|
PureBytes Links
Trading Reference Links
|
What you sound like you are describing is a weighted model. In this
case you would take your intial signal and apply perhaps more than one
filter.
Once you have found your filters, then you have to determine the
amount of influence both positive and negative they have on the
outcome of the trade. Once you know that, you then combine them into a
binary equation without giving all of the filters equal weight. Filter
one, when it is positive, might contribute 0.5 to the equation, where
filter two only contributes 0.25 when it is positive.
In order to have a trade signal, some combination of the filters have
to contribute enough fractional points to sum to one.
The filters are your constraints and the weightings are the
coefficients of the contraints.
Weight testing is not for beginners, but once you figure it out, it
changes the whole realm of system design.
Filters are not equal in their contribution and should not be treated
as if they are. If all constraints in the equation have to have a
coefficient of 1, then it limits the possible solution sets in a big way.
Roy's newsletter is great. But anyone who thinks they won't benefit
from it shouldn't read it. I don't know who that might be but .....
Have fun!
--- In equismetastock@xxxxxxxxxxxxxxx, mgf_za_1999 <no_reply@xxxx> wrote:
> Thanks super, I will have a look at this trend quality indicator.
>
> BTW, are you trying to convince me to subscribe to Roy's newsletter?
> You would never do such a thing so I am probably getting the wrong
> impression, but I think you may yet convincing me....
>
> I am really wondering about bringing the optimisation and actual
> trading process closer together. It is easy to optimise (easy to
> optimise any, even a bad system, not so easy to find a good system) on
> say forecasting performance or correlation between signal strength and
> market returns. If you optimise only on entry and exit point (as I
> often do) then you don't really take the signal magnitude into account
> - it would really help if you could say a signal of say 3 is very
> bullish, but in this case you can't. If the signal crosses zero, then
> it is bullish, regardless of its magnitude.
>
> This is what I am contemplating. If the market rises from say 50 to
> 100, then over this time a new kind of trading system may predict a
> higher average return than usual. This is all fine, but for such a
> system, when do you actually enter or exit the market?
>
> Maybe the best way is, as Paul suggested, to band the signal and
> optimise on the width of the band or pick it arbitrarily, maybe as a
> fraction of historical price movements.
>
> I know Ehlers did a lot of work on signal to noise ratio using the
> Hilbert transform and we implemented a lot of these ideas. Another
> idea would be to use some fuzzy logic, neural network based
> classification system to pick a 'buy' or 'sell' class and apply that
> to the trading decision.
>
> I once read an article (in International Journal of Forecasting) where
> the guys, in stead of trying to predict the market, just tried to
> predict a + or - if it would rise or fall, and then trade on this.
> The article was really about how this approach outperformed the
> traditional forecasting way of predicting 'how much' the market would
> rise or fall. This is also really what a traditional trading system
> is, a + or -, since the magnitude does not matter.
>
> Regards
> MG Ferreira
> TsaTsa EOD Programmer and trading model builder
> http://www.ferra4models.com
> http://fun.ferra4models.com
>
>
>
>
> --- In equismetastock@xxxxxxxxxxxxxxx, superfragalist <no_reply@xxxx>
> wrote:
> > You have to test various filters. I think you already know this. Each
> > filter has to be designed to improve specific issues with the system
> > you're designing. Each filter will remove something and add something
> > else. It's always a balance between the good and evil of every filter.
> >
> > From your description of your hypothetical system, you only want to
> > remove the noise, which would be the problem based on what was
> > described. Noise causes the whipsaws with crossovers.
> >
> > A noise filter that will work for this specific case was developed by
> > David Sepiashvili from Carnegie Mellon. It's called the trend quality
> > indicator, or filter.
> >
> > It's in the April 2004 issue of Technical Analysis of Stocks and
> > Commodities. However, the code in the magazine is wrong and so is the
> > code on the Equis site.
> >
> > Roy published the correct code in his newsletter. He won't post it on
> > the site. However, I think you will be able to find the errors and fix
> > the code based on your own experience and skills.
> >
> > You may have to read the article several times. David is a briliant
> > time series mathematician, but his writing is often sketchy at best.
> >
> >
> >
> >
> >
> > --- In equismetastock@xxxxxxxxxxxxxxx, mgf_za_1999 <no_reply@xxxx>
> wrote:
> > > Hi there,
> > >
> > > This is a general, almost philosophical question, but some of
you may
> > > have had to deal with it before or may have some ideas on it. It
> > > pertains to automated trading and system building. If you build an
> > > automated trading system, you often create some 'signal' and trade
> > > based on this. Now, this signal could be say the difference between
> > > two moving averages, and you trade whenever they cross which will be
> > > exactly when the signal crosses the zero line.
> > >
> > > When you evaluate such a system, you buy when the signal goes
positive
> > > and sell when it goes negative. All of this is fairly general and
> > > fairly common. Note in this example, however, that the actual value
> > > or magnitude of the signal does not play any role - the important
> > > thing is when it crosses the zero line. Again, no rocket science in
> > this.
> > >
> > > I am toying with the idea to build a system based on the
*strength* of
> > > the signal rather than when it crosses zero, which introduces some
> > > problems. The question really is, when do you then enter a trade?
> > >
> > > One obvious way is to put some band around zero and trade when the
> > > signal goes outside of this band. This complicates matters as you
> > > also have to create this band. Another way is to still trade on the
> > > zero-line crossover, which is fine but then you could just as well
> > > stick with the original where the whole system is built around
this.
> > > You could use a moving average of the signal line and trade when
they
> > > cross, but this is just the same as trading a zero line crossover.
> > >
> > > What ideas do you have for automatically trading a system using some
> > > signal, that are not triggered by this signal crossing zero?
> > >
> > > Regards
> > > MG Ferreira
> > > TsaTsa EOD Programmer and trading model builder
> > > http://www.ferra4models.com
> > > http://fun.ferra4models.com
------------------------ Yahoo! Groups Sponsor --------------------~-->
In low income neighborhoods, 84% do not own computers.
At Network for Good, help bridge the Digital Divide!
http://us.click.yahoo.com/EpW3eD/3MnJAA/cosFAA/BefplB/TM
--------------------------------------------------------------------~->
Yahoo! Groups Links
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/equismetastock/
<*> To unsubscribe from this group, send an email to:
equismetastock-unsubscribe@xxxxxxxxxxxxxxx
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
|