Hi Nadra, I am sorry that I am not an expert in SAS. Just from a theoretical point of view, every time series model can be cast into its state space representation. Could it be the case that SAS statespace provides you with time variant coefficients of the model? In any case, regarding your question, I would go for a bivariate VAR model (two equations, one per variable) and then test via Engle-Granger tests whether variable 1 can influence / predict variable 2 and vice-versa. If your data is not stationary but the series are cointegrated (test for it) you can use the VAR model still (or use a VECM model), if they are non-stationary and not-cointegrated you shoud take first differences to make the series stationary before estimating the model.

Hope that helps,

Martin

Nadra

State Space is one form of writing ARIMA models (an analogy can be made with scalar value and matrix representation of linear equations).

But if you are interested in checking if X causes Y or Y causes X or both, you have feedback, you are dealing with multivariate time series . You can use Multivariate Arima models or its state space representation

Basilio

And what correlation do you get between x and y variables? =).

I would agree with Martin's suggestion. I think the issue is not so much what procedure in SAS (or in any other software) to use but how to model the time series.

I also agree with Marti. You have to perform a causality test and for doing so, you need to adjust a VAR or VEC modelo depending on the nature of the series. Recently there has been some discussion about those models in this forum. It may help you.