VAR and degrees of freedom
Posted: Fri Aug 23, 2019 7:21 pm
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 ?
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 ?