It depends on what kind of NN you used and how it was accurately developed. I compared associative self-building NN that I used in intraday trading model against real Excel system that was used for real trade. It contained about 25 indicators and filters. I used only implemented indicators in NN training and result was quite optimistic. Historical data was S&P500 with 1sec. ticks.
Of course it depends on such things. The models I used in these experiments I had already applied, quite successfully, to other domains such as pattern recognition in electrodiagnostic systems. But there are less complicated ways to become profitable. GS
I've read the articles on the site. The only thing i can say is: what a laugh ! Their neural network that can't even average 1 point a day in daytrading the ES? I can even do better without a neural network. And they want people to pay for it? Is this what they mean when they say that NN can improve results spectacular. If NN were so spectacular they would take over almost anything from budgetting, salesforecast, research on medicines, predicting the outcome of the horseraces, predicting the lotto....... I think NN that are sold are comparable to the sales of 3000$ tradingsystems. They often only benefit the seller of the software. I don't say NN have no potential. The main problem appears to be the human being that has to setup the NN.
My late father was an expert in artificial intelligence - Neural Nets, Genetic Programming, Rough sets. He started out witha NN system and eventually that broke down. Then he made a 100% mechanical system and eventually that stopped working. He then devised a discretionary system and that is still working. In other words: there are no easy shortcuts: you'll have to learn to "read" the markets. Maria
garbage in = garbage out if you know what to feed the NN then you do not need the NN, you can more accurate "visualise" it yourself how does a NN know when to include a spike due to news or not to include it? When does the NN take care of a holiday or luchtime? etc etc Waste of time
I trade a set of mechanical systems in a discretionary way. Fully mechanical systems can never achieve that kind of returns. It was much easier to feed my Tradestation with info than to feed the NN.
I feel this is perhaps underestimating the pattern recognition capabilities of a neural net not to mention the sheer computational advantage of dealing with very large amounts of information that a neural net might have. There are many tasks that can be done by humans but where it can be more efficient to be done by computer. This I'm sure is stating the obvious. How do you know when to include a spike etc? What information do you have available to make those decisions that a neural net does not? Is it feasible to provide the neural net with the same information that you have? e.g. www.mediasentiment.com, calendar of holidays, trading hours for relevant instruments. More simply, if you train the neural net with day of week, hour of day, volume etc. do you not think it is capable of "learning" about holidays, lunch times all by itself? Food for thought. MoMoney.
Well, the only thing i'm interested in is the results. Whether the NN uses very complicated, highly scientific and very advanced algorithms that only can be understood by a genius, or not, i don't care about. Even if nuclear power might be used and Martians would have developed the NN i don't care about. Some people are impressed by the description of the NN, i'm only impressed by the return it generates. And the published returns are poor.