Two stage least squares

[kungal]

Does any one know why it seems to be impossible to get the coefficients that EViews estimates in prab by doing prab manually.

I understand that the standard errors should be different but I can't see how EViews is coming up with its parameter estimates either.

[EViews Gareth]

It is relatively straight forward to mimic the results manually.

Of course it is pretty difficult to tell why you are unable to do it, without knowing what you are doing.

[kungal]

I suppose that would help.
I cannot see what I am doing wrong. I have a set of instruments a, b and c and I regress each of the explanatory variables on a, b and c and click on prab button to create series with the predicted values of each of my explanatory variables. Then I regress my dependent variable on the prab series I have just created. However, the results I obtain are completely different from the results I get if I use the prab option for estimation.
I feel like I must be overlooking something rather important.

[EViews Gareth]

Code: Select all

create u 100 series x=nrnd series y=nrnd series z=nrnd 'simple 2SLS equation eq2sls.tsls y x @ z equation eq1.ls x z c eq1.fit xf equation eq2.ls y xf

[EViews Gareth]

Get's a bit more (and only a bit) if you have one of the regressors as an instrument (which is usually the case, since C is usually both an instrument and a regressor):

Code: Select all

create u 100 series y=nrnd series x=nrnd series z=nrnd equation partialy.ls y c z partialy.fit yf equation partialx.ls x c z partialx.fit xf equation tsls.tsls y c x @ z c equation manual.ls yf c xf

[kungal]

Thanks a lot. I got it now