TSLS (2SLS): computation of Std. Error

[mectricsdonk]

Eviews version 6

Dear Forum,

I'm currently verifying my 2sls estimator of Matlab with Eviews. The coeficients are the same, but the standard error is not. Hence my question, does one of you know how the std. error of the two stage least squares in Eviews is computed?

My model has 1 endogenous regressor and 12 instruments (including a constant).

The user guide does not give an explicit expression. When I apply basic theory I get different results.

Basic theory states (under homoskedasticity):

sigma^2 = s^2*(X'*Pz*X)^(-1)

where s^2 = (1/n)*uhat'*uhat

where uhat = y - X*betaiv

and X is the endogenous variable.

Hope you guys know it!

Thanks in advance

Mectricsdonk

[startz]

The formula is given in equation (21.3) of the help system. EViews estimates s^2 by dividing by n-k rather than n.

[mectricsdonk]

Hi Starz,

Thank you for the quick response. Unfortunately Chapter 21 of the user guide (of eviews 6) is about "Working with Spools", it does not contain formula (21.3).

I can't access the quick help reference on the web for some reason.

Is the computation of the standard error the same as mine except divided by n-k rather than n?

[startz]

My reference was from EViews 8. Yes, I believe the difference is n versus n-k. Since these are asymptotic approximations, it shouldn't make too much difference which you use.

[mectricsdonk]

I didn't solve the problem yet, but this gives me reassurance of what is wrong and what not. Your help is much appreciated! :D