You can definetly add value by applying some NN techiniques to your system. Basically, from my experience no single Neuraral Net to my knowledge can step forward w/o significant error. However, where it proves invaluable is tweaking the system that you already have, i.e. you have particular system that you're evaluating, and NN algorythm can be very timesaving as far as tweaking the optimal historical entry/exits relative to time/ volume etc. But as I said, i haven't seen neural network that would say S&P up 10 points tomorrow.
Nope. Me neither. You have to learn to live with low r^2 values. But forecasting stochastic on the ES 3 minutes into the future is a little easier, and might be an edge.. I'm experimenting with using basic indicators as a 'bias', and then NN as the trigger..
I agree that traders would get more benefit out of tweaking an already existing trading system using neural networks.
NN's have an inherent lag in their analysis and decision making ability, even on the 1min or tick charts. In addition to this, other similar rule based systems work best from a good distance from the market and for position trading which allows for recoupment of their loses and slippage issues. To attempt to adapt this methodolgy for day trading requires both deep pockets, and steel nerves and a solid belief that one's earnings are derived from some other endeavour. Steel nerves...
Most NN are crap ! Why... 1) They lag cause calcs are based on historical data 2) Unless you can forecast a turning point with 80% success then your wasting your time (even www.vantagepoint.com cant do that !) 3) the curve fitting will result in your account getting kicked when the market changes its form. Stick to price patterns, sup/res, trendlines, different time frames. Good Luck