|
PureBytes Links
Trading Reference Links
|
Hi
I'm seeing GA etc embedded into many XL workbooks. Most trading and analysis
workbooks have at least one or two "advanced" procedures used routinely.
Hurst Exponent, GARCH (in the form of GA1 or GA2), etc. appear often. Other
lesser know techniques have been adapted to workbooks and are in regular
use.
The traders that use them have usually identified a niche with an edge,
i.e., a sustainable competitive advantage and only work within their niche.
I don't know anyone who uses it in "scanning" large stock populations. This
often indicates a lack of an identified sustainable edge that exists prior
to the use of a trading system to harvest the edge.
Pre-processing is important and often includes time series techniques.
Prognosis is a newer program that has uses standard time series techniques
and is easy to use.
http://www.profiware.com/index.html
Demetra will also do batch processing. A major re-write will be appearing
soon.
Usually the traders have sufficient tools and don't have to proceed to
neural nets.
Best regards
Walter
----- Original Message -----
From: David Jennings <davidjennings@xxxxxxxxxxxxx>
To: <metastock@xxxxxxxxxxxxx>
Sent: Thursday, February 08, 2001 3:37 AM
Subject: Re: Artificial Intelligence, expert and neural
> Jeff,
> Amazing how some of this stuff comes back, I had forgotten about
> leptokurtosis - sounds like a disease doesn't it.
> I did a significant study in this area on the Brent Oil market in about
> 93/4. I suspect I threw it into the bin when I left Shell. I havn't
picked
> up Peters' book in a number of years either.
> I think I'm going to have to take some time out and fiddle with the
> equations. I'd forgotten how much I enjoyed this area.
>
> As regards yr. other posting, I don't think you are being confrontational
at
> all. I completely agree. I guess, that my sense of it is that an excellent
> fundamental analyst will always outperform an excellent chartist. What
will
> let him down will be the number of stocks he can manage. Thus an average
> chartist will tend to out perform a good analyst. At the limit, I see a
> black box outperforming the individual purely because of scope and freedom
> from emotion.
>
> On your other point regarding data, I couldn't agree more. If you look at
a
> chart on the screen, one can ignore the bad tick. In a network, you can't.
> Cleanliness of data is paramount.
>
> I must confess I've given up and what little success I've had tends to
come
> from technical analysis rather than NNs. However, I would still commend
> using a GA to search for indicator settings. It's extremely fast.
>
> Take care
>
> DJ
> ----- Original Message -----
> From: "Jeff Haferman" <haferman@xxxxxxxxxxxxxxxxxxxxxxx>
> To: <metastock@xxxxxxxxxxxxx>
> Sent: Wednesday, February 07, 2001 4:00 PM
> Subject: Re: Artificial Intelligence, expert and neural
>
>
> > David Jennings wrote:
> > >
> > >As an aside, whilst they fit markets quite well, different parameters
are
> > >required over different time frames. Thus the question to you guys is:
if
> > >this is corect (and I believe it to be) it flatly ontradicts any
fractal
> > >market hypothesis. Any assistance would be helpful.
> > >
> >
> > My opinion is that it doesn't contradict any FMH. I'm cheating
> > here and looking at Edgar Peter's "Chaos and Order..."
> > specifically his Eq. (9.2). If you don't have the book,
> > this is an equation for a Pareto-Levy distribution, which has
> > 4 characteristic parameters. [I would reproduce the Eq here,
> > but it would take some work!] Call the parameters A, B, C, and D.
> >
> > Peters states that a normal dist'n results when A=2, B=0, C=1,
> > and D=1. A fat-tailed, skewed-to-the-right (i.e. leptokurtotic)
> > dist'n results when A is between 1 and 2, and B approaches 1.
> > It turns out A is actually the fractal dimension of the dist'n.
> >
> > Now, what is the mean of this second distribution? It depends
> > on the time frame! For daily prices, if the mean is M, then
> > for 5-day returns the mean is 5*M. So, this is a case where
> > the distribution of the returns has a time-dependent parameter,
> > but the distribution is fractal.
> >
> > So, to apply this to GARCH, it really comes down to this:
> > is the fractal dimension of the distribution independent of
> > the time frame? Any other parameters of the distribution
> > do not need to be independent of the time frame.
> >
> > Jeff
> >
>
>
|