|
PureBytes Links
Trading Reference Links
|
If you want to get academic, the system buy and hold has 0 parameters
and it is very likely to make money on equities.
Gabriel
-----Original Message-----
From: Bilo Selhi [mailto:biloselhi@xxxxxxxxxxx]
Sent: Tuesday, February 03, 2004 3:39 PM
To: marc miller; omega-list@xxxxxxxxxx
Subject: Re: e-mini...search for the grail continued
how can that be not so?
even the simplest of the statements close > close[1]
has 3 parameters,
1. series = close
2. close is same as close[0] meaning [n] is the parameter
3 .close[1] is again [n] is the parameter
we are dealing with expert type systems in TS, meaning
we are trying to fit rules to data, you can't have rules without
parameters or data without parameters.
every system is a curve fit and every curve fit is optimized to
find the best parameter set for the model.
the questions is not the curve fit but
how long will this curve fit hold up for before you have to refit the
model to the data? the reason why it does not hold up is because data is
drifting out of the model fit, not vv.
if it holds up then profit can be taken out.
obviously the longer it holds up the better, but it won't hold up
indefinitely unless data is stationary or deterministic ( not the case
with price data ), or rules and parameters are so generalized that they
will fit any data ( not the case, impossible ) so then the remaining
part is adaptation, like everything else is in life. environment
changes, rules change , parameters change.
so data change --> adapt rule or parameters or both,
ideally you want to find rules that generalize well to keep rules same
and
just change the parameters to fit the data.
even non parametric models are just called non-parametric in the
statistical point of view, all of them are parametric.
to summarize: data changes, rules better not, parameter tweaking is ok
as long as the model holds up for time enough to extract profit.
optimization is necessary to find optimal
parameters, model can only be validated on unseen data and if data
parameters don't
change then model will hold up until data holds up, else it's a curve
fit.
Helix developer claims that he found rules and parameters that are
stable enough for
5 years, so that they will work for next N? years... this maybe correct
with 10 point stop
on 5 minutes but eventually data will change and parameters or-and rules
will have to also.
so far Striker shows about $1000 profit for half January, developer
claims out of
sample of 14 months... ES data may not have changed for the past 14
months, hard to believe or
who knows the truth, rules and parameters could have been changed, we
simply don't know. Striker reports and Futures truth stats will show.
bottom line, statistical models can not work indefinitely without
tweaking if data is non stationary... and tweaking is ok and even
necessary, whether it is done manually or automatically.
bilo.
----- Original Message -----
From: marc miller
To: omega-list@xxxxxxxxxx
Sent: Tuesday, February 03, 2004 2:58 PM
Subject: Fw: e-mini...search for the grail continued
And I forgot to add this:
ANY system that "optimizes" or "trains" its indicator inputs on a
dataset is IMO total crap.
Best Wishes,
Marc Miller
----- Original Message -----
From: "marc miller" <marcmiller@xxxxxxxxx>
To: <omega-list@xxxxxxxxxx>
Sent: Tuesday, February 03, 2004 11:42 AM
Subject: Re: e-mini...search for the grail continued
>
> Even the VERY BEST system will get run over by forces more numerous
> and perhaps more powerful than can be programmed into that system OR
> meet the eye. For Instance:
> Take a dominant cycle high or low point, forecast in measured 'cycle
time'
> to occur at a future date (i.e., connect previous lows/highs, then
project
> out... If the market has been trading sideways, just "meandering"
just
> before this occurs, then "takes off like a rocket"- most
> math/stat/probability "systems" are going to signal you to exit or
enter
> prematurely. Every one of us that have been around even for a short
period
> of time have witnessed this.
> Even then, "phase" of cycles are known to "shift". This is just ONE
factor
> that NO-Body can program into ANY system.
> In my life experiences, having a good math/stat/probability program,
> adhering to its rules, but more importantly, diversifying its use
> among as MANY instruments as you can keep track of helps alot. On a
> down day in
the
> emini, your model is going to have you both long and short on many
> instruments, helping to remain more delta neutral. The challenge and
> fun is being able to recognize the market's
metamorphosis
> yet is secondary to making $$.
>
>
> Marc Miller
>
> ----- Original Message -----
> From: "Bilo Selhi" <biloselhi@xxxxxxxxxxx>
> To: "Infinity Brokerage" <jeffg@xxxxxxxxxxxxxxxxxxxxx>; "'Ron Hudson'"
> <ron@xxxxxxxxxxxxx>; <omega-list@xxxxxxxxxx>
> Sent: Tuesday, February 03, 2004 11:16 AM
> Subject: Re: e-mini day trading systems
>
>
> > every day is a new day...
> > every day a new sucker is born...
> > every trade is a new trade ...
> > every day there is new system...
> > every day is new line of code...
> > every day brings a new idea...
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > but Duke&Duke always makes the commission !
> >
> >
> >
> >
> >
> >
> >
> >
> > bilo.
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > ----- Original Message -----
> > From: Infinity Brokerage
> > To: 'Ron Hudson' ; omega-list@xxxxxxxxxx
> > Sent: Tuesday, February 03, 2004 1:58 PM
> > Subject: RE: e-mini day trading systems
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > Every trader needs to think systematically and understand how past
> > price movement may influence future prices. As you said, there is
> > probably not
a
> > system that will always work.
> >
> >
> >
> >
> >
> >
> >
> >
> > This is why Tom Skilling goes to work everyday and why I go to work
> > everyday. Everyday is a new day.
> >
> >
> >
> >
> >
> >
> >
> >
> > J Gil
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
>
|