Application of Chart-Aware Neural Networks with Evolving Topology in Forex Trading

It is quite hard to find a Forex trader that has tried to automate his trading and has never heard of neural networks. Creating a trading robot that is capable of evolving and learning from the market itself is a very attractive idea. Artificial neural networks used in computation resemble the neural networks in human brain as they are built with multiple interconnected nodes and work as a single system. Their structure allows adaptive learning, which is great for analyzing such complex systems as financial markets.
Today, I present a free download of the article on an interesting case of neural networks’ application in trading, published by Gene I. Sher from the University of Central Florida. In his work, titled Evolving Chart Pattern Sensitive Neural Network Based Forex Trading Agents, the author presents an innovative use of TWEANN (Topology and Weight Evolving Artificial Neural Network), which is able to work with price charts rather than usual linear input to analyze the market, learn and carry out trading decisions. Unlike a normal neural network (NN), whose evolution is based only on connection weights between neurons, TWEANN’s evolution also changes the very connections inside the network. The author tests two hypotheses in his research:

The article will be very interesting to the developers of expert advisors, especially to those who have tried creating or using NNs (or any other self-learning systems) in trading. It is necessary to mention that the article will seem very confusing if you do not have some basic knowledge of neural networks. On the other hand, there is no difficult math involved, while the concepts that are not explained in there are easily googled about.
Sadly, there are two significant disadvantages in this article. First, it contains quite a few errors. It looks more like a draft rather than a more or less final version. When dealing with such a tough topic, additional obstacles in form of grammatical errors do not help at all. Second, Sher is using only one (and rather short too) period of EUR/USD chart history to train and test his NNs. The results would be more reliable and informative if he used several longer periods across different currency pairs. But despite these cons, I still highly recommend it to any trader who seeks to find out more about neural networks:

If you have any questions, comments or suggestions regarding this e-book or about adding more free books to EarnForex.com, please feel free to post them in the comments below.