ML-estimation

[hanke1983]

Is it possible, in the logl-object, to estimate an regression with an embedded likelihood-function, as the equation below suggests:

eq.07.png (15.22 KiB) Viewed 5304 times

I have always thought so, since the idéa with maximum likelihood is to create a likelihood for the product of all residuals (sum likelihood), well at least that´s what I thought. But now Im not sure anymore, and I don´t know how to write it in code, its always some error. Is there some way to shorten it to several parts?

[EViews Gareth]

I don't think anyone can answer that. I'm not sure what an embedded likelihood is. Also, I don't see any likelihood looking thing in the equation.

[hanke1983]

ah, sorry if I didn´t explained the equation properly.

The phi_l squared is the likelihood estimate that people act rationally towards inflation expectations, phi^e. What I meant with an embedded likelihood function is that the parenthesis is just a part of the whole equation (but maybe I described it bad). At first I thought that I would estimate the parenthesis separately, but my supervisor told me otherwise. But now Im stuck with the problem on how to write this in eviews. I don´t know if there is any other way, other than the logl-object, and I would greatly appreciate any help.

[EViews Gareth]

The LogL object is the only chance you have. The main restriction with the LogL object is that you can only optimise something that is the sum of individual observations. Looking at your equation, I see no summation, so I'm not sure if that is the case.

[hanke1983]

Thanks for your answer, it helps a lot. At first, I thought exactly the same, and tried to code it in the logl-object. The thing is that phi_l is approximated by several different specifications, as seen in the picture below.

9_and_10.png (27.85 KiB) Viewed 5277 times

Its just that my supervisor told me that phi_l should be estimated separately, but on the other hand he does not know eviews.

But do you know if its possible to code equation (7) and substitute phi_l with eq. (9) or (10)? I´ve tried that but found it very difficult.

[EViews Gareth]

I'm reasonably confident that cannot be estimated in EViews yet.

[hanke1983]

ah.. ok. Thanks for your help anyway, you are the only one being clear and direct to the point. Thanks!