system of equations

[mboschloo]

I have a data serie “xt” and I am trying to estimate a system of equations on the back of that (a heterogeneous agent model). The model consists of 4 equations of which 3 are (non-observed) endogenous / based on lagged values of xt. The system reads:

xt – xt-1 = c(1) - (1 + mt-1) * c(2) * xt-1 + (1 - mt-1) * c(3) * (xt-1 - yt-1) + εt-1 * √(St-1)
yt = c(4) * yt-1 + (1 – c(4))*xt
St = c(5) + c(6) * St-1 + c(7) * [εt-1 * √(St-1) ] ^2
mt = tanh [(xt – xt-1) * (-c(8) * xt-1 – c(9) *(xt-1 – yt-1))]

so yt, St and mt are not observed, but are actually based on lagged values of xt, the error term from equation 1 and (in another model) based on some of the variables of the first equation. Unfortunately I cannot collapse the system into one “reduced form” equation that thereafter I could estimate. So I tried to enter the four equations in a “new object”- “system” window EViews indicates that “Y” has not been defined.

Does anyone know if it is possible to have the entire system solved (either iteratively or as one system by QML or so)? And if so, how should I then enter it into EViews?

Thanks, Meint

[EViews Gareth]

I think the best bet for you is to write your system in a LogL object. You really only have one error term, with a series of equations specifying the error term. That is exactly what the LogL object works with. The only down side is that you have to figure out the likelihood function for your error term.

[mboschloo]

Thanks Gareth. regards, Meint

[Leonel]

Hi:
I try to estimate a system of three equations for a panel data in eviews.
I Include dummies of individual (without constant term) and the program says "near singular matrix".
Of course, I exclude one dummy variable, and I verificate correlation with other variables.
How can I resolve this?
Thanks