|
PureBytes Links
Trading Reference Links
|
Mr. Chan, you make an interesting point.
I have often wondered where the priorities should lie.
1) Is it more important to pre-ascertain a market's "signature" and mate
that market to a particularly effective system (trend/reversal et al.), then
just feed the "good monkey" and spank the "bad monkey" and is this in
essence, optimization in reverse... or...
2) Is it more important to work harder at designing a system that
auto-adapts for multiple signatures/conditions, or ...
3) A combination of #1 and a non-trending filter that effectively shuts the
system down when a particular level of volitility is present (or not).
So far, (and I'm admittedly early in the curve here), and perhaps because of
my daytrading background, I've tended to gravitate towards creating a robust
trending system and feed it pre-screened stocks based on their ability to
bubble high up on a list of APR and Profit% in a custom OmniTrader RealTime
module I wrote.
Running the OT module at EOD lets me catch when stocks are changing their
"character" simply because they better "fit" or "fail" the trending system.
In this way, I detect the failure before the main TS system rolls over.
Of course certain stocks tend to stay near the top of this list nearly all
the time, and there is still descretion to filter out "news", and to set a
threshold for how good or bad a monkey becomes... ;-)
Food for thought... and very open to suggestion.
Gene
<DarkMagic>
http://www.UndergroundTrader.com/
----- Original Message -----
From: "Lawrence Chan" <stnahc@xxxxxxxx>
To: "Bilo Selhi" <citadel@xxxxxxxxxxxx>; <orcas@xxxxxxxx>; "List, Omega"
<omega-list@xxxxxxxxxx>
Sent: Monday, April 30, 2001 10:53 AM
Subject: Re: Trading Systems
>
> Bilo,
>
> First time I agree on something you said :)
>
> My experience in developing and using a trading
> system is that one do not wait for the system
> to fail to perform to realize that it has failed.
>
> Most of my work in trading a system is not
> monitoring the system, instead, it is the end of day
> analysis on the data collected to see if they still
> demonstrate the same statistical behaviour that
> the system was designed to exploit.
>
> backtesting to me is the analysis of the statistical
> behaviour of the data - to see if some sort of
> consistency exists overtime, not wasting time to
> twist a system from not profitable to profitable.
>
> thus, real trading systems do not fail at all because
> they are all attached with their own premises. It
> looks like they fail to perform because they are
> applied to data that they are not suppose to be used
> on!
>
> e.g. 70% congestion 30% trend data with 1000 bar
> consistency. then one day, you notice the latest
> 1000 bars has about 60% congestion 40% trend -
> something has changed.
>
> unluckily, most people using TS will not even know
> what those premises are :)
>
> -Lawrence
>
>
> --- Bilo Selhi <citadel@xxxxxxxxxxxx> wrote:
> > you don't need any historical data to test this type
> > of system.
> > you can create synthetic price time series with
> > various
> > key statistical characteristics and test the system
> > on that.
> > there are several variables that you can play with
> > there.
> > trendiness, noise, volatility, etc...
> > you can recreate virtually any price time curve and
> > study
> > how your system works on that curve.
> > < snip all the rest >
>
> =====
> Lawrence Chan http://www.tickquest.com
> Home of trading tools NeoBreadth and NeoTicker series
>
>
|