[R-sig-finance] [QUAR] Re: some doubts in garch models

Tim Hesterberg timh at insightful.com
Thu Mar 2 23:10:41 CET 2006


Note that different people use different terminology.

The terminology I see most is:
	kurtosis = E((X-mu)^4) / var(X)^2 - 3
With this terminology, kurtosis > 0 indicates longer tails than Gaussian.
The S-PLUS kurtosis function computes this kurtosis.

Alternate terminology:
	excess kurtosis = E((X-mu)^4) / var(X)^2 - 3
	kurtosis = E((X-mu)^4) / var(X)^2

Tim Hesterberg

========================================================
| Tim Hesterberg       Research Scientist              |
| timh at insightful.com  Insightful Corp.                |
| (206)802-2319        1700 Westlake Ave. N, Suite 500 |
| (206)283-8691 (fax)  Seattle, WA 98109-3012, U.S.A.  |
|                      www.insightful.com/Hesterberg   |
========================================================

>Are you sure that you are getting kurtosis and not
>excess kurtosis?  I've never seen market data that
>weren't fat tailed.  If you have an example, I'm keen
>to see it.
>
>Using 'qqnorm' should give you a good visual clue.
>
>Patrick Burns
>patrick at burns-stat.com
>+44 (0)20 8525 0696
>http://www.burns-stat.com
>(home of S Poetry and "A Guide for the Unwilling S User")
>
>Ricardo Zambrano Aguilera wrote:
>
>>	Thank a lot mr Diethelm Wuertz and mr Patrick Burns my best regards..., but then i have another doubt... and it is a big 
>>	if my data... the exchange rate (dollar (USA) - peso(Chile)), between 1-1-2001 - 12/12/2005, (i´m working with the log-returns), my first intention is to aplly an ARMA -GARCH , but  their Kurtosis is 1.182 < 3,  ¿ can i use an garch modelling??? , because sometimes ,like the 40% , when i did simulate garch procces ( i made like 1200 obs) , the kurtosis is nongreater to 3?? 
>>	atte a ustedes Ricardo Zambrano Aguilera



More information about the R-sig-finance mailing list