04-09-2013 05:53 PM
I have a question about the technique details of SUR employed in proc syslin or proc model.
We know that SUR will use the cross-equation variance-covariance matrix to do the estimation. However, if I have more equations that number of observations (in each equations). For example, I have 30 equations and 20 observations in each equation. The variance covariance matrix would be 30 by 30. However, I only have the matrix of residuals as 30*20. In this case, would the variance covariance matrix used in SUR be deficient? How does SAS deal with this problem? I expect there should be at least some warning from SAS, but it is not.
04-11-2013 04:49 PM
I spoke with the developer of proc model and he had this response, "
For the SUR method MODEL computes a pseudo inverse of the OLS
residuals’ covariance matrix. When the OLS covariance matrix is
singular a warning is produced explaining which equations’ rows in the
covariance matrix are linearly dependent, and PROC MODEL continues with
the SUR estimation of parameters. If the user isn’t getting this warning
message for a known deficient matrix then something else is going on "
He also requested that he would be happy to look at your code and data if you wouldn't mind providing and example of the issue. - Ken