Page 1 of 1

Mixed equality and inequality restrictions

Posted: Wed Jun 16, 2010 3:24 am
by simonius
Hi,

Could anyone provide me some clues of what I'm doing wrong?

I'm trying to estimate an equation with the restriction that my coefficients are all non-negative and sum up to one. After reading several post and the user guide, I still don't get the solution.
I tried:
y = C(1)+(EXP(C(2))/(EXP(C(2))+EXP(C(3))))*x1+ (EXP(C(3))/(EXP(C(2))+EXP(C(3))))*x2
and
y =C(1)+(1/(1+EXP(C(3))))*x1+ (EXP(C(3))/(1+EXP(C(3))))*x2
(here I tried to apply the suggestion of Glenn, http://forums.eviews.com/viewtopic.php? ... aint#p1887 )
However, for both equation I get a this WARNING: Singular covariance - coefficients are not unique
I do want to have a constant in the regression.

To be sure, I added my workfile, where eq1 and eq2 are the estimated equations above respectively

Many thanks in advance, Simon
eviews 6 user

Re: Mixed equality and inequality restrictions

Posted: Wed Jun 16, 2010 6:51 am
by startz
You're just having bad luck. Reset the coefficient vector to zero and re-estimate the equation.