VECM state space form specification

[shorty_85]

Hello everyone!!

I would like to estimate a VECM with time varying factor loadings. The VECM is the one in the attachment with the equations (1.1) and (1.2) as signal equations and equations (1.3) and (1.4) as the state equations. I am trying to estimate the time-varying factor loadings lambda, but i have some problems with the specification since i allways get realy stange results when I look at the state series that eviews creates. The error terms eta and epsilon are indipendent and have normal distributions with mean zero and constant variance. Optimal number of lags is 2 according to SC.

Here is my specification of the model:

@signal D(S) = SV1*( S(-1) - 1.02290145625*B(-1) - 14.9266426833 ) + C(2)*D(S(-1)) + C(3)*D(S(-2)) + C(4)*D(B(-1)) + C(5)*D(B(-2)) + C(6) + [var=1]
@signal D(B) = SV2*( S(-1) - 1.02290145625*B(-1) - 14.9266426833 ) + C(8)*D(S(-1)) + C(9)*D(S(-2)) + C(10)*D(B(-1)) + C(11)*D(B(-2)) + C(12) + [var=1]

@state SV1 = SV1(-1) + [var=1]
@state SV2 = SV2(-1) + [var=1]

@mprior v1
@vprior m1

Can anyone tell me if this is the right specification?? Since my result are really strange I doubt that this is correct.

Please let me know if anyone can give me some suggestions!! I need this for my thesis and it would be a great if someone could help me!!!

Thanks!!

[trubador]

Your syntax seems OK. Are you really sure about restricting the variances of disturbance terms? Other than that, the method you are using is very sensitive to data, so it is quite normal to experience some estimation problems. Here are few suggestions:

1) Try a more flexible variance structure like [var = exp(c(1))]
2) Try smaller lag lengths.
3) Try estimating the parameters of cointegration equation (i.e. alpha and beta) along with the model.
4) Try different starting values for the parameters of your model.
5) State initialization can be tricky. Unless you are confident about these values, let EViews handle them for you.

There are similar discussions in the forum, which you may find very useful...