Risk Models

["Bilo Selhi" <citadel@xxxxxxxxxxxx>]

would like to bounce some ideas and get some
pointers on existing models for risk ( volatility )
measurements as well as risk forecasting.
( with an exception of getting into VAR or beta models )

as far as i know there are three classes of those:
- standard classic risk model, aka risk = sigma
   which is not a forecasting model per se.
   easily implemented in TS.
- riskmetrics JP model, aka based on EMA
   which is what they call 1 bar horizon
   seems to be the most practical with quite a
   following. this one is very easy to implement in TS.
- ARCH garden variety
   consensus is that it's less practical and more theoretical.
   hard to implement, although i remember a promise blip
  on Arima/Garch for TS...
 ( which i am not really interested in )

i was wandering if there are any other new or
old approaches in risk measurement?
any alternative approaches?
any good papers on those?

some papers i read recently mention extreme value
modeling of fat tails which seems to make most
sense. anything practical on that?

and finally, anyone seen any good papers on risk/volatility cluster analysis
of risk in  financial time series ( not arch based )?
seems like basic cluster analysis may help in risk measurement ie.
low, ave, high risk grouping, although i have not explored that path yet.
anything on that?
bilo.
ps. i have a good documentation on riskmetrics if anyone wants and
      the original paper on garch.