State-space modelling

[creg]

Hi,

I want to estimate a multi-variate state-space model. My signal equation is as below (all observed variables):

@signal gdp = c(1)*gdp(-1) + c(2)*trade(-1) + [var = exp(c(3))]

Now I want to include an unobserved state variable which follows a random-walk process without a drift.

Please help! Many thanks.
Creg

[startz]

Hi,

I want to estimate a multi-variate state-space model. My signal equation is as below (all observed variables):

@signal gdp = c(1)*gdp(-1) + c(2)*trade(-1) + [var = exp(c(3))]

Now I want to include an unobserved state variable which follows a random-walk process without a drift.

Please help! Many thanks.
Creg

something like

Code: Select all

@state sv1 = sv1(-1) + [var = exp(c(4))]

[creg]

Hi,

Many thanks for the reply. I tried the following model:

@signal gdp = c(1)*gdp(-1) + c(2)*trade(-1) + [var = exp(c(3))]
@state sv1 = sv1(-1) + [var = exp(c(4))]

However, I get the following warning and the results are incomplete, as shown below.


WARNING: Singular covariance - coefficients are not unique

Coefficient Std. Error z-Statistic Prob.

C(1) 0.875527 NA NA NA
C(2) 0.018035 NA NA NA
C(3) -9.696770 NA NA NA
C(4) 0.173547 NA NA NA

Have you got any idea, please?

[startz]

The state variable needs to be linked into the data in the model somehow. Otherwise it's not identified.

[creg]

Hi,
Thanks again. Can I clarify this please; my state variable is meant to be the long run trend value of one regressor?

In that case, how to modify the following state equation, making a link to the regressor in the signal equation, pleasse?
@state sv1 = sv1(-1) + [var = exp(c(4))]

PS: I tried with the mean of the regressor "as a mean prior" but I get the same result.