The RATS Software Forum

Dear Tom:

If my state space equation model does not converge, but the t-value of the estimated state is significant and the result is reasonable and meets expectations, can the model estimation be considered successful?

Best Regard
Hardmann

I'm not sure what you mean by the "t-value of the estimated state".

Quite a few state-space models have border solutions which are "valid" but indicate that you may have more fundamental shocks than necessary. Those can manifest in zero variances or + or -1 correlations. If you look at a number of the state-space model examples, the final estimates require pegging a parameter or two, which is done because the completely free model doesn't converge.