|
PureBytes Links
Trading Reference Links
|
Adrian, without splitting hairs or getting bogged down on semantics,
any process that randomly (or pseudo/semi-randomly) looks for optimal
results, is considered curve-fitting in my books.
For example, run any old price/MA crossover system on any optimizer,
and it will come up with the best parameters to fit the past data.
Now, take the signals from this optimized system, and visually inspect
the entry/exit points: the result is most likely crappy signals. The
entries and exits are probably lagging so much, or getting us
whipsawed, the system has no chance - except for those two or three
signals that "caught" the odd major past swings, never to be repeated
again.
The only true way to develop a basic tradeable system, is to use some
very specialized hardware/software combination, the ultimate pattern
recognition combination: our own eyes and brain.
My point is:
Forget the System Tester and Explorer.
Design your trading system in a visual way. Look and see for yourself
what your entry & exit system signals are doing. Look for
correlations to other relative market data.
Once you have developed promising system signals, by all means
backtest it in a way that mirrors the real world as close as possible.
Regards,
jose '-)
http://users.bigpond.com/prominex/pegasus.htm
--- In Metastockusers@xxxxxxxxxxxxxxx, "Adrian Pitt" <apitt@xxxx>
wrote:
> So Jose,
>
> Going by your comments, you seem to define optimisation as the
process
> of introducing a variable where you have to select a number from a
range
> of numbers. Some people suggest that by using price patterns instead
> that they aren't optimising, but we both know this to be wrong as
well.
> So my question is Jose, how do you go about designing a system
without
> using optimisation at one point or another? And just picking a
number
> for a variable and saying you didn't optimise it doesn't mean that
at
> all...it merely says you randomly chose a number and then stuck your
> head in the sand hoping it would work. Your email says to "identify
a
> characteristic within the data".. How does one do that without some
> level of data mining..i.e optimisation?
>
> Bottom line, regardless of what anyone on the planet thinks, every
> single model and strategy has been optimised. And that includes buy
and
> hold. People have done endless research to show that buy and hold
'has'
> been an effective strategy for the market as a whole. In other words
> they optimised the strategy process, even though there are basically
> ZERO variables. Ultimately to arrive at a model that in some way
> accurately reflects how the markets work requires optimisation..end
of
> story, and no one can prove otherwise. There is a subtle line
though
> between optimising and data mining. Data mining is simply the
scanning
> of data for patterns, random of otherwise until you find something
that
> happens to do well. This is highly likely to be just a random
> occurrence though unless you have some way of showing it not to be
> random.
>
> So back to you Jose. IF you don't do any curve fitting or
optimisation,
> how do you go about designing a trading model and finding a
> characteristic 'within the data'?
>
> Regards,
> Adrian
>
> -----Original Message-----
> From: Jose [mailto:josesilva22@x...]
> Sent: Saturday, 31 July 2004 3:28 AM
> To: Metastockusers@xxxxxxxxxxxxxxx
> Subject: [Metastockusers] Re: Trading System Presentation
>
>
>
> Having coded & backtested 000's of trading ideas for paying clients
> over the last five years, these are my personal comments based on my
> experience:
>
>
> "Identify a characteristic within the data."
> "Pick a trading time frame."
> "Test entry ideas."
>
> "After finding a good entry, test exits."
> "Add filters to improve the system performance."
> "Test money management rules."
> "Complete out-of-sample testing."
>
> Reasonably good system development ideas.
>
>
> "Optimize the entry signal across as many commodities as possible."
>
> Curve-fitting.
>
>
> "He defines curve fitting as either the overuse of rules, filters,
or
> stops when developing a trading system on a sufficient amount of
past
> data;"
>
> Not so.
> One can curve-fit system results with just one simple set of entry &
> exit rules, on any amount of past data.
>
> Curve-fitting is simply tailoring (mostly illogical) system rules to
> fit past results. Often these rules manage to (mostly by
coincidence)
> capture one or two large price swings, enough to make a final
> difference to test results.
>
>
> "He is confident that if he can get enough trades then total profit
is
> the only meaningful statistic to prove that a system works."
>
> Pfff...
> Buy & Hold fits this description exactly, yet drawdowns and lack of
> sufficient profit generated over time make this strategy totally
> unsuitable to most active traders with limited funds.
>
>
> Nevertheless, perhaps Mr Fitschen's presentation is worth listening
> to. Has anyone been to any?
