horsehandsome wrote:Hi, I'm trying to estimate a state-space model in which the state vector of the transition equation is a mix of state and observables. The reason for us to have this setting-up is to allow for the interaction between the state variables and the relevant observables. Is there a way for the RATS program to estimate this kind of model? I will greatly appreciate any input! Thanks.
I'd have to see more of what you want, but that sounds like a standard state-space model, where one of your observation equations will take the form y=unit vector * X with a zero error (zero component in the SV matrix). If you look at the procedure
armadlm.src you'll see the following structure for the state vector:
* The state space representation is from Jones(1980), "Maximum
* Likelihood Fitting of ARMA Models," Technometrics, pp 389-395. The
* state space representation of x(t) uses a state vector consisting of
* x(t),x(t+1|t), x(t+2|t),...,x(t+r-1|t), where x(s|t) is the best
* linear predictor of x(s) given information through t.
Note that the actual x is part of the state, along with unobservable forecast series.