A VECM model includes a set of "cointegrated long-run relationships" which, although they are NOT structural equations, they CAN be expressed (reparametrized) as structural equations (usually depending upon the appropriateness of the long-run adjustment error-correction coefficients, say, the dynamic stability of the VECM model). You can argue about/explain the specific parameterization in terms of a theoretical structure, even though you did select it based on the most appropriate empirical representation of the VECM. I would advise to use a VECM (estimated by Maximum Likelihood) model instead of a simultaneous equation model (estimated by 3SLS) because the former is robust against the empirical possibility of including spurious long-run relationships.

Dr. Bento´s answer is quite relevant because the first contrast in the literature was between "old" simultaneous equation models (SEMs) and the VARs. Christopher Sims made VARs win over SEMs thanks to his argument against the "incredible theoretical restrictions" used in SEMs (in favor of the non-orthogonal structural restrictions in "structural VARs"). But when the "new" statistical discussion about testing the stationarity or the non.stationarity of any particular time series took place, it turned out that if all the individual time series included in a VAR model are individually (made) stationary, then the whole VAR model is stable. What if at least one variable is not stationary? Then the VAR model will tend to be unstable and thus the econometrician should obtain its first-differenced version, which is stationary, and include only this version into the VAR model. The last paragraph of Dr. Bento's answer is related to the important issue about " long-run information" If there are at least two variables which are non-stationary each BUT they are intimately related by a long- run relationship, then the errors of such cointegrated relationchips should be added (with a lag) to the original VAR model as exogenous variables, and thus we obtain a VEC model.

Dear Saadia.

VECM is a long-run adjustment error correction mode which can be used as an alternative to simultaneous equation model.

Regards

There is no direct answer to your question. If you are sure of your theory and have sufficient data to identify a structural equation model then such a model may be better. Your question does not indicate how you would intend to identify such a model. A VAR type model can always be regarded as the reduced form of a structural model. If you intend to use a VAR type model you must provide identification if you wish to do impulse response etc. analysis. Regardless you must use your economic theory. I suspect that there may be omitted variables in your model. I have listed below some points that you should consider for your analysis.