VAR and degrees of freedom

[idomen]

Dear forum participants


For deciding the specification of VAR, there is a trade-off between omitted variable bias and autocorrelation and I want to fit a model with 5 variables and 4 lags with quarterly data

You suggest the rule of thumb np<T/3 that give us model with less than 5 variables

I have ended up with 3 specifications :

1st follow np<T/3 rule of thumb when n=4 df=-4

2nd follow VARX specification with one exogenous variable, at n=4 df= 60-48=12

3rd follow a VARX specification with 2 endogenous and one exogenous variable at n=4 and df=36

Which of the 3 specifications do you think that will give more robust results ?