SSPACE: lag structure, parameter values

[skiab]

Dear all,

we use an Eviews SSPace object to do econometric forecasting. The idea is to mix monthly and (then partly unobserved) quarterly data.
The model consists of about 10 equations which are separately specified using OLS.
Two problems occur when plugging them into the SSPACE object:

1) deterministics and exogenous variables are automatically lagged one period in the estimations, e.g. a dummy variable shifts into the wrong period. Therefore, if we want lag 1, we may have to create contemporaneous relations which is not allowed in stace space representation - what can we do? Moreoever, we are not sure whether the same happens with interactions of the dummies and other variables which would create nonsense regressors.

2) Parameter estimates change substantially. Whereas a univariate AR(1) is equal in OLS and SSPACE Kalman Filter ML, it starts to change values as soon as we put in more variables. Changes increase when inserting a second equation. So, model specification in the sspace object would lead to completely different conclusions and hardly any significance. To allow or not allow for residual correlation does not mitigate the problem.

We are grateful for any hints. Thank you! Kind regards, Sabine