Question regarding basic VAR Specification

[tonidastier]

Hello everybody,

I got a short and hopefully easy to answer question:

For a valid VAR one needs stationary time series?
So I test for stationarity via a unit root test.
The unit root test gives strong evidence, that the time series are stationary in the first difference.
Now do I have to difference all the series one time to make it stationary and then set up the VAR with the differenced series or do I have to set up the VAR with the original series?

Thanks a lot for your help!

[trubador]

The question might be simple, but answer is not. In practice, you can see them both. Differencing in VAR context is not similar to that of ARIMA modeling, where stationarity is crucial in terms of convergence behavior and the validity of results. When you difference the data, you actually lose the long term information, which may be an important component for the purpose of your study. If it is not, then whether or not differencing the data should not be much of a concern. I, for instance, prefer not to work on differenced series when it comes to VAR modeling.