>
>
> jose '-)
> http://users.bigpond.com/prominex/pegasus.htm
>
>
>
>
> --- In Metastockusers@xxxxxxxxxxxxxxx, "luigilms_2000"
> <luigilms_2000@xxxx> wrote:
> > This is an announcement of a free presentation by Keith Fitschen,
> > the developer of the Aberration Trading System, in regard to
> > developing trading systems and the pitfalls involved. I'm
confident
> > that they will learn valuable information in regard to system
> > development. Below are the details:
> >
> > SATURDAY, AUGUST 7, 2004, FROM 10AM - 2PM, DOUBLETREE HOTEL OHARE
> > AIRPORT, 5460 NORTH RIVER ROAD, ROSEMONT, IL. (847)- 292-9100
> >
> > SATURDAY, AUGUST 21, 2004, FROM 10AM - 2PM, HYATT SAN JOSE AIRPORT
> > 1740 North First Street, San Jose, CA. 95112,(408)- 993-1234
> >
> > Important Seminar Information,(Note: All pre-registered attendees
> > will receive a free trading system, $500 value,RSVP to
> > Leonard@xxxx or call 800-858-2340.
> >
> > Trading System Development Outline
> >
> > Keith Fitschen presents a detailed study on the theory of system
> > development, an application of that theory in which he designed a
> > profitable system, and an overview of his commercially available
> > Aberration Trading System. Aberration is a multi-commodity trading
> > system that was released in 1993. Futures Truth Magazine has twice
> > named it "ONE OF THE TOP TEN TRADING SYSTEMS OF ALL TIME" (1997
and
> > again in 2000). The speaker is sponsored by Angus Jackson, Inc.,
> > (www.angusjackson.com), a futures and commodity brokerage firm
based
> > out of Fort Lauderdale since the mid-1980s. Angus Jackson Inc.
> > specializes in executing various mechanical trading models for its
> > clients.
> >
> >
> >
> > System development, as defined by Fitschen, is a process to
identify
> > a means of profitably exploiting noisy market data. Fitschen
> > summarized the process as a series of steps:
> >
> > . Identify a characteristic within the data.
> >
> > . Pick a trading time frame.
> >
> > . Test entry ideas.
> >
> > . Optimize the entry signal across as many commodities as
> > possible.
> >
> > . After finding a good entry, test exits.
> >
> > . Add filters to improve the system performance.
> >
> > . Test money management rules.
> >
> > . Complete out-of-sample testing.
> >
> >
> >
> > Curve fitting. The biggest pitfall to system development, in his
> > opinion, is curvefitting, which over-optimizes past data to
generate
> > results that can't be achieved in real time trading. He defines
> > curve fitting as either the overuse of rules, filters, or stops
when
> > developing a trading system on a sufficient amount of past data;
or
> > the proper application of rules, filters, or stops on an
> > insufficient amount of past data.
> >
> > Demonstration. Turning from theory to application, Fitschen
> > demonstrated the system development process. He followed the steps
> > he had outlined earlier.
> >
> >
> >
> > Another Example. Donchian breakout entries are a well-known
> > commodities trading technique. This system buys on a close higher
> > than the highest close in the look back period and sells on a
close
> > lower than the lowest close in the look back period.
> >
> >
> >
> > Performance Metrics. The most important measure of a system to
> > Fitschen is total profitability. He is confident that if he can
get
> > enough trades then total profit is the only meaningful statistic
to
> > prove that a system works. He also likes to look at the gain/pain
> > ratio, which he defines as the average annual gain divided by the
> > average annual drawdown.
> >
> >
> >
> > Guidelines for Buying and Trading a Mechanical System.
> >
> >
> >
> > Overview of Aberration Trading System.
> >
> >
> >
> >
> >
> > Keith Fitschen has developed commodity systems for the last 20
> > years. He retired from the Air Force in 1988, and has been
trading
> > commodities since then. Aberration has been marketed since 1993.
> > Keith has also developed Aztec, a shorter-term commodity system,
and
> > Imaster, a short-term index system. Additional information about
his
> > work can be found at www.trade-system.com.
------------------------ Yahoo! Groups Sponsor --------------------~-->
Make a clean sweep of pop-up ads. Yahoo! Companion Toolbar.
Now with Pop-Up Blocker. Get it for free!
http://us.click.yahoo.com/L5YrjA/eSIIAA/yQLSAA/zMEolB/TM
--------------------------------------------------------------------~->
Yahoo! Groups Links
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/Metastockusers/
<*> To unsubscribe from this group, send an email to:
Metastockusers-unsubscribe@xxxxxxxxxxxxxxx
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
|