If I told you about a technical analysis tool that might take up to a year to fine tune, could be rife with built-in errors and would eventually become obsolete, would you be interested in using it? Me, neither. Lots of people are, however, including corporations.
I'm talking about neural networks.
Why even consider setting up neural networks, Dima Vonko asks in "What Can Neural Networks Do for You?" (Stocks & Commodities, V. 24:12, 60-61). Vonko insists that neural networks give traders an edge over those who don't use them.
Neural networks might best be described as a form of artificial intelligence. The simplest explanation is that you show the neural network a set of inputs with a desired output then train it to sort through new sets of data and recognize patterns that might not otherwise be apparent.
Proponents of neural networks speak of the architecture of the network and the training algorithms employed to teach the neural network. Fortunately, you don't have to have a Masters in Software Engineering from Carnegie Mellon, as my son-in-law does, or understand each detail of the architecture or algorithms to set up neural networks. Software, either purchased or freeware, does that for you via a graphic-user-interface (GUI) program. Traders considering using neural networks do need to understand their basic uses, strengths and shortcomings.
What can the neural networks be trained to do? Although Justin Keupper ("Neural Trading: Biological Keys to Profit") says in an article found through Investopedia that neural networks can be trained to "predict future price movements," others argue against expecting neural networks to predict prices. Both Vonko and Connie Brown ("Neural Networks with Learning Disabilities," Stocks & Commodities, V. 11:5 (207-214)) warn that neural networks are better at picking out patterns or predicting whether a trend, once in place, will continue than at predicting prices. "It's not price forecasting," Vonko warns in the Stocks & Commodities article and in one for Investopedia.
What neural networks can do is to thoroughly analyze all sorts of complex inputs and anticipate the next direction, revealing opportunities or probabilities that might not have been apparent to the trader. Elements of "fundamental analysis, technical analysis, markets sentiment, economic factors, and even (arguably) randomness" can all be input and analyzed by neural networks, Keupper claims. Brown's article offered examples of neural networks being trained to analyze ADX, CCI and RSI figures together with volatility bands on underlyings such as cotton or the Deutschemark per U.S. dollar. Various offsets and other parameters were tested.
Pattern recognition can be a strength of neural networks. Yet Brown warns in her article of the ways that neural networks can go wrong. Like the unlucky first grader who supposedly learns to read but is in fact only memorizing sight words rather than understanding phonetics, neural networks can memorize data rather than learn from it. During the training phase, data from a trending market might need to input out of chronological order, Brown warns, to keep the network learning rather than memorizing. Vonko cautions that it might take up to a year to set up or train your neural networks. Keupper warns that they "are neither perfected nor proven."
Several writers advise against buying into all the claims of the commercially available services and applications. If an application works so well that profits are guaranteed, Keupper questions, why would a company sell it? Vonko worries that recent claims of faster training of the networks may compromise the quality of the results. Technical traders already understand that faster signals may not equate to more reliable ones, and Vonko's warning suggests that may be true of neural networks, too. Boris Schlossberg, Senior Currency Strategist, FXCM, offers another caution in an article linked to Investopedia. These networks eventually become obsolete, he warns, and must be discarded.
Yet Vonko wrote in both articles about that edge that traders gain from using neural networks. Keupper offered links to freeware such as Merchant of Venice and Joone that allow for different approaches to setting up neural networks.
I was surprised that only one of the writers spoke of the one trait that I had thought might be most advantageous about neural networks. In the efforts to mimic the human brain, no effort has been made to mimic our emotional response to trading decisions, of course. Market pundits speak quite often about the advantages of removing emotion from trading decisions. Not true, Schlossberg says when writing about the benefits of keeping a trading diary. "In fact," he writes, "I have witnessed the results of hundreds of systems trade in real time and not one of them was profitable in the long term. Trading requires all of our emotional and analytical capabilities in order to produce success."
During the course of researching this article, the knowledge that there is a
freeware proves appealing. If I can find spare time from my writing duties, I
might even investigate it. If those other writers are right, though, don't
expect to hear back from me about my results for up to a year.