I am beginning to study VAR models, and I have a simple question.

  • If I have three variables and one of them is nonstationary, can I estimate a standard VAR model? If not, what should I do?
  • 1
    $\begingroup$ No you cannot. If only one series is unit root non-stationary you should remove the unit root by first (or second) differencing the non-stationary series. Then continue the analysis using the the differenced series. $\endgroup$ – Plissken Jul 20 '16 at 17:11
  • $\begingroup$ thanks. I Got it. In this case I would run the VAR and interprets the beta parameter as a function of different variable. The interpretation for the impulse response function would take place in the same way? Right? $\endgroup$ – Linkman Jul 20 '16 at 17:14
  • 1
    $\begingroup$ Yes. You would continue the analysis as you would if all three would have been stationary in levels. See the notes by Eric Zivot: eco.uc3m.es/~jgonzalo/teaching/timeseriesma/…. Page 11-12 discuss exactly what you are asking. $\endgroup$ – Plissken Jul 20 '16 at 18:12
  • 1
    $\begingroup$ Sorry. It was actually p. 15 - combination of I(0) and I(1) variables. The link is here: eco.uc3m.es/~jgonzalo/teaching/timeseriesma/… $\endgroup$ – Plissken Jul 20 '16 at 19:19