EViews.com

Posted: **Tue Jun 07, 2011 12:37 am**

Hello,

I would like to estimate the following model using sspace object:

(1) Y = alpha + beta_{t}*X + eps_{t}
(2) beta_{t} = B_{t} + C_{t}
(3) B_{t} = B_{t-1} + w_{t}
(4) C_{t} = phi*C_{t-1} + v_{t}

where eps_{t}, w_{t} and v_{t} - random disturbances, beta_{t} - time-varying coefficient.

How should I represent properly the above model in Eviews 6.0? Since the eq. (2) can not be putted directly (state equation must be in the one-period lag of the states), my solution is:

Code: Select all

@signal Y = c(1) + (sv1+sv2)*X + [var = exp(c(2))]
 @state sv1 = sv1(-1) +   [var = exp(c(3))]
@state sv2 = c(4)*sv2(-1) +  [var = exp(c(5))]

Is it correct or there is more compact way to represent this? I'm asking because during the estimation I'm experiencing the singular covariance matrix warning.

Thank You in advance.

Posted: **Tue Jun 07, 2011 1:45 am**

The error message could be related to starting values. If you prefer a different specification, you can try the following:

Code: Select all

@signal Y = c(1) + sv1*X + [var = exp(c(2))]

@state sv1 = sv2(-1) + sv3(-1)
@state sv2 = sv2(-1) + [var = exp(c(3))]
@state sv3 = c(4)*sv3(-1) + [var = exp(c(5))]

param c(1) .0 c(2) .0 c(3) .0 c(4) .0 c(5) .0

EViews.com

State-space model specification

State-space model specification

Re: State-space model specification