Discrepancy between a VAR and a VECM

[Reynaldo2007]

Hello,

I am performing cointegration analysis using Eviews and the DOLS estimator in the cointegrating equation. After saving the residuals of this equation (they are stationary then the variables are cointegrated), I estimated the short run equation including the lagged error term, one lag of the dependent and independent variables taken in differences (my sample is not large), and a constant. I got the coefficients of the VECM. However, I was thinking that in this case that I have included only one lag, if I estimate an unrestricted VAR including the dependent variable and the independent variables and the estimated error term, the coefficients of the VAR should be EXACTLY the same of the VECM that I estimated. The only difference is that when estimating the VAR, I will get also the equations considering the independent variable and the error term as dependent variable. The result was that the coefficients were extremely similar, but NOT EXACTLY the same. Why do this happen?

[dakila]

Check your sample size.

[Reynaldo2007]

Very much thank you for your reply. I have 61 observations that when I apply DOLS the sample use to be 56 or 57

[dakila]

Sorry, you mean that the samples were different?

[Reynaldo2007]

sorry, what I mean is that because i am using DOLS the sample reduces to 57 and I save 57 observations of the error term. Then when I estimate the VECM it is estimated with 57 observations because is limited by the error terms observations. When I build the VECM with a lag in all the variables (including the error term) and compare with a Var with 1 lag, there arise the discrepancy that is very small, but there is

[dakila]

Ok. Could you post the results?

[Reynaldo2007]

Dear dakila, lot of thanks for your attention and help. I already found the source of the problems. It was very very simple. The point is that the VECM allows for one year more in the data (2011). When I estimate the VECM with data until 2010 the results are exactly the same. My only problem was that the VAR cannot reach the 2011 observation because it is including also the residuals as variable determined by the system (exogenous) and the VECM (when estimaded by DOLS) not (the lagged residuals is like an independent variable). This is the reason, right? Sorry for taking your time.