I had a brain storming idea, and I wondered if anyone here had thought of training nerural networks to make buy/sell/stop etc. descisions based on momentum trading theory. Let the computer choose the advancers and let it determine where a stop-loss should be set etc... What do the experts think? I''m really thinking about having fun with it...

Its been done.

Ill try to dig up a link later when i have time.

My father and I toyed a lot with similar ideas. We used formulas to rate stocks, and tried to optimize them by simulating with the last 5-6 years of market data. The best formulas managed to do over 100% of gain / year (of course they were optimized for this particular data, but still...)

If that can be of some help, we found that being invested 100% of the time yields better results than having the "buy at the right time, sell at the right time" mentality. You don't "sell" stocks. You just replace them with better ones.

My father has tried this during the last 6 months, and it works well. It's really nice.

I don't know about neural networks, but there was a discussion of this on comp.ai.neural-nets not so long ago. There were interesting links. Make a google search.

Cédric

(Edit) Link: http://www.tradingsolutions.com

[edited by - cedricl on August 19, 2002 10:25:03 PM]

Yes, it''s been done. Search for "Recurrent Network" which
have been applied to stock market and other fun things.

---

Kami no Itte ga ore ni zettai naru!