|
PureBytes Links
Trading Reference Links
|
well, i wrote a TS Risk modeling toolbox with which
one can play with 11 risk models: naive, sigma, risk metrics,
arch, arv stochastic, garch, kalman/jma, truerange EMA type models.
the parameter estimation is not included so you have to
hand pick the parms ( guidelines are included ) otherwise all models are there
( except VAR and realized volatility ) and
you can tweak them and see how they work.
i included comments and results of my research.
the conclusions are interesting, in short:
risk metrics is a decent kiss type model.
arch is no good.
ar stochastic models can be useful if combined with garch.
garch(1,1) is not fundamentally sound in my view and i altered it to make it better...
jma model is the best but sadly optimal parameter estimates are not possible.
so the best model is mixed breed garch/arv smoothed by jma where
risk is forecasted by mean reversion garch where risk proxy is smoothed
out by low weight jma.
who want's it, i have it.
bilo.
ps. the search for the best risk model is narrowing fast...
i would like to hear some opinions and suggestions
especially about:
- realized volatility implementation in TS
- VAR risk models
- alternative estimation techniques for garch type models
- non parameteric risk models, kernels, wavelets, adaptive filters, non linear methods
etc...
|