IRF from VAR-BEKK

[FaeK]

Dear all,
I have written this code to estimate VAR-BEKK with impulse responses based on a previous post in the forum:
http://www.estima.com/forum/viewtopic.php?f=11&t=359

system(model=var1)
variables dlx dly
lags 1
det constant
end(system)
garch(p=1,q=1,model=var1,mv=bekk)
impulse(model=var1,step=20,result=impulses,factor=%identity(2))

However, I am not sure about the nature of the calculated impulse responses whether they are originated from the mean equation or the variance equation. Please, I need some clarification. One last thing, if there is anyway around to calculate their standard errors..

thanks in advance for your cooperation.
Faek

[TomDoan]

Those are the responses of the mean model. To get volatility responses, see http://www.estima.com/forum/viewtopic.php?f=8&t=1466. The confidence bands can be generated using Independence Chain Metropolis as shown in the OILGDPGARCHIRF program in http://www.estima.com/forum/viewtopic.php?f=8&t=1189

[FaeK]

Dear Tom, thanks very much for your feedback. just one last question, is it possible to get the generalized impulse responses of the mean model instead of the traditional ones generated above by the code or you think it is not straight forward to get them.

Many thanks
Faek

[TomDoan]

Dear Tom, thanks very much for your feedback. just one last question, is it possible to get the generalized impulse responses of the mean model instead of the traditional ones generated above by the code or you think it is not straight forward to get them.

Many thanks
Faek

It's not that they're not straightforward---it's a linear model in the mean, so the GIRF's are just linear combinations of the innovation responses. It's just that there is no fixed covariance matrix in a GARCH model to use for the contemporaneous correlations. You'll get different responses using the covariance matrices at different time periods.

[FaeK]

Many thanks for your fruitful feedback.