I suggest you to do stationary test first. If stationary, you can try LCF/PCF to see the what is the optimal lag to use in autoregression.
In VAR, you can put the optimal leg in equation, say, y(t)=y(t-1)+y(t-2)...
you have to look at residual and check the co-variance matrix of your residuals. If the co-variance matrix is not independent, you have to decompose it and use GLM. Indeed, you DCF test is for the stationary of the time series. you don't need to redo it if you change the VAR.
Hope this helps.
|