Page 1 of 1

How to estimate a fully modified VAR for impulse response analysis?

Posted: Thu Jun 14, 2018 11:58 pm
by Nico Lai
Hi community,

I am interested in impulse response analysis, variance decomposition and granger causality in EViews 10. However, my variables exhibit cointegration of order 2 as well as integration of the form I(2), I(1), I(0). Thus, the ususal reduced form VAR is inappropriate for inference. I would like to avoid VECMs as their impulse responses are often persistent rather than dying out in the long run. Therefore, I tried a lag-augmented VAR as suggested by Toda and Yamamoto (1995) but found that I do not have enough observations for the extra lags required. My sample covers 18 years of monthly data for 6 different variables.

I came across Peter Phillips‘ so-called fully modified (fm) VAR approach [1995], which extends FMOLS and was shown to be applicable in my case. I was, however, unable to find any practical guide on how to implement this approach in EViews 10 - or any other program for that matter.

Does someone know how to implement FM-VARs (or any other suitable alternative) in EViews?

Any guidance will be much appreciated.