# Lag length selection in levels-VAR before VECM: inclusion of exogenous variables

I am trying to estimate a VECM and I read in Asteriou´s book "Applied Econometrics" that

"The most common procedure in choosing the optimal lag length is to estímate a VAR model including all our variables in levels (non-differenced data). This VAR model should be estimated for a large number of lags, then reducing down by reestimating the model for one lag less until we reach zero lags."

However, I will introduce a dummy variable in my model.

Question: should I include the dummy in this VAR or should I include the dummy only afterwards when I estimate the VECM?

• What do you think about the answer? – Richard Hardy Sep 11 '17 at 12:56

I would include the dummy variable(s) right away. For example, if the data generating process is well described by a VAR(1) and (strong) additive seasonality, then a sensible lag selection procedure will correctly suggest lag 1 if you include the seasonal dummies in the models among which you are choosing. However, if you do not include the dummies at the model selection stage, a VAR($p$) model could be suggested by the procedure instead, where $p$ could be equal to the seasonal period (e.g. 4 for quarterly data), as the (strong) additive seasonality will induce (high) autocorrelation at lag 4.