Skip to content

About AR(1)-GJR-GARCH(1,1) MODEL

3 messages · XI ZHANG, Alexios Ghalanos, Jeff Ryan

#
sorry to bother all,
I am recently doing a topic about spillover effect between two markets. And
I want to use AR(1)-GJR-GARCH(1,1)-M Model. I find that rgarch package has *
functions* for univariate GARCH model, including GJR.
My GJR model is in the attachment.



where ?_(j,t-1)^2 is the squared volitility of counterpart market,
?_(i,t-1)^2  is the squared volitility of domestic market. S represent good
news (S=0)or bad news(S=1). and i write the R code as followings,
external.regressors=lagvoljp)
distribution.model="norm")
But then I figure out that in normal GJR model, the fomula is
and the function written in rgarch package is also based on it
which means it catches news from domestic market.

But in my model, it should catch news from counterpart market,
I was thinking of taking S part as an another external regressors
since if ?<0,S=1 elseS=0,so I write codes as
external.regressors=cbind(lagvoljp,voljpgjr))
distribution.model="norm")
But the result is still disappointed that the coefficient which reflect
volatility spillover effect is always not significant and almost 0. I tried
other dataset, but the result is the same.

Here is my question,
How could i change the code to let the S reflects the effect from
counterpart market? Are there any codes I write have problems and
un-reasonable part?

Thisi is important for me!
Thank you so much!

Sincerely,
Zoe
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://stat.ethz.ch/pipermail/r-sig-finance/attachments/20110802/a881d7ca/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: gjr.jpg
Type: image/jpeg
Size: 13969 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-sig-finance/attachments/20110802/a881d7ca/attachment.jpg>
#
Dear Zoe,

The study of volatility spillover effects between markets is usually 
studied using multivariate GARCH models, which take into account lagged 
as well as own shocks and volatility. See for example the BEKK, VEK and 
Generalized DCC models. For bivariate modelling this 'should' be 
computationally feasible/straightforward, but you should look outside 
the rgarch package at present for these type of multivariate GARCH 
models (hint: mgarchBEKK, ccgarch).

Regards,

Alexios
On 02/08/2011 15:55, XI ZHANG wrote: