|
PureBytes Links
Trading Reference Links
|
problem solved, had simple problem with equation,
parms estimating well, in matlab for now.
whoever wants the matlab simulation and
estimation demo for this log range model , let me know,
i can send the m files ( need optimization toolbox for that ).
model works pretty good on simulated as well as on
real time series data.
i believe that this rather simple stochastic model is probably
the best backbone type statistical risk model for intraday trading, a good
starting point at least but needs to be combined with other
techniques to get those good looking tight entries and exits around the
pivots.
it is obvious that typical garch type models are not applicable to adaptive
entries and exit theme not because of the complexity but because of
the difference in the traditional proxy used. same goes about sigma models
and risk
metrics... which is why this model is the best at this time.
strangely only a couple of people
seem to be interested in this absolutely key topic...
dudes, risk in trading is everything...as volatility fluctuates so does risk
and you have measure it adaptively... average true range won't cut it.
questions are still have are:
who wants to code TS DLL for parameter estimation and adaptive entries/exits
?
who's using what risk models in their own systems?
anybody else using log range stochastic models?
regards...
bilo.
----- Original Message -----
From: "Bilo Selhi" <biloselhi@xxxxxxxxxxx>
To: <systems-only@xxxxxxxxxxxxx>; "Omega List" <omega-list@xxxxxxxxxx>;
"Code List" <code-list@xxxxxxxxxxxxx>
Sent: Wednesday, November 14, 2001 5:32 PM
Subject: SO_Re: adaptive entries, exits. MLE, QMLE est. in Stoch. Vol.
models
> ok, going to add more in hope that someone could help out.
> i use the following log range stochastic volatility model,
> Rn = mr + p(Rn-1 - mr) + b*et
> where
> R is normally distributed with mean mr
> Rn-1 last bar log range
> mr is mean of log range
> p - is persistence parameter ( mean reg. coeff )
> Rn - current bar log range
> et - normally distributed innovations ( disturbances ) N[0,1]
> b - innovation std parm
>
> this is a near gaussian log model with three parameters:
> mr, p and b. mr need not to be estimated.
> 1. i need to be able to estimate p and b with mr known.
> 2. i need to be able to estimate b only with p and mr known.
> the second case is more important to me...
>
> from what i can infer, in case 1. the best estimator for this model
> is quasi-maximum likelihood estimator or straight max.
> likelihood estimator. however in case 2. i am not sure that i need
> mle/qmle for that since i only need to estimate one parm, might be
> able to accomplish that with transformation.
>
> so,
> i need to be bounce a few mainly math questions, ideas with someone
> who knows mle or qmle or stoch volatility models parameter estimation
> ( similar to arch, garch ). qmle uses kalman but i got some questions
> on that technique.
> thanks. hope someone out there can help out a bit.
> bilo.
> ps. this is not garch p,q type model.
>
>
> > working on adaptive entries and exits for
> > a trading system i ran into a wall on
> > parameter estimation of my stochastic
> > volatility model, specifically,
> > qmle/kalman and mle based parm estimation for
> > this specific model.
> > need minor clarifications there.
> > if anyone is familiar with those parm estimation methods or
> > know those econometric techniques or are interested in adaptive
> > enties / exits for a trading system ( risk modeling ),
> > please e-mail back asap.
> > good trading.
> > bilo.
> >
>
>
> http://www.markbrown.com/systems-only
>
|