The RATS Software Forum

A newer version of the Chan and Maheu Jump GARCH models was posted (10 August 2012) at http://www.estima.com/forum/viewtopic.php?f=8&t=1578

Im trying to combine VAR and jump GARCH, anybody can help to do the programming ? Tks:)

I think before you ask about programming, you think about what it means to combine those ideas. The jump GARCH models are for univariate processes. For a univariate process, the addition when the Poisson process fires is a univariate Normal with mean theta and variance deltasq, thus 2 free parameters. If you try to generalize that to a multivariate process, you now need to add a Normal with a mean and covariance matrix with a total of N(N+1)/2 free elements. Adding something with a fixed (but unknown) covariance matrix is quite a bit more stringent than simply adding variance.

Thanks for your response.
My purpose is to investigate the jumps and volatility linkages between several markets(say,three). How can I modify the Rats example procedure(which is a univariate model ) into a multivariate GARCH model?
Thanks again.


Zhang

Unless you have a paper that you can cite that's doing what you want, then this is not yet a programming question - it's a theory question about how you would extend the jump model to a multivariate case.

there's a paper"Jumps and dynamic volatility relationship between equity and bond returns" by Elton Daal etc. who use a multivariate GARCH jump model, i can't find the link, May i have you email address so i can email you the soft copy?

Thanks alot.

Zhang

Hi,

I am using bivariate EARJI-EGARCH jump model similar to Chan in article above, but I do not know how to do codes in RATS, maybe somebody could help me? in attached file you will find model estimation and results I would like to obtain in my paper work.

linaUtkaite wrote:Hi,

I am using bivariate EARJI-EGARCH jump model similar to Chan in article above, but I do not know how to do codes in RATS, maybe somebody could help me? in attached file you will find model estimation and results I would like to obtain in my paper work.

Have you checked with the authors about getting their data set?

TomDoan wrote:

linaUtkaite wrote:Hi,

I am using bivariate EARJI-EGARCH jump model similar to Chan in article above, but I do not know how to do codes in RATS, maybe somebody could help me? in attached file you will find model estimation and results I would like to obtain in my paper work.

Have you checked with the authors about getting their data set?

Yes, I wrote them: but one of the author left academia and do not have access to the data sets and from other I did not get any responds in my two e-mails.

Teachers at university refering to this model as very complicated one and I did not get any help, this is why I noticed that model introduced above is very similar with one I would like to use and now trying to find solution how to implement it.

Yes, it's quite a bit more complicated than the univariate case. They don't just have a single Poisson process, but two independent ones. Have you tried estimating univariate models to see if there is strong evidence for jump GARCH in them?

Thank you Tom for your replay, I will try to look first at univariate model.
Could you explain me one more thing, I am really new with RATS software, and I willing to understand it.
I am interested in the data file dija.txt and cannot understand how you got logret and logrange? You calculated it?

linaUtkaite wrote:Thank you Tom for your replay, I will try to look first at univariate model.
Could you explain me one more thing, I am really new with RATS software, and I willing to understand it.
I am interested in the data file dija.txt and cannot understand how you got logret and logrange? You calculated it?

The data file was provided by Chan and Maheu with those series already constructed. However, the following will construct all but the first value for the log return:

set logrange = log(high/low)
set logret = log(close/close{1})

The first data point for logret would require the closing price for the last data point before the data set. Obviously, they had that and just exported the data with the transformed series.

Dear Tom,
Please is it possible to use @GARCHFore to forecast the series out of sample as in Chan and Maheu(2002) paper?
Thank you

Dear Tom,
please does it make sense to use:

group garchmod hfrml>>h ufrml>>u
forecast(model=garchmod,from=2008:2,steps=5)

Not sure where i should start.
Many Thanks

No. That doesn't work because it doesn't take into account the jump process. The one-step variance calculations are included in the newer version of the programs.