[R-SIG-Finance] Problem with garch (tseries)

[Adrian Trapletti]

Volatility of financial time series often exhibits a very persistent 
behaviour which is not well represented by plain GARCH. Many models have 
been developped to address this issue among others IGARCH and long 
memory GARCH. My personal point of view is that this almost unit root or 
long memory behaviour is due to the fact that many structural breaks 
(exogenous shocks) of different size occur randomly from time to time. 
Appropriately simulated time series of such structural break models 
cannot be distinguished from long memory GARCH for example.

To give you some input to your question: A very robust solution is to 
use simple running averages or medians (or other robust measures of 
location) of recent realized volatility. There is a lot of recent 
research about estimating realized volatility, one very simple measure 
is to use the range (high minus low of the period) appropriately scaled 
to represent standard deviations. According to my experience forecasts 
based on these simple models are often better than forecasts based on 
complex statistical time series models, in particular if the loss 
function is not a statistical one, but an economical one.

Best regards
Adrian

>Message: 6
>Date: Fri, 18 Aug 2006 17:50:22 -0700 (PDT)
>From: michael mathews <muckjail at yahoo.com>
>Subject: Re: [R-SIG-Finance] Problem with garch (tseries)
>To: Patrick Burns <patrick at burns-stat.com>
>Cc: r-sig-finance at stat.math.ethz.ch
>Message-ID: <20060819005022.79785.qmail at web38912.mail.mud.yahoo.com>
>Content-Type: text/plain; charset=iso-8859-1
>
>As I mentioned to Joe off list I am dealing with natural Gas prices at
>houston ship channel. While I could assemble a longer price series it
>seems that nat gas has entered into a new price/volatility regime in
>the last two years which I think makes using older data somewhat
>problematic.
>
>If garch does not like "small" samples what would be a robust way to
>estimate the volatility?
>
>thanks for your input
>
>michael
>