GARCH Models

anyone use them? If you do, what is the rationale behind Lambda, gamma, alpha, and beta? This was for a GARCH (1,1) I guess for Lambda, the recommended risk models use 94%. Why is that? Also, I know that gamma+alpha+beta=1. When I saw someone using an example (didn’t explain in detail why), they threw out Gamma 5% Alpha 5% beta 90% Why the dramatically larger beta? Thanks.

“If you do, what is the rationale behind Lambda, gamma, alpha, and beta?” You either estimate them using past data and then assume they will hold in the near future. Or you use some market prices of derivatives to estimate the values implied by the market.

it helps to think of the constant, alpha and beta as your weights for the historical variance, news, and the previous days forecast for today. Together they should sum to a number less than 1, if they sum to 1 than you have an IGARCH process. The weights you use for news and forecast will give the speed of convergence to the historical variance and what you usually find is that the bulk of information will come from your previous days forecast. So if in the early morning ur sitting on a stock with hv of 1.5%, that moved 3% yesterday and your MA charts point to a forecast of 2% for today. You may reach the conclusion that the current period is atypically volatile, which suggests the forecast for today might be even higher. But because the long-term average is only 1.5 you’ll want to take that into account too. In that situation GARCH will help with the weightings. j.p morgan came up with the 6% for the news weighting in their riskmetrics EWMA which means their forecast weight is 94% when historical volatily is zero.

thanks Fatman, that helps a lot.

use them as amusements,as long as you realize that at a fundamental level it is all horribly flawed

yeh i agree these models are pretend tools that will perform miracles if certain facts should turn up in the right form

But boy do they impress people at parties!

I just need something to anchor my bias! Another question. If Lambda at a 94% level is 0 volatility, what is an appropriate level for the S&P? I saw a guy use 80%, is that about right? thx.

bchadwick Wrote: ------------------------------------------------------- > But boy do they impress people at parties! G(orgeous) A(nd) R(ich) CH(echnyan) Models certainly do impress at parties. Watch out though, good models require alot of inputs and constant maintenance.

bchadwick Wrote: ------------------------------------------------------- > But boy do they impress people at parties! Who the hell talks about GARCH models at a party?

mark@dirtbags Wrote: ------------------------------------------------------- > I just need something to anchor my bias! > > > Another question. If Lambda at a 94% level is 0 > volatility, what is an appropriate level for the > S&P? I saw a guy use 80%, is that about right? > > thx. before Bear Stearns went under your weights for the S&P were ('91-2008) unc.v 6.020020e-07 news 5.378848e-02 forecast 9.405899e-01 so about 94% Since then ('08/03/10->) unc.v 5.077884e-06 news 1.108235e-01 forecast 8.806917e-01 so about 88%