SSpace structure

[pcruz]

Hello, I succesfully estimated an SSpace model of the following form:

Y = H*X + w
X = F*X(-1) + e

The model has 7 signal equations and 30 state variables (including the auxiliary ones like X1=X(-1))

Now I am trying to compute the following matrix (M):

M = (I-(I-K*H)*F)^(-1)*K

where (matrix dimensions in [ ])
I = @identity(s)
K = Kalman Gain [s,y]
H = Signal equation parameter matrix [y,s]
F = State equation parameter matrix [s,s]

s = # of state variables in the system
y = # of signal equations in the system

In order to do so, I need to understand well how Eviews structure the F,H and K matrices so here are my questions (Im calling "model" the Sspace obejct):

1) Are F and H in a structure like the one displayed in the matrix using model.structure? In this case, this matrix would correspond to F stacked above H.
2) How can I put the view model.structure into a matrix that I can manipulate?
3) The kalman gain (model.@curr_gain) is displayed in a @vec for each period. I need to @unvec it in an [s,s] matrix. If I do so, would it be consistent with the structure of F if (1) is true?

Please help!
(Im using Eviews 9)

[EViews Glenn]

1. yes, but with the state equation offset by one period (see the manual for discussion)
2. you'll have to create the matrix yourself -- you should be able to copy-and-paste
3. yes

[pcruz]

Thank you for your help.
So the only way to work with the structure matrix is by copying it "by hand" using copy-paste?

[trubador]

Actually, I would expect the following command to do the trick, but it does not:

Code: Select all

ttom(model.structure,smat)

Anyway, you can always create a blank matrix (e.g. smat) and fill in the values via copy-and-paste.

[EViews Glenn]

You can use ttom but if you use the coefficient view you can run into trouble. If you use the value view, it should work.

Code: Select all

ttom(model.structure(v),smat)

[pcruz]

Excelent, thank you!