I am using Dynare to estimate a model which has the property that all the unobservable variables are linked to the observable variables through identities. My question is: Can I get Dynare to initialise the Kalman filter using actual observations?

I am using Dynare to estimate a model which has the property that all the unobservable variables are linked to the observable variables through identities. My question is: Can I get Dynare to initialise the Kalman filter using actual observations?

Thanks,

Ida[/quote]

Hi Ida,

I may be wrong, because I don’t know your model, but if your unobserved variables are deterministic combinations of your observed variables you don’t really have any unobserved variables in your model. So you don’t need to run a kalman filter to evaluate the likelihood of your model. But I wonder how you could evaluate the likelihood. Deterministic restrictions between observed variables sounds to me like a singularity problem. I guess I am misunderstanding something…

As I understand it, in the current version Dynare will use steady state values in the stationary case or a diffuse prior in the case non-stationary filter. For the stationary case, I would like to be able to manually set different initial values for the different state variables.