[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

From |
orozco@toulouse.inra.fr |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: st : how to treat endogeneous variables with NLSUR |

Date |
Wed, 07 Nov 2007 16:13:13 +0100 |

Thank you for your reply AbdelRhmen, In fact, my example was just a small one to illustrate my question shortly. The idea wasn't to interpret it, but just to have an idea of my methodological problem. It is true that in reality I have in mind the AIDS system that I used to estimate with "reg3" when I took the linear index of prices (stone index), but now, I want to estimate the model with the nonlinear one, that's why I am interested in the NLSUR command. I want to instrument Prices and expenditures. Thank you. valérie /*********************************************************/ I'm not a user of Stata 10 but i have a few remarks on this: 1) It looks like an estimation of AIDS (Almost ideal demand system) and i don't understand why you try to instrument log price 2. 2) the theorie bhind the specification of price and it's square lnp2 and lnp2^2 as a regressors in item 2 of your system is not clear. 3) Usually what is instrumented is the total expenditure (lnexp) price are supposed exogenous 4) I guess that NLSUR can not treat endogeneity problem 5) conditional on your price index lnP you have a linear system of equation and you could estimate it by reg3 6) Finally if a suppose that your specification is correct you still have a linear system just generate P2=lnp2^2 and replace it on your system HTH AbdelRahmen Selon orozco@toulouse.inra.fr: > > Hi, > > I already posted my question last week, but I didn't receive any ideas of > answer. Maybe I was unclear, so I try again to explain. > > I just would like to know if it is possible to treat the problem of > endogeneity > estimating a system of equations (more especially using NLSUR). > In fact, I suppose some variables to be endogeneous but I don't know how to > specify my equations of instrumentation (differently from my main system) and > test the validity of my variables (like with the overid command after > ivreg2). > > I tried to write a little program estimating the following system of 2 > equations : > w1 = a1 + a11 lnp1 + b1 lnp2 + b11 * (lnexp - lnP) > w2 = a2 + a12 lnp2 + b2 lnp2^2 + b12 (lnexp - lnP) > with lnP = a1 lnp1 + a2 lnp2 + a11*a12 + lnp1*lnp1 + lnp1*lnp2 + lnp2*lnp1 + > lnp2*lnp2 > > and I want to instrument the variable lnp2 that I suppose to be endogeneous > (with 2 instruments : expfd and p1 for example). > But it doesnt work and I have no idea how to correctly specify it. > > Thank you for trying to help me. > > valérie > > > capture program drop nlsurtest > > program nlsurtest > version 10 > syntax varlist(min=5 max=7) if, at(name) > tokenize `varlist' > args w1 w2 lnp1 lnp2 lnm expfd p1 > > tempname a1 a2 a11 a12 > scalar `a1' = `at'[1,1] > scalar `a2' = `at'[1,2] > scalar `a11' = `at'[1,3] > scalar `a12' = `at'[1,4] > > tempname b1 b2 b11 b12 > scalar `b1' = `at'[1,5] > scalar `b2' = `at'[1,6] > scalar `b11' = `at'[1,7] > scalar `b12' = `b11' > > tempname t1 t2 > scalar `t1' = `at'[1,8] > scalar `t2' = `at'[1,9] > > quietly { > tempvar toto > gen double `toto' = `a1' * `lnp1' + `a2' * `lnp2' + `a11' * `a12' > forvalues i = 1/2 { > forvalues j=1/2 { > replace `toto' = `toto' + `lnp`i'' * `lnp`j'' > } > } > replace `w1' = `a1' + `a11' * `lnp1' + `b1'*`lnp2' + `b11' * (`lnm' > - > `toto') > replace `w2' = `a2' + `a12' * `lnp2' + `b2'*`lnp2'^2 + `b12' * > (`lnm' > - > `toto') > replace `lnp2' = `t1' * `expfd' + `t2' * `p1' /*instrumentation > equation*/ > > } > > end > > use http://www.stata-press.com/data/r10/food,clear > > nlsur test @ w1 w2 lnp1 lnp2 lnexp expfd p1 ,parameters(a1 a2 a11 a12 b1 > b2 > b11 ) neq(3) ifgnls > > > > * * 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/

- Prev by Date:
**RE: st: Computation of Correlation Coefficients - COMPLEX** - Next by Date:
**Re: st: model for fractional data with panel data** - Previous by thread:
**Re: st: st : how to treat endogeneous variables with NLSUR** - Next by thread:
**st: regress with 4000 x's is very slow** - Index(es):

© Copyright 1996–2016 StataCorp LP | Terms of use | Privacy | Contact us | What's new | Site index |