If you want to design trading systems NNs are a total waste of time. Consistency is the key word here. For each different market the neural topology is in principle different. NNs tend to perform in unpredictable and inconsistent manner. Data mining for rule-based statistically significant patterns of price action is the way to go. There is one program that I know of which combines data mining with entry/exit to find patterns from historical data that exceed certain limits in performance parameters values you set in advance. It says now on the website that is not sold any longer (website) There is another one I have found, TSL, but its price tag is prohibitive (website). This one is based on Genetic Programming.
From what I've seen, NN's need to be adjusted often. Just when you think you have a NN that works, it doesn't. A successful NN seems to be a short lived thing and retraining isn't the problem.
Sorry, I'm not your free research person. If you want someone to tell you "they are great" then will you then accept that? I followed a lot of people who tried to implement them for many years on sites, forums and other places. The general consensus was that it was difficult and not particularly useful. If you want a scientific study to prove it, then "good luck."
Just curious. Is there a difference between a computer that can think for itself (as in "artificial intelligence") and a computer that is statistically significant?
artificial intelligence is not really "thinking for itself". it is using a limited set of facts and limited set of rules to solve a constrained problem. It is why IBM's Deep Blue could defeat a chess champion a few years, but could not plug in its own power cord or make a martini. It is why a showering robot with shampoo in its eyes would poke holes in the shower stall trying to find the cap. A showering human would be able to feel around when the problem/circumstances changed. Basically, we have not yet created a system that could match a cricket for overall functionality and self mobility in the real world
This thread was an interesting read, I would recommend that you take the action to research nn's yourself and arrive at your own person conclusion and not rely on others. I personally think they are great given the proper circumstances in which to operate.
How about the open-source neural network packages? Or to do it right, do you need software specifically geared for analyzing stock prices?
There is an absolutely free tutorial on how to build your own NN for trading with free OS software here... http://intelligenttradingtech.blogspot.com/2010/01/systems.html Once you are familiar with how to build it, you can start to research public papers on the subject. Here are some researchers actively engaged in the area with public work: http://works.bepress.com/bruce_vanstone/ As with any other complicated mathematical processing, it all comes down to what your I/O variables are and how you are utilizing them. Hearsay is one way to gather information, another is to learn for yourself.
we pulled the plug on nn research when the need for data filtering became aparent. If this is no longer the case 'would love to know...