EViews.com

Hi all,

I am using e-views version 7.

I am trying to estimate a system of linear equations in which the constants are restricted to be equal to the variances of the error terms. I would like to use full information maximum likelihood to this end. So, e.g., the first of my equations reads as follows:

RET1 = C(1) + C(2)*MYP + C(3)*UI + C(4)*DSV,

where C(1), C(2), C(3) and C(4) are the unknown parameters, and RET1, MYP, UI and DSV are my variables. Now, a full information maximum likelihood estimation should give me estimates not only of C(1), C(2), C(3), C(4), etc. (there's more than one equation, remember), but also of the variance-covariance matrix elements of the error terms. I want to restrict C(1) to be equal to the variance of the first error term, and I want C(5) (the constant of the next equation) to be equal to the variance of the second error term, etc..

Is this possible? I'm a new user. If this question does not make much sense, please let me apologize for it.

Thanks in advance,
Kevin

Unfortunately, I can't think of any way to get FIML to do the kinds of variance restrictions that you are asking for.

Okay, thanks!

Can I ask one more question, please? Is there a way in e-views that I can see (obtain) the estimated elements of the variance-covariance matrix of the error terms in a full information maximum likelihood estimation? If there were, then it may be possible to first run the FIML estimation without variance restrictions, and then to use the variance estimates of the error terms in a second-stage estimation as constants (i.e., I would change the first equation from:

RET1 = C(1) + C(2)*MYP + etc. to RET1 = (estimated value variance error term 1) + C(2)*MYP + etc.)

Iterating on this approach may lead to convergence of the constants and the variances of the error terms.

I hope that I have been able to explain this in a way that it makes some sense.

Many thanks for your very prompt help!

Best wishes,
Kevin

You can use the system object to estimate a FIML equation.