cocky_pusher
12 posts
msg #37620
- Ignore cocky_pusher |
8/24/2005 3:25:53 PM
Now the Financial Genetic Programming Engine described in this site http://www.tau-zweig.com has gone thru some transformations to be able to deliver results written in StockFetcher code.
|
mrbobolina
19 posts
msg #37629
- Ignore mrbobolina |
8/25/2005 5:47:13 AM
Please explain in more detail how this is done.
Thanks in advance
|
Koronbock
201 posts
msg #37632
- Ignore Koronbock |
8/25/2005 9:55:37 AM
Yes, show us an understandable example - with actual SF code - how to apply your method to make us better traders and/or letting us find better stocks.
|
cocky_pusher
12 posts
msg #37636
- Ignore cocky_pusher |
8/25/2005 1:16:02 PM
As requested … please visit again.
|
Koronbock
201 posts
msg #37637
- Ignore Koronbock |
8/25/2005 4:03:05 PM
^Well, you won't convince people this way. I have gone to your side and consider myself as willing to read, but that site is so convoluted and requires a lot of background knowledge, so nobody will take note.
What I would like to see is some - basic and short - example (not pages of text) where you show us that your approach is better of what we already have here.
Why don't you just take 2 or 3 indicators, combine them in any way you want and show us - with SF code - how your approach is better.
|
yepher
359 posts
msg #37640
- Ignore yepher |
8/26/2005 12:03:47 AM
cocky_pusher,
The problem I have with Genetic Programming or Neural Nets for that matter is it depends on a set of pre defined filters.
It seems to me it would be much more interesting to use a Hidden Markov Model to try and adapt to the changes of the stock market and not worry about what the actual underlying true cause is for price changes. I originally started thinking about HMM to try and discover new "chart patterns" that are not regularly noticed but have high significance. (market intelligence recogonition instead of Speech Recognition :)
I think the real work for GM, NN, or HMM is not performing the calculations but in aggressively pruning the trees so you can process large numbers of data points/nodes.
Here was an interesting article I read some time ago about using HMM for such an application:
http://www.cs.sfu.ca/~anoop/students/rzhang/rzhang_msc_thesis.pdf
Obviously this was a simplistic implementation but I think it is in the right direction.
With all that said I strongly feel that simple is generally better.
My $.02
Yepher
|
Koronbock
201 posts
msg #37644
- Ignore Koronbock |
8/26/2005 3:34:57 AM
Yepher,
great post. I am alos a great fan of "simple solutions". I have never seen a trader making money with these "super advanced", super-curve-fitted models. They fall apart when applied to real markets.
|
cocky_pusher
12 posts
msg #37656
- Ignore cocky_pusher |
8/26/2005 2:06:30 PM
Thanx Yepher for your comments, I´ll check them out. To Koronbock, mrbobolina and the others that did find the page somehow convoluted. There is now a FGP in a nutshell page to help people out. Hope it is not as convoluted as the rest of the page for that matter. Just remember that I am not imposing any trading system, neither selling one and much less criticizing those who don’t like it. Just sharing utter financial algorithms information.
Yepher talking about Markov, you know, a teacher I had wouldn’t stop telling me about him, but you know I kind of prefer the empirical knowledge found in GP than theoretical melodrama. Never disregarding Markov’s theories. There are so many things one could use to create tactics the subject is just overwhelming.
|
yepher
359 posts
msg #37970
- Ignore yepher |
9/15/2005 1:12:27 AM
cocky_pusher,
I have been researching your web page a little more deeply and have some incoherent babbling on your findings.
You state:
"volume indicators did not even show up.....it means that if a genetic program did not find any patterns linking them to price prediction......"
I expect that "GP tactics were unable to link them properly" is the correct answer. Here is why I say that.
From what I understand of your analysis you were only trying to predict price one day in advance. I personally believe that volume provides much more predicative power than price does. My theory would be that the volume is a better indicator when trying to detect longer term price action. Where for a very short term prediction like one day the odds are in your favor that price will continue in its current direction.
It is interesting that GP tending to get better results with MACD, MA, and Stochs. I wonder if there is any research that shows smoothed data performs better in general with GP compared to chaos data.
Have you tried a “Tapped Delay Line Neural Net” or “Back Propagation”?
Thanks for the great report,
-- Yepher
|
yepher
359 posts
msg #38205
- Ignore yepher |
9/26/2005 6:58:52 PM
cocky_pusher asked if I would share this link to his forum
http://www.tau-zweig.com/forums/index.php
From the welcome message:
Welcome to tau Zweig’s investing strategies forums. These forums are here for your personal use, free of charge and uncensored. It is a place designed specially for the private investor community. Although, many private investors have forums at their trading platforms site, this is a place for everyone to get together and discuss. The subjects; Stock picking, computer modeling, filter debugging, news, etc…
....
|