st: Re: overidentification test after reg3

[Kit Baum <baum@bc.edu>]

It is not apparent that this system is simultaneous. The dependent  
variables are dw1,dw3, dw3, exp, p1, p2, p3, p4 = 7. I do not see any  
of those variables on the RHS of other equations. If that is the  
case, there is no simultaneity here, and each equation can be  
estimated by OLS (or, with constraints across equations, as a SUR via  
sureg).

If there is no simultaneity, there is no question of identification,  
and thus no need for a test of overidentifying restrictions. It is  
not surprising that the routine referenced should come up with a huge  
number of d.f., as the number of exogenous variables is huge, and  
that will be multiplied by 7. By my count you're estimating 48 RHS  
coefficients (plus constant terms) minus 6 constraints, or 42.  
Somehow, I knew the answer should be 42.


Kit Baum, Boston College Economics
http://ideas.repec.org/e/pba1.html
An Introduction to Modern Econometrics Using Stata:
http://www.stata-press.com/books/imeus.html


On Sep 28, 2006, at 2:33 AM, Valerie wrote:


>  am trying to run an over identification test for a constrained
> simultaneaous model (with reg3). It is the  following :
>
> /*symmetry*/
> constraint 1 [dw1]dlnp2 = [dw2]dlnp1
> constraint 2 [dw1]dlnp3 = [dw3]dlnp1
> constraint 3 [dw2]dlnp3 = [dw3]dlnp2
> /*homogeneity*/
> constraint 4 [dw1]dlnp1 + [dw1]dlnp2 + [dw1]dlnp3 + [dw1]dlnp4 =0
> constraint 5 [dw2]dlnp1 + [dw2]dlnp2 + [dw2]dlnp3 + [dw2]dlnp4 =0
> constraint 6 [dw3]dlnp1 + [dw3]dlnp2 + [dw3]dlnp3 + [dw3]dlnp4 =0
>
> reg3(dw1  ddeflexp dlnp1 dlnp2 dlnp3 dlnp4  period  dlnyear) /*
> */  (dw2  ddeflexp dlnp1 dlnp2  dlnp3 dlnp4   period  dlnyear)/*
> */  (dw3  ddeflexp dlnp1 dlnp2 dlnp3 dlnp4  period  dlnyear) /*
> */  (exp  explag PIB period)/*
> */  (p1   pxlaitcomteinterp  pxlaitcomteinterplag1     ind_sal_interp
> pourc_vol_ean_1 pourc_vol_MDDHD_1)/* */  (p2    pxlaitstandinterp
> pxlaitstandinterplag1  ind_sal_interp  pourc_vol_ean_2  
> pourc_vol_MDDHD_2)/*
> */  (p3   pxlaitcomteinterp  pxlaitcomteinterplag1   pxlaitstandinterp
> pxlaitstandinterplag1  ind_sal_interp  pourc_vol_ean_3  
> pourc_vol_MDDHD_3
> pourc_vol_parm pourc_vol_beauf) /*
> */  (p4   pxlaitstandinterp  pxlaitstandinterplag1  ind_sal_interp
> pourc_vol_ean_4 pourc_vol_MDDHD_4),    cons(1-6)
>
> The last 5 equations correspond to my instrumentations.
> As it isn't possible to use the overid command (because I used reg3  
> and
> not ivreg2), I have to rebuild the sargan test of overidentification.
> I have found a the little and useful program written by Christopher  
> Baum
> to rebuild the  J stat. I just don't well understand the degrees of
> freedom calculated in his program (df = #eqns * #exog - #betas  
> estimated
> (net) =  rowsof(esigma) * colsof(w) - (colsof(beta) - `nconstraint')).
>
> With this, I obtain a df of 150 which is enormous. I'm probably  wrong
> somewhere. If someone has any suggestions....
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/