Covariance Restrictions

[caburdick]

I have 4 dependant variables w,x,y, and z with z = w+x+y. I want to model these four variables as a system with a common set of explanatory regressors. The coefficient restrictions implied by z=w+x+y are straight forward to impose, but how can I impose the covarince restrictions implied by ez = ew+ex+ey? I've tried using named errors with sspace, but the Kalman filter is really not the right tool for this job and only works if I add additional observation noise to avoid singularity problems.

Is there a way in Eviews 6 to use GLS with a user supplied covarince matrix? If so, I can use an EM algorithm to estimate the system. Or is my best bet to model this as a structural svar? Any tips or suggestions will be appreicated.