based on cmp

["David Roodman (droodman@cgdev.org)" <DRoodman@cgdev.org>]

Use "matrix list" to display your matrices before posting, then debug.

If you can send me a good example of standardization improving stability, I may well at it as a feature to cmp.
Best,
--David


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] 
Sent: Wednesday, October 24, 2012 2:33 AM
To: statalist-digest@hsphsun2.harvard.edu
Subject: st: eclass program based on cmp
Dear members,

This is follow-up of a previous question to this forum. I want to construct an eclass program that estimates a multiequation model using the cmp routine (Roodman 2012). I have estimated using a multiequation model using cmp directly, but the procedure becomes numerically unstable. I figure out that by standardizing the regressors, the performance improves a lot. The problem is that I want to recover the parameters in its original scale after estimation and store them in the usual e(b), e(V) and other eclass scalars.
I posted my program at the end of this message. When tested, I got I am rookie in program writing in Stata, so I got the following message

estimates post: matrix has missing values
r(504);

Moreover, even though I get the e(b) and e(V), they do not include the re-scaling I'm trying to make.

Even though I have some experience in do-file programming, I'm a rookie on eclass programs. 
Any guidance in order to solve this problems will be appreciated

Best,

Benjamin

*******************************************************************

cap program drop ecmp
program ecmp, eclass
version 11.2
syntax anything(equalok) [if] [in] [fw pw aw iw] [, *]
tokenize "`anything'", parse("()= ")

local num=1
local cvar=1
local sanyth
local listvar
local prev
while "``num''" !="" {
*Standardize variables except the dependent one - first after "(" -
local term "``num''"
if  "``num''" !="(" & "``num''" !=")"  & "``num''" !="=" & "`prev'" !="(" {
qui summ ``num'' `if' `in' `weight'
tempname mean``num''
sca `mean``num''' = r(mean)
tempname sd``num''
sca `sd``num''' = r(sd)
tempvar s``num''
cap gen `s``num''' = (``num''-r(mean))/r(sd)
local vv_`cvar' ``num''
local ++cvar
local term "`s``num'''"
local listvar `listvar' ``num''
}
local sanyth "`sanyth' `term'"
local prev "``num''"
local ++num
}

di "`listvar'" 
di "`sanyth'" 
tempname MEAN
matrix `MEAN'=J(`cvar',1,0)
tempname SD
matrix `SD'=`MEAN'
local k=1
foreach vv in `listvar'{
matrix `MEAN'[`k',1]=`mean`vv''
matrix `SD'[`k',1]=`sd`vv''
local ++k
}

  cmp `sanyth' `if' `in' `weigth' `exp', `options'

  tempname newb newV oldll oldN newll newN
  matrix `newb' = e(b)
  matrix `newV' = e(V)
  forv k=1/`cvar'{
matrix `newb'[1,`k'] = `newb'[1,`k']/`SD'[`k',1]
forv j=1/`cvar'{
matrix `newV'[`k',`j'] = `newV'[`k',`j']/`SD'[`k',1]/`SD'[`j',1]
}
}

ereturn post `newb' `newV'
end

*
*   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/