Page 1 of 1

Panel Var

Posted: Wed Jul 16, 2014 3:21 am
by arpaial
HAllo!

How can I Interpret with eviews 8 the IRF provided with estimation of a var in panel format (stack vector of 30 countries)? How does it work with Eviews the identification of structural VAR estimated when variables are in a panel format?

thank you
Alfonso

Re: Panel Var

Posted: Wed Jul 16, 2014 8:45 am
by EViews Gareth
EViews does not use any panel estimation techniques for VARs. It simply estimates a standard VAR on the stacked data, with the exception that lags behave properly (they do not cross-across cross-sections).

Re: Panel Var

Posted: Wed Jul 16, 2014 1:37 pm
by arpaial
EViews does not use any panel estimation techniques for VARs. It simply estimates a standard VAR on the stacked data, with the exception that lags behave properly (they do not cross-across cross-sections).
Thank you for your reply. If I understand well Eviews ignores the panel structure doing a pooled estimation. In terms of algebra the parameters are computed as

beta= (X'X)^(-1)X'Y

where X is an nTxK matrix and Y is nTx1

So fixed effects are ignored and estimates of beta can be biased.

However, one could demean the data taking the deviation from means for different units. This would be (almost) equivalent to make a panel VAR with fixed effects. I say almost because one would ignore the possibility of cross-section heteroschedasticity. But one could deal with this issue expressing the demeaned data in terms of common factors.

So working with demeaned variables (ie x(i,t)-x(i) where x(i) is the individual specific mean ) purged from common factors would allow to interpret the IRFs "correrctly" as in a proper panel VAR which controls for individual specific heterogeneity possible cross untis correlation.

Is this interpretation correct? If not how one can I interpet the IRFs

It would be nice to see the algebra of what Eviews does. Ihave not found an explanation in the manuals; but please correct me if I'm wrong

Thank you again for your help on this important topic

Alfonso

Re: Panel Var

Posted: Wed Jul 16, 2014 1:45 pm
by EViews Gareth
Your formula for the calculation of beta that EViews does is correct, although the lags complicate the structure of the Y and X matrices, and T.

The econometrics behind computation of panel VARs when you believe there are fixed effects are complicated and still under study. If you imagine the complexity of the estimators for univariate dynamic models (DPD models), VARs are a step beyond.