Hi sp500, Do you use NN in your trading at all? For what purpose? Your comment also sounded like it can work under proper use. Thanks, KC
I have contacted Katz and McCormick for their CD a long time ago. They seems like friendly people and they trade themselves. A few months (or could be a year ago), they asked (no hype in the email) if anyone interested in a trading system based on genetic algorithms because it doesn't fit their style of trading. I never bothered with it thinking if it is very good, they would fit their style to it. Now it got me curious. KC
unfortunately i'm part of the 90% that will never be able to make it work. The only NN that works for me is my head. And the purchase was a bargain because i got it for free.
The sample size is too small for now... hopefully we can get to 30 votes. The poll shows that the success rate doesn't look as bad. I also wonder if the NN has improved over the years. What other areas of research shows promise besides genetic algorithms? I think there is something there: 4 25% I lost money and waste time: 7 43.75% I made money but not worth the risk: 1 6.25% I made money consistently: 4 25%
Indeed, ya wrote a lot about a lot of stuff. If you keep this up, you might actually come close to matching Jack' s renown by the time you reach your seventies. nononsense
A couple of years ago I built and tested several NN based models for stocks. I tried many variations regarding network topology, input data normalization, point of insertion of the network into the system and algorithms for weight updating. Since I was not satisfied with the ubiquitous back propagation rule for learning, I finally constructed a more "advanced" version where weights were updated via a quite sophisticated multiple crossover genetic algorithm. I used out-of-sample data, walk forward retraining etc etc. Then I gave up NN and only kept the GA as a very fast and effective form of curve fitting. While I don't feel so arrogant (I'm arrogant though) as to say mine is the last word on the subject, all I can say is that if NN and GA can make money, I wasn't able to find how. Not a single one of the systems I tested was able to net positive when tested on a non trivial amount of out-of-sample data. A much simpler pullback system of about 10 lines would have been much better, for a while. GS