TSPGo,
I first was exposed to neural network artificial intelligence in 1993 when I was working for the US Air Force as Electronics Engineer GS-0855-12.
Since then I spot the potential to use it to predict weather, market, etc...
I am not an IT persay... nor mathematician... I am an engineer from accademics... You are absolutely right the neural network learns from past experience patterns recognition etc... but not necesary a lot of data is need it nor the universe to predict each time. sample teories can be applied as well. once it learn you do not need to run every day all the data (universe) you just need couple days... and update... and I did run the full data back to june 2003 again... during the testing... the key is on your neurons... I have two input neurons and three output neurons... the hidden neurons allowed me to develop the leaning models... that's where the real secret is... if I run back all the data available (universe) each day the pronostics will not be processed till the following day and it gave me the same resuls as if I only runs for couple days (sample) ahead... enough time to made the
interfund transfer... you may find more info in my undevelop scratched website
www.tspneural.com predict the market with artificial intelligence is simple, piece a cake... after the fact of course...
it gives you the 8 patterns similar to the ones EBB is using, I got great respect for Ebb perhaps he discovered them without artificial intelligence. I do not know.. Ebb never mentioned artificial intelligence... But He did it ! and I admire Him very much. Bottom line the neural network artificial intelligence I am using performed as follow...
It Predicts Up Trends 30% of the total observations
It Predicts Down Trends 28% of the total observations
It Predicts Side Trends 30% of the total observations
Total Correct Predictions 88% of the total observations
It is not perfect so it is Wrong 12% of the total observations
It is as good as or close as good any of the systems in this website.
As I keep repeating myself... let be the time be the judge...
If it works it works... if it does not I will be square one again...
meanwhile I will be sharing my interfund transfers for those willing
to consider the pronostics for their investing decisions...
Hope my pronostics help you in enjoying the ride and profit from it...
It seems that you have a software that has nothing to do with artificial neural networks. The basic concept behind neural network is that it learns from past experience ( lots of data). The more data it has the more accurate the system becomes. In a true "artificial neural networks" system three days, three months or even thee years of market data the output would not be reliable. You must be dealing with something else and by mistake it has been called "artificial neural network". I have been posting two graphs for the S&P in my website generated by an "artificial neural network" software for a couple of years now. The projections of these graphs are product of the analysis of all the data available for the S&P. I would not advise anybody to follow these predictions blindly. Those predictions you see in my account talk are derived from neural network software but they have other components added. I run the last neuron of the system in the morning before I make the final recommendation. Sometimes that last neuron causes the prediction to be modified. The software analyzes all the data available since June 2003 and all the data available for the S&P. In my experience I have not seen all the neurons be 100% in agreement. That indicates that they are still learning. The best you can get out of three days or out of three months (60 points of data) would be an average. It seems you came across a good idea that back testing seems to promise something. That's is fine. However, it should not be call product of artificial neural networks analysis if you have you only three days or three months of data.
Good luck
http://www.tsptalk.com/mb/showpost.php?p=118894&postcount=500
