Correlation between error term of endogenous variable in panel data?

[Jose Ferreras]

Is it possible to correlate endogenous variables in STATA SEM when analyzing panel data? In my model, only one variable is included as dependent and the data is collected from 2006 to 2014. I read in differents forums that the error terms associated to the endogenous variables can be correlated, however I could not find information if this approach also apply when analyzing panel data. Can anyone offer me an explanation about this issue in case that I'm missing something important here.

Thank you very much in advanced.

[Phil Bromiley]

You can allow a wide variety of error covariances in SEM. There are a couple of problems. Your estimates depend on the imposed assumptions (like any other analysis) and the implications in SEM may be a little less obvious than in regression or similar tools. You can create a model that is not identified. Normally, if the model is not identified, Stata will not give you interpretable results (it won't converge or will converge but with no standard errors, etc.).