Does anyone have any recommendations on neural net software? I've used a plethora of products, all of which were not directly related to trading. I understand the pros and cons of NN's-just curious on what others enjoy using for trading.
What do you want to accomplish with them? If you are looking to use them for price series forecasting I have to warn you that their applicability in the markets for that purpose was disproven many years ago. If you are looking for something along the lines of system development we have discussed this subject here extensively: http://www.elitetrader.com/vb/showthread.php?s=&postid=2968698#post2968698 http://www.elitetrader.com/vb/showthread.php?s=&threadid=239129&perpage=6&pagenumber=1
No looking to forecast an indicator. Also might try price forecasting with indicators I'm already successfully using - NN might find a more efficient way to trade them.
it is useful. if you are trading systematically. I am a discretionary trader, most time buy then sell. nn works like that. you need feed in some trade examples, to setup intial parameters or trading signals (they call training). then feed in your real market data, let it give out signal (match your examples), after many executions, those paramters will be refined, will better reflect statistic edge. NN is not complicated stuff.
Sounds so easy, but ... What's about overlearning, what's about using too many parameters, to big nn's? It's absolutely easy to train a nn but it's more than difficult to develop one that keeps the positiv results out of sample. I've worked a lot with nn, my first application - not in finance - was used already in 1992 for cost prediction in industry. But predicting marktes is something else. Recently I tried some newer methods like support vector machines and others from data mining. The results are a little better than with simple nn, but ... So, before you start investing lot of time look at some other threads
In my experience, using any type of soft computing technique is an indication that you don't have insight into the problem you are attempting to solve. Once you gain insight, there are typically computationally more efficient ways of addressing the initial goal than using NN.
Precicely. Well said. NNs and things like genetic programming and genetic algorithms are curve-fitting tools by definition. Very useful in many application in science, not very useful for developing trading systems because of that.
Nice first post. Another way of putting this... You need to have Actual Professional EXPERTISE in trading... Especially expertise in some specific niche of trading. Buying software and books is not a substitute for human expertise... Which takes roughly 10,000 hours to develop in any given field.