"In trading aren't you really trying to find only two types of object, Buy, Sell. "
No. That is the decision one makes after the question, "Does the most recent price/volume pattern (or any of its derivatives) indicate the price might rise or fall."
Now that might not seem significant, but its not just semantics. What one asks a neural net to do is pattern recognition. Thats all it can do along with associative extension - which even the best commercial products only do rudimentally. There really isn't suh a thing as a hidden input - only associations (patterns) that are not obvious - and they are patterns that are reinforced when repetitive success is reinforced.
But, I understand what you are asking, and the best I can answer that is to refine what i do - which is not to train specifically for any one equity, but rather to screen for candidates with potential. I don't want to know if my nets can predict the close or % change or anything else statistically often. I want the larger picture. Are financial sectors moving, are tech stock heading down, are industrials going to sustain some climb afterthe SPX drops from net stock bombshells? Given that sort of view I can then look for those price change leaders (again - something the net can do well - not just how much did it climb once, but how often and with what cycle or frequency can it be possible to continue).
This probably isn't anywhere as good as focusing on one stock or sector, but then I use nets as one tool in the toolbox. neuroShell Trader has the capability to run buy sell strategies on 99 candlestick formations. I have tested lots of them. But I don't see tham as any more noncontinuous than occidental buyy sell signal methods - or more successful. But that could just be a shortcoming of the depths of my analysis.
We have to know some level of relationship to use nets, or standard TA or just plain chart trendlining. We have to know what we are looking for, because without that objective and some understanding of why we are looking or that relationship, there is nothing what will tell us to buy and sell - regardless if you use one input or one thousand.
The magic isn't in some black box - it is in our own understanding of the markets and individual stocks. Nets are a finishing tool, a compass, or perhaps a screen. But buy itself, back-propogation mathematics holds no secret.
lastshadow |
