Support / Knowledge Emporium / Neural Networks: Understand to Win
Neural Networks: Understand to Win
The Way Things Really Are
Most people have never heard of neural networks and, probably, don't
need to know what they are because of what these people do for a living.
It comes as no stunning surprise if a mail man or a bar waitress does
not possess a profound knowledge of the issue. But what's really surprising
it's the fact that a huge number of those who could benefit
richly from the neural
network technology have either never even heard of it,
take it for a lofty scientific idea or think of it as of a slick marketing
gimmick. There are also those who pin all their hopes on neural networks,
lionizing the nets after some positive experience with them and regarding
them as a silver-bullet solution to any kind of problem.
All the above attitudes are quite understandable, for all of us get bombarded with TV commercials, pop-up ads, spam messages and door-to-door offerings -you name it- that make you feel like you live for the sole purpose of being sold stuff to. Wherever we are and whatever we are doing - watching football at home or roaming African velds through the Web, there's always someone trying to pawn something off on us. However, the most important is to always be able to spot the real McCoy in this avalanche of debris. And high technologies are no exception.
Of course, neural networks are no quick-fix that will allow you to strike it rich by clicking a button or two, neither are they a new marketing stunt to make you shell out more money. And of course, they are not an uncorroborated theory or fruit of work of some crackpot scientists. Neural networks are a new, unique method of technical analysis, intended for those who take a thinking approach to their business and are willing to contribute some time and effort to make this method work for them. When applied correctly, neural networks can bring a profit on a regular basis. Well, after all, the latter is time-proven.
Why Neural Networks?
You may say, there are all kinds of things invented these days. Why
should one pay that much attention to this specific method? The answer
is very simple if you look at the current situation on the market.
There is hardly anything that can give you some edge though - the
market teems with top-notch professionals, conversant with every possible
widely adopted method. And all of them are after that buck or pound
you want too. Moreover, the traditional trend-following and technical
analysis systems yield mediocre results and are geared toward dabblers
and beginners who, lets face it, hardly ever make anything these days.
No other currently used method of technical analysis can detect
those hidden subtle non-linear interdependencies and patterns that
neural networks can detect. For a serious thinking trader,
neural networks are a next-generation tool, offering much better possibilities
than any of the tools the crowd tinkers with.
The Most Powerful Technical Analysis Tool
A major misconception is that many traders mistake neural networks for a forecasting tool that can offer advice on how to act in a particular market situation. Neural networks do not make any forecasts, they analyze price data and uncover opportunities. But with them you can make a trade decision based on thoroughly analyzed data, which all those who use the traditional technical analysis methods do not merely have at their disposal.
Which of the Nets Are Better?
Just like any kind of great product or technology, neural networks have started attracting all those who were looking for a budding market. Torrents of ads about next-generation software flooded the market; ads celebrating the most powerful of all the NN algorithms ever created. Even in those rare cases when advertising claims resemble the truth, a 10% increase in efficiency will be the most you will ever get; and still there will be some data sets and task classes for which the previously used algorithms remain superior. Remember this: it's not the algorithm that does the trick. Well-prepared input information on the targeted indicator is the most important component of your success with neural networks.
Fast Convergence: Good or Bad?
Many of those who have already appreciated the true value of neural networks mistakenly believe that the faster their net gives results to them, the better net it is.
These people are in delusion. A good network is more of a compromise between the velocity at which it trains and the quality of the results it produces.
How Do Traders Normally Use their Nets?
Alas, improperly, most of the time. Why? Because they place too much trust in the software they use all without having been provided with proper instructions on how to use it correctly.
To use a neural network the right way and, thus, gainfully, a trader ought to pay due attention to all the stages of the network preparation cycle. It is the trader and not his net who is responsible for inventing an idea, formalizing this idea, testing and improving it, and, finally, choosing the right moment when to dispose of it when it's no longer useful. Let us enumerate and consider in more detail all the stages of this crucial process:
1. Finding and Formalizing a Trading Idea
Many traders make the mistake of following the simplest path - they rely heavily on and use the approach for which their software provides the most user-friendly and automated functionality. This simplest approach is forecasting a price a few bars ahead and basing your trading system on this forecast. Other traders forecast price change or percentage of the price change. This approach seldom yields better results than forecasting the price directly. Both the simplistic approaches fail to uncover and gainfully exploit most of the important longer-time interdependencies, due to which the model quickly becomes obsolete as the global driving forces change.
A trader should fully understand that his neural network is not intended for inventing winning trading ideas and concepts. It is intended for providing you with as trustworthy and precise information as possible on how good is a trading idea or concept invented by you.
Therefore, you should come up with an original trading idea and clearly define the purpose of this idea and what you actually expect to achieve by employing it. This stage is the most important in the network preparation cycle.
2. Improving the Parameters of Your Model
Next, you should try to improve the overall model quality by modifying the data set used and adjusting the different the parameters.
3. Disposing of the Model When it Becomes Obsolete
Every neural network-based model has a life span and cannot be used eternally. The longevity of a model's life span depends on the market situation and on how long the market interdependencies reflected in it remain topical. However, sooner or later any model becomes obsolete.
What should one do when that happens? You can either re-train the model using completely new data (i.e. replace all the data that has been used), add some new data to the existing data set and train the model again, or simply never use this model again.
So what is the most optimal approach one should take with neural networks?
A successful trader will focus and spend quite a bit of time on selecting the governing input items and adjusting their parameters. He will spend from (at least) several weeks, and, sometimes, up to several months on deploying the network. A successful trader will be adjusting his net to the changing conditions throughout its life span.
How Can One's Neural Net Be Improved?
The sad fact is that most of the time, most traders cannot possibly improve a neural network. But not all the news is bad - in fact, you don't really need to, simply because it isn't your neural network that needs to be improved for you to achieve better results. What really needs to be improved to that end is the data sets you use. From our extensive experience with neural networks, we can say that the following methods are the most efficient in improving neural network data sets:
1. Simplification of the modeling object
Hitting the Bull's Eye: How to Achieve the Best Results?
Given the enormous multitude and diversity of the factors that influence future prices and their combinations, it is a lot more prudent to evaluate the current situation and the impact it may have on the future price than try to guess precisely what the price will be say, in two days. As an evaluation tool, a neural network can beat any of the other commonly used methods and if you make the targeted output simple enough your chances for success will be a lot better. For example, instead of forecasting future prices directly, you can forecast whether the trend will continue or whether the price will change direction.
2. Expert selection of the input information
While selecting input information for your network, you should bear in mind that the network should be fed only the cream-of-the-crop data on the main driving forces. The network should not be overloaded with information. On the other hand, all the required important factors should be reflected in your dataset.
3. Using committees of neural networks
Each neural network can cover a relatively small aspect of the market. Another powerful asset of neural networks is that you can use as many as appropriate. Each of these multiple nets can be responsible for some specific aspect of the market, thus giving you a major advantage across the board. We recommend that you keep the number of the nets used by you simultaneously within the range from 5 to 10.
4. Combining neural networks with classical methods
Neural neworks should be used together with one of the classical approaches. This will allow you to better leverage the results achieved in accordance with your trading preferences.
To speak the truth, in any event, provided the technology is applied correctly, you will gain more than you will lose, but you will be a real success with neural nets only when you stop looking for the best and, naturally, non-existing net. The key to your success with neural networks lies in having a strong idea about how to develop a committee of neural networks, whereby each of those would model a simplified object, and in combining this committee with classical filters and money management rules in sync with your risk preferences and trading style.