The human brain is one of the most complex objects in the known universe. It is not its superior processing speed and storage space that make it extraordinary, but its ability to learn and adapt. Recently, computer scientists have focused their efforts on writing software that allows computers to mimic this learning ability. Such efforts are accomplished through what is now known as neural networking. This technology has several applications, particularly in sales forecasting and trading. This article looks at how neural networks work and how they can be applied to trading.
Neural Network Basics
Neural networks are essentially a collection of interconnected neurons, each containing several inputs and outputs. These inputs vary in terms of their weight (importance) and frequency. Meanwhile, outputs are a function of net inputs. The image below of biological neurons demonstrates the interaction of neurons:
|Image from ANNEvolve.Sourceforge.Net|
Here we see the blue neuron sending an impulse to the yellow neuron. The yellow neuron may be receiving other inputs (varying in strength) from other neurons, yet it sends out only one signal (a function of all the inputs).
Now, here's how we can apply this input-output process to a computer:
The diagram above shows a series of inputs with varying strength being entered into a function that produces an output. The building of artificial neural networks gets far more complicated from here; it involves neural models and mathematics that are beyond the scope of this article. However, if you'd like to learn more about how they function, click here.
Although not nearly as complex as their biological counterparts, these components will eventually mimic the way our brain works, in order to make quicker and more accurate decisions.
Trading with Neural Networks
So, what makes these networks so special when applied to trading? Well, many factors go into making a trade: fundamental analysis, technical analysis, market sentiment, economic factors and even (arguably) randomness itself. Making sense of all of this can become a problem. Many trading applications are capable of considering one or two of these factors, but none can take all of them into account. Neural networks are used to fill that void.
What Do Neural Networks Do?
Neural networks used in trading vary greatly. They range from those that rely purely on genetic concepts to those that involve complex neural network models.
The neural models that use pure genetic programming utilize historical data (inputs) and randomly generated equations (functions) to create effective buy/sell rules. The process starts with the creation of an "individual" by running the inputs through a given function. Then random individuals (of which the better performers are given preference) are taken to create a second-generation hybrid. This process continues through several generations, each perfecting the equation. The result, in theory, is a perfect equation that will be able to generate profitable buy/sell signals.
Fortunately, much of this can be accomplished via an easy-to-use graphic-user-interface (GUI) program. Here is an example from Merchant of Venice, which is a free project on equity analysis:
Here you can simply input the number of generations the program is to run, the number of individuals in each generation, the number of random individuals that should be selected from each generation to create the hybrid, and so forth.
The other types of neural trading applications let you create actual neural networks as opposed to simply using the genetic concepts. One such application is Joone, another freely available program. They offer plug-ins that get input from sources such as Yahoo! Finance, and allow you to create your own neural networks. Here is what this application looks like:
Here we can see the input plug-in from Yahoo! Finance, the function being applied and a chart showing correlations being printed out. Simplified versions of this type of application are also available commercially.
Who Uses Neural Trading?
There are various financial areas in which financial professionals use neural-based applications: in various bond trading to determine the probability that a given company will default on its debt, in sales forecasting to predict future sales based on a series of inputs and in equity trading to predict future price movements. Many professionals and institutions are testing and even using this technology to help predict events and profit.
Why Doesn't Everyone Use It?
The application of neural networks to trading is relatively new. As such, it is neither perfected nor proven. The genetic programming example above yields large equations that become impractical or too biased to the past. Meanwhile, regardless of their GUI environment, programs like the one above often involve complex mathematics and NN models. Factors like these can limit the usefulness of neural networks.
Also, many critics say that the idea of software being able to "learn" the markets is flawed. If humans can't predict markets with certainty, how is it possible to create software that can accomplish something we don't even fully understand?
As you might guess, there are several commercial alternatives available. However, approach with caution, as many of these signal services and applications can be misleading. After all, if somebody developed a system that made guaranteed profits without any work on the part of the trader, why would that person sell it to the public? Be sure to do your research before purchasing a system that makes guarantees. Or, if you are ambitious, try creating one for yourself using one of the tools mentioned above.
The Bottom Line
Neural networks operate by taking a series of weighted inputs and putting them through a function to provide an output. These outputs are in turn applied to subsequent generations to effectively "learn" how to predict events more accurately. Many companies offer applications that allow you to create neural networks, and even more companies that sell you neural networks. The theory is slowly being improved and even adopted by some in the trading community. Time will tell the extent to which neural networks will change markets in the future.