Modelling VAR or VECM

Asked Feb 27 '21 at 16:44

Active Feb 27 '21 at 17:26

Viewed 35 times

I have 5 series, I did ADF on my series and found that 3 series have unit roots(x1, x2, and x3) and 2 don't(x4 and x5). Still using the level variables I did Engle e Granger test and Johansen test( Trace test indicates 2 cointegrating) and found that there is cointegration between the series.
how should I proceed with this? Can I mix differentiate with non differentiate variables? Should I do a VAR for the first differences in the series or a VECM?

edited Feb 27 '21 at 17:26

asked Feb 27 '21 at 16:44

RoutoCharles

Cross posted at Economics SE. – Richard Hardy Feb 27 '21 at 18:34
See these threads. Your question is likely a duplicate of one of these. – Richard Hardy Feb 27 '21 at 18:35
Does this answer your question? VAR or VECM for a mix of stationary and nonstationary variables – Richard Hardy Feb 27 '21 at 18:37
HI Richard, i saw your answer and i have a question: I changed my variables and now I have 3 OF THEN ARE STATIONARY (x1, x2 and x3) and 2 are NOT.
-Test each pair of the nonstationary series (x1 and x2; x1 and x3; x2 and x3) for cointegration using the Johansen or the Engle-Granger test . i did this for the stationary the series cointegrated using Engle-Granger and then using Johansen Trace test indicated 3 cointegrating.
– RoutoCharles Feb 27 '21 at 19:04
@RichardHardy and For the non-stationarity series they not cointegrated in level, but cointegrated in first differences.
What should I do? I didn't understand how to do "If (C) then build a model where..." this
– RoutoCharles Feb 27 '21 at 19:13
Series that are I(0) cannot be cointegrated. If a bunch of series are I(1) and not cointegrated, they cannot be cointegrated in first differences for that reason; check the definition of cointegration which requires the constituents to be integrated, never I(0). – Richard Hardy Feb 27 '21 at 21:19

0 Answers0