# st: Help needed for three-step FGLS estimation

 From "Jakobsen, Kristian Thor" To statalist@hsphsun2.harvard.edu Subject st: Help needed for three-step FGLS estimation Date Mon, 22 Jun 2009 14:38:05 +0200

```I am trying to estimate the vulnerability of Nicaraguan households using the three-step FGLS method that has been suggested by Tesliuc and Licndert (2002) and Chauduri et al (2002). But I am having trouble at step three, so I am hoping that somebody can help me out here:

We assume that ln c=x\$B&B(J+ e, (1)

where c is consumption and x is household characteristics.
We also assume that the variance of e is given by e^2=x\$B&H(J + n (2)

I therefore first estimate (1) using OLS. Then I save the squared residuals and regress these with the same household characteristics in order to estimate theta from (2).

But now I am lost. I need to transform (1) into:

ln c/\$B"e(J(x\$B&H(J)= x\$B&B(J/\$B"e(J(x\$B&H(J)+ e/\$B"e(J(x\$B&H(J)

But how do I do that? So far I have:

.reg logconsum miembros_01 s2p3 s2p4 genage agesq primary secondary /// .university badhouse durindex landsize landsizesq healthaccess ///
.schoolaccess childratio oldratio drought pest district_012-district_0117 .[pw=peso2], robust
.predict residuals,resid
.gen e2=residuals^2
.reg e2 miembros_01 s2p3 s2p4 genage agesq primary secondary university /// .badhouse durindex landsize landsizesq healthaccess schoolaccess
.///childratio oldratio drought pest district_012-district_0117 [pw=peso2], .robust
.predict yhat

And I need to be able to estimate both the expected log consumption (E(ln c|x)=X\$B&B(J(hat)) and the variance (V(ln c|x)=X\$B&H(J(hat)), so that I can estimate:

v=Pr(ln C<ln z|X)= \$B&5(J((ln z - x\$B&B(J(hat))/ \$B"e(J(x\$B&H(J))

Can anyone help on how to do the final step in STATA?

Cheers
Kristian

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```