decipherlinda
133 posts
msg #71237
- Ignore decipherlinda |
2/6/2009 10:17:15 AM
Anyone tried using "black box" neural systems? How did it work?
|
Walid
130 posts
msg #71240
- Ignore Walid |
2/6/2009 12:10:43 PM
As a software engineer - with special interest in AI - I know one thing about automated trading systems. The best - and some times only - way to make money out of them is to sell them to others.
|
chetron
2,817 posts
msg #71241
- Ignore chetron |
2/6/2009 12:20:30 PM
= {)
|
guru_trader
485 posts
msg #71245
- Ignore guru_trader |
2/6/2009 2:36:08 PM
I've tried a variety of neural systems and some of them can make money but it's tough to tell the difference between them being lucky vs them really being able to forecast direction. Also, the time frame you use to train neurals may be critical. Do you train over the last 1 year, 2 years, last 1 month, last 6 months? I had one neural system that worked better if I trained it over the last 3 months on a rolling basis. The problem is when market tendencies change quickly neural systems often struggle to keep up.
Make sure you research how to improve and/or optimize neural networks properly before you start training them.
|
stocktrader
294 posts
msg #71252
- Ignore stocktrader |
2/6/2009 10:21:26 PM
I spent one month a couple of years ago analyzing this web site and crunching/tracking the data.
The result, flipping a coin was just as good.
http://www.stock100.com
|