|
PureBytes Links
Trading Reference Links
|
I have been using NN models a lot and have developed some amazingly
profitable models, amazing to me, because these models most of the
time beat my best and only mechanical signal detection system when
comparing the annualized return and percent wins. They seem to be
generating lesser number of trades when compared to my mechanical
system, however. But when tested over a large period, the number of
trades are greater than 30 which makes the statistics statistically
valid. Also, I have the exact same signals every time I generate
outputs with my models.
I don't want to be stirring up a controversy, but in my opinion, the
merger of NN with mechanical systems running on a super-computer is
the ultimate combination in the world. I don't know for sure which
is superior, the human brain or the super-computer, but I predict a
draw, because I love to predict! :)
By the way, I have been using TradingSolutions to develop my NN
models and also my Entry-Exit Trading Signal Detection and
Verification mechanical System, because of its spreadsheet format
which makes it easy to detect and verify signals. I use AB as my
main Interpretation platform for the detected and verified signal.
In my view Verification is more difficult than Detection and
Interpretation is more difficult than Verification and only AB has
been able to handle it, for which I am indeed very thankful to TJ
for producing such a wonderful software and for the AB user
community for creating several great systems.
The NN uses the mechanical system signals as inputs, and the NN came
out with superior results when compared with the detected signal
system, but it is not superior when compared to the detected,
verified and interpreted signal system which is actually done by the
human brain, but again when the NN output goes through the
verification and interpretation stage it is equally good. So, I
have to predict a draw.
Regarding Vantage Point, in my opinion they don't know what they are
doing, because they seem to be using different models for different
instruments (stocks, bonds etc.) withe the result that most probably
the models will die after some time. I use the same models for all
of my instruments which in my opinion is the way it should be
because as far as the computer is concerned they are all a bunch of
numbers and so the stocks/markets personality doesn't and shouldn't
matter.
rgds, Pal
--- In amibroker@xxxxxxxxxxxxxxx, "DIMITRIS TSOKAKIS"
<TSOKAKIS@xxxx> wrote:
> I have tried some [most promissing] NN in the past.
> My test was quite simple : I ran the algorithm for a 1000-bar
famous
> stock and saved the result. Then, 10 min later, I ran the same
test
> on the same stock. The new result was quite different ["quite"
means
> a "Sell" instead of "Buy"]. I repeated 2-3 more times the test. I
> also noticed differences in the NN training series. I carefully
> copied my tests and sent them to the NN company with my Best
Regards.
> I still wait for their reply. I am patient enough but, I am afraid
> they will not send any kind of reply. Then I sent a suggestion for
> their own use. Instead of running 100 individual NNs for the 100
N100
> stocks [and enjoying, perhaps, the perfect 50%-50%], it would be
> better to run a composite StochD for the whole N100 market. It was
> the day of a bearish MeanStochD divergence and, my NN-less
prediction
> was the beginning of a [new] downtrend. They didnt make any
comment.
> I suppose they make money selling NNs. I also suppose they trade
the
> MeanStochD divergences. But, I have not received any kind of
> confirmation for these thoughts. The [absolute] silence of these
NN
> vendors is, perhaps, the best way to communicate in the 21st
century.
> Dimitris
> PS : If anybody knows an NN which give the same output for 10
tests
> on the same 1000-bar data series, please let me know...
> If you also know any NN software "understanding" the
directionality
> of the market, please respond.
>
> --- In amibroker@xxxxxxxxxxxxxxx, Yuki Taga <yukitaga@xxxx> wrote:
> > I'm afraid I have to agree about the worthlessness of NN. Here
in
> > Japan, with *very* powerful computers, people have attempted to
use
> > NNs to extract money from the stock market. The market has
> extracted
> > money from them instead. Mostly, the effort here has been
abandoned
> > until sometime in the future when there is some kind of
fundamental
> > change in computing power. Other than that, there are the usual
> > collection of hustlers selling black boxes that are just about
> > guaranteed to be next to worthless.
> >
> > The idea that these programs can "learn" in any real sense is
simply
> > not true. Not in any real sense, and not compared to even how
> > chimpanzees learn, let alone human beings. The state of
computing
> > does not appear to be anywhere near this realm yet. Maybe in
> another
> > (human) generation or more.
> >
> > Yuki
> >
> > Sunday, September 5, 2004, 10:32:07 AM, you wrote:
> >
> > m> I also vote NO on Neural Networks. The Neural network you
were
> > m> born with is thousands of times better than any digital Neural
> > m> network can ever be. We simply fail to use it as well as
> Einstein
> > m> used his. Ron D ----- Original Message ----- From:
> > m> Michael.S.G. To: amibroker@xxxxxxxxxxxxxxx Sent: Friday,
> > m> September 03, 2004 6:52 PM Subject: RE: [amibroker] Re:
Neural
> > m> Netowrks in AmiBroker
------------------------ Yahoo! Groups Sponsor --------------------~-->
$9.95 domain names from Yahoo!. Register anything.
http://us.click.yahoo.com/J8kdrA/y20IAA/yQLSAA/GHeqlB/TM
--------------------------------------------------------------------~->
Check AmiBroker web page at:
http://www.amibroker.com/
Check group FAQ at: http://groups.yahoo.com/group/amibroker/files/groupfaq.html
Yahoo! Groups Links
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/amibroker/
<*> To unsubscribe from this group, send an email to:
amibroker-unsubscribe@xxxxxxxxxxxxxxx
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
|