State Space Estimation, Observable state variables

[cntantamis]

Hi,

Trying to work something similar to Diebold, Rudebusch and Aruoba (2006) paper using Sspace, I encountered the following issue and I am not sure of whether I am doing something wrong or it is just the way Eviews works.

The model tries to fit a term structure model using 5 factors: three of them, l1, s1, cu1, are unobserved (latent), but the other two, rgdp_gr, cpi_gr, are observed. The states are assumed to follow a VAR(1), and the yields are explained only by the latent factors.

When setting up the SSpace, I entered the interest rate yields as the signal equations and I set all the other factors are state equations:

@STATE L1=C(1)+C(2)*(L1(-1)-c(1))+C(3)*(S1(-1)-c(8))+C(4)*(CU1(-1)-c(15))+C(5)*(RGDP_GR(-1)-C(22))+C(6)*(CPI_GR(-1)-C(29))+[ENAME=HL,VAR=EXP(C(7))]
@STATE S1=C(8)+C(9)*(L1(-1)-c(1))+C(10)*(S1(-1)-c(8))+C(11)*(CU1(-1)-c(15))+C(12)*(RGDP_GR(-1)-C(22))+C(13)*(CPI_GR(-1)-C(29))+[ENAME=HS,VAR=EXP(C(14))]
@STATE CU1=C(15)+C(16)*(L1(-1)-c(1))+C(17)*(S1(-1)-c(8))+C(18)*(CU1(-1)-c(15))+C(19)*(RGDP_GR(-1)-C(22))+C(20)*(CPI_GR(-1)-C(29))+[ENAME=HC,VAR=EXP(C(21))]
@STATE RGDP_GR=C(22)+C(23)*(L1(-1)-c(1))+C(24)*(S1(-1)-c(8))+C(25)*(CU1(-1)-c(15))+C(26)*(RGDP_GR(-1)-C(22))+C(27)*(CPI_GR(-1)-C(29))+[ENAME=HRG,VAR=EXP(C(28))]
@STATE CPI_GR=C(29)+C(30)*(L1(-1)-c(1))+C(31)*(S1(-1)-c(8))+C(32)*(CU1(-1)-c(15))+C(33)*(RGDP_GR(-1)-C(22))+C(34)*(CPI_GR(-1)-C(29))+[ENAME=HCP,VAR=EXP(C(35))]

@SIGNAL US3M=L1+S1*((1-EXP(-1*C(46)))/(C(46)*1))+CU1*(((1-EXP(-1*C(46)))/(C(46)*1))-EXP(-C(46)*1))+[VAR=EXP(C(47))]
@SIGNAL US6M=L1+S1*((1-EXP(-2*C(46)))/(C(46)*2))+CU1*(((1-EXP(-2*C(46)))/(C(46)*2))-EXP(-C(46)*2))+[VAR=EXP(C(48))]
@SIGNAL US1Y=L1+S1*((1-EXP(-4*C(46)))/(C(46)*4))+CU1*(((1-EXP(-4*C(46)))/(C(46)*4))-EXP(-C(46)*4))+[VAR=EXP(C(49))]
@SIGNAL US2Y=L1+S1*((1-EXP(-8*C(46)))/(C(46)*8))+CU1*(((1-EXP(-8*C(46)))/(C(46)*8))-EXP(-C(46)*8))+[VAR=EXP(C(50))]
@SIGNAL US3Y=L1+S1*((1-EXP(-12*C(46)))/(C(46)*12))+CU1*(((1-EXP(-12*C(46)))/(C(46)*12))-EXP(-C(46)*12))+[VAR=EXP(C(51))]
@SIGNAL US5Y=L1+S1*((1-EXP(-20*C(46)))/(C(46)*20))+CU1*(((1-EXP(-20*C(46)))/(C(46)*20))-EXP(-C(46)*20))+[VAR=EXP(C(52))]
@SIGNAL US7Y=L1+S1*((1-EXP(-28*C(46)))/(C(46)*28))+CU1*(((1-EXP(-28*C(46)))/(C(46)*28))-EXP(-C(46)*28))+[VAR=EXP(C(53))]
@SIGNAL US10Y=L1+S1*((1-EXP(-40*C(46)))/(C(46)*40))+CU1*(((1-EXP(-40*C(46)))/(C(46)*40))-EXP(-C(46)*40))+[VAR=EXP(C(54))]
@SIGNAL US15Y=L1+S1*((1-EXP(-60*C(46)))/(C(46)*60))+CU1*(((1-EXP(-60*C(46)))/(C(46)*60))-EXP(-C(46)*60))+[VAR=EXP(C(55))]
@SIGNAL US20Y=L1+S1*((1-EXP(-80*C(46)))/(C(46)*80))+CU1*(((1-EXP(-80*C(46)))/(C(46)*80))-EXP(-C(46)*80))+[VAR=EXP(C(56))]
@SIGNAL US30Y=L1+S1*((1-EXP(-120*C(46)))/(C(46)*120))+CU1*(((1-EXP(-120*C(46)))/(C(46)*120))-EXP(-C(46)*120))+[VAR=EXP(C(57))]

My problem is that when I estimate the State Space model, it seems that the two observed variables (rgdp_gr, cpi_gr) are treated as if they are latent; when estimation is complete, what you get for these two state variables have nothing to do with the realized values.

Thus, I have two questions.
1) Does EViews allow for State variables to be observed, of it treats all such variables as latent?
2) Conditional that Eviews allows states variables to be observed, any thoughts of how to set this up? I know that setting them as @signal won't work.

Many thanks in advance for any help/suggestions.

PS EViews 8 (64-bit enterprise edition), Aug 2013 build

[startz]

I don't think that state variables are ever allowed to be observed in a state space model. That's largely what distinguishes state from signal variables.