[Limdep Nlogit List] Regression and Minimize

[William Greene]

The reason that the standard errors are not the same is that 
MINIMIZE does not know that the second derivatives are supposed
to be scaled by the sum of squares.  If you want to replicate
OLS with a nonlinear optimizer, you should use NLSQ.  Even if so,
the standard errors will not be identical to OLS, but they will
be close.  NLSQ is going to use use the outer products of the
derivatives, then scale the OPG matrix appropriately.
/Bill Greene

----- Original Message -----
From: "Andreas Drichoutis" <adrihout at aua.gr>
To: "Limdep and Nlogit Mailing List" <limdep at limdep.itls.usyd.edu.au>
Sent: Monday, October 22, 2007 11:12:49 AM (GMT-0500) America/New_York
Subject: [Limdep Nlogit List] Regression and Minimize

Dear all, 
 
I'm trying to replicate the results of a multiple regression by using the
MINIMIZE command (see below the syntax I use). I get the same parameter
estimates but completely different std. errors. In specific, with the
minimize command, std. errors become extremely small and are not consistent
with the results from the regression. Why is this? (sample size is 350, if
it matters.).
 
Thanks,
Andreas Drichoutis
 
NAMELIST ; X1=ONE,gend, age, north, south, west, east, educ2, educ3, inc2,
inc3, inc4 $
 
REGRESS ; LHS=EFFIC ; RHS=X1 $
 
MINIMIZE ; FCN=(EFFIC-A1'X1)^2
; START=0,0,0,0,0,0,0,0,0,0,0,0
; LABELS=12_A $
_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au


-- 
Professor William Greene
Department of Economics
Stern School of Business
New York University
44 West 4th St., Rm. 7-78
New York, NY   10012
http://www.stern.nyu.edu/~wgreene
212.998.0876