For difference stationary/ non-stationary variables do I use VECM or VAR estimation?
Dear Mike!
In time series analysis, I suggest you follow procedure:
1. You test the stationary of variable
2. If they all are stationary => you can use every models, but VAR and GARCH or ARCH are usually better OLS
3. If they are non - stationary => you test the cointegration of variables => there are 3 cases
- If they integrated + all variables are stationary at level 1 => use VECM
- If they integrated + variables are stationary at differences level => use ARDL
- If they do not integrated => make them stationary => use every models as the option 2
4. In the case of the monthly data or quarterly data => you have to check the seasonry problem before testing the stationary
I hope it will be helpful for you
Best
Canh
Your series are I(1). You can VECM if the variables are cointegrated. If not use VAR.
Mike,
Order of integration is only known through stationarity test. Besides there is no general order of integration for all the variables. Individual variables have their own order of integration. This is the first to do to know the statistical properties of your variables and the appropriate model to use.
So first thing first. Use the ADF or PP test or others in literature to determine the order of integration of each of the variables ( also known as test of stationarity). The result from here will determine whether to use VAR or VECM.
Use VECM only if ALL the variables are known to be integrated of order 1 and they are ALL stationary after differencing (that is they are cointegrated). The rest is considering VAR at level or VAR at difference or Toda and Yamamoto (1995) VAR of integrated series of different order.
- Badges
- Science topic
- Similar topics
- Correction