panel-var

[seda]

Hi,

I'm trying to modelise a var model on panel data sample. i would like to know how i can make it.
my equation model is like this:
u i,t =a i,t + b*u i,t-j +c*g i,t-k+ d*Dummy i,t-l + E i,t


where "t" is the time
j, k, l are the optimal lags for the variables.
i work with panel data for OECD countries (27) for "i"
1980-2007 for "T"
1 endogeneous: u
2 exo: g, dummy.

to select the optimums lags, to verificate the integration degree of the series, to write the model, to identificate the parameters of the var models, to make the impluse response and the decomposition of the variance, do i must use the same "procedure" as a "normal var"?
and for the series's test, which test must i do in this conditions?
how can i do?

thank you very much if somebody can help me!

seda

[TomDoan]

Panel VAR's are mainly panel data techniques, more than VAR techniques (which is why I moved the topic). If everything about the model is homogeneous (same coefficients, same variance), you can basically treat it like a standard VAR. When you estimate the model, RATS will respect the boundaries between individuals. However, it's almost never interesting to assume that the model is homogeneous, so the question becomes: what's the same, what's different? What techniques apply depend upon your decision. If the intercepts are different (which you're assuming), do you have enough data to ignore the "LSDV" small sample bias? At your sample size, probably not. By the way, is there actually a "V" in the VAR? You seem to indicate that you have only one endogenous variable, in which case what you have is a dynamic panel model.

[TL]

Hi

Panel VAR's are mainly panel data techniques, more than VAR techniques (which is why I moved the topic). If everything about the model is homogeneous (same coefficients, same variance), you can basically treat it like a standard VAR.

If I allow for different intercepts across countries (cross-section units) and other coefficients in VAR model are the same, can I still estimate the model like a standard VAR? Just include country dummy variables as exogenous variables. For imposing sign restriction, Can I use the same program as time series technique in this case? Thank you.

Tim

[TomDoan]

Hi

Panel VAR's are mainly panel data techniques, more than VAR techniques (which is why I moved the topic). If everything about the model is homogeneous (same coefficients, same variance), you can basically treat it like a standard VAR.

If I allow for different intercepts across countries (cross-section units) and other coefficients in VAR model are the same, can I still estimate the model like a standard VAR? Just include country dummy variables as exogenous variables. For imposing sign restriction, Can I use the same program as time series technique in this case? Thank you.

Tim

See Holtz-Eakin-Newey-Rosen for a discussion of the issues regarding panel VAR's when the T dimension is small. If T is large, the small-sample bias from individual effects (or, equivalently LSDV's) is small, and is not worth the the loss of efficiency in using instrumental variables techniques.

If you estimate the model in LSDV form, then, yes, the same basic techniques would apply, since that would be sampling the fixed effects along with the VAR coefficients. If you difference out the data first, it would be only approximately correct, since that isn't treating the fixed effects as separate parameters.