Forex Software und Trading Plattformen
4 stars based on
This is a short primer on applying Artificial Intelligence AI techniques to your trading activities. AI can bring a fresh approach and a different perspective to anyone willing to invest the time and effort to learn how to combine some simple programming trend-tradercc erfahrungen and common trading knowledge.
First off, I will fully admit that while I am not an expert in either trading or artificial intelligence, I know trend-tradercc erfahrungen bit about trend-tradercc erfahrungen, which is probably just enough to be dangerous. Furthermore, I will make the claim that being a mediocre programmer and a semi-skilled trader puts you ahead of either world-class programmers that trend-tradercc erfahrungen nothing about trading, or top-notch traders that do not understand the basic concepts, advantages, and limitations of AI.
Think about your current skills and where the biggest bang for invested time lies. If you buy commercial AI trading software to save some effort, expect that just throwing some data and a few preselected indicators at it will not be very productive.
AI in a trading sense is a computer program that has the ability to change or adapt the way it processes data, over time or as new information becomes available, or even as the program is run multiple times.
This capability differentiates it from programs like Excel. Running an AI program many times, over and over, typically improves the results as the program learns from its errors, whereas running Excel over and over gives you the same result.
AI programs synthesize an appropriate output for which there is no existing data, but use similar historical situations to decide what is most likely to occur. This primer will review four types of AI: Judging solely by the number of books and papers on the topic, NN are the most frequently types of AI applied to trading and is the most detailed review here; feel free to skip directly to the section on NN if you just want to read the most popular topic.
And since developing, training, and using AI is a heavily time intensive trend-tradercc erfahrungen, few people using AI will tell you much about their successes or failures.
Most trend-tradercc erfahrungen the papers and books come from academic sources, not from traders. Whatever traders know will likely be highly proprietary. Fortunately, searching on any term in the AI field will yield a wealth of links and far more information than you could ever read. Expert systems are created by specially trained programmers that collect and categorize the knowledge, opinions, and experience of multiple people that have first-hand experience in a particular field.
Expert systems have found wide application in such diverse areas as medical diagnosis, automotive repair, and manufacturing optimization.
An expert system, whether used in trading or elsewhere, will be in essence the same as having the most knowledgeable senior person locked trend-tradercc erfahrungen in your computer, trend-tradercc erfahrungen they are as valuable as any other Intellectual Property. Applying expert systems to trading, for example, if you wanted to build an expert system to trade emerging market bonds, you would interview the best EM bond traders available, asking them what particular events, data, trend-tradercc erfahrungen and triggers that they use to trade on and this trend-tradercc erfahrungen only be done entirely in-house.
Of course the expert must be able to accurately describe the why and how of the trades they do, or the system will trend-tradercc erfahrungen. Expert systems can be used in combination with other AI techniques such as Neural Networks. Neural Networks NN, also called Artificial Neural Networks, ANN mimic in some primitive fashion the way humans use axons, neurons, and dendrites to process information and make decisions.
Photograph of a spinal cord neuron from Anatomy and Physiology by Kenneth Saladin. Signals received by the dendrites are processed bythe neurosoma and sent out along the axon to other cells. In a real neuron, electrochemical processes trigger a reaction from dendrites to axons. The human mind is reported to have about 86 billion neurons. In an artificial neuron, input data one or more signals is operated on by common linear mathematical functions such as addition and multiplication; however in the neuron there is a complex transfer function that can perform non-linear operations on the signal.
As more and more neural elements are paralleled or cascaded in ever more complex arrangements, the computational processing power increases exponentially. A fairly simple NN can perform incredibly complex operations that are beyond the capacity of most users to understand. A NN is composed of as trend-tradercc erfahrungen as one or two to as many as hundreds of artificial neurons, interconnected by various paths which have weightings that adjust the strength of the signal applied to each input.
In a NN computer program the neurons and pathways are represented by signal levels, which are themselves binary information in the computer running the AI software. A drawing of an artificial neuron is shown below. Multiple identical neurons will be interconnected to build the complete neural network. Each neuron is composed of two sections — a summer that adds the input signals into a single signal, and an activation function that converts the summed signal into a signal transformed by a specific transfer curve.
Typical transfer curves trend-tradercc erfahrungen shown in the drawing below, with the input signal on the X-axis, and the resulting output signal on the Y-axis, for three possible functions. The sum of the input signals is transformed to the output signal via one of the three transfer functions. The output is limited to a range of 0 to 1, while the input range is unlimited.
The next drawing is of a simple NN. Note there are three layers. The input layer is merely a series of connection points to distribute input signals to the neurons in the hidden layer. The hidden layer does most of the processing. In this network every possible interconnection has been made, although this is not always the case.
Elements 1 through 4 are basically tie points, Elements 5 through 9 are single neurons as shown above. During the development process, data sets and goals are presented to the NN in a process called learning. During learning the various pathways, weighing values, and number of hidden layers and neurons will be varied in a specific manner and an overall trend-tradercc erfahrungen value calculated.
The error value represents trend-tradercc erfahrungen far away the typical output is from the goal. Choosing the best data and appropriate goals is a major part of building a successful NN. NN excel at pattern recognition and pattern optimization, although not necessarily at the same time. For example, NN can be taught to trend-tradercc erfahrungen license plates in photos, and trend-tradercc erfahrungen the image to make trend-tradercc erfahrungen license numbers indistinct.
This is a form of pattern recognition — start with many different photographs of plates, digitize them in some manner so the NN can work with them, and present them to the NN. Training with additional sample photos is continued until the NN can recognize any new photo. This is called generalization. Conversely, if shown a rectangular plate of the trend-tradercc erfahrungen size with numbers and letters printed on it, for example a street sign, the Trend-tradercc erfahrungen will recognize it is not a trend-tradercc erfahrungen plate.
Another potential application is recognizing human faces, for photography or security. NN have often been applied to reduce credit card fraud and identify bankruptcy candidates. Neural networks seem to have had ups and downs in popularity. Major flaws have been found several times, people lose trend-tradercc erfahrungen, and then someone discovers trend-tradercc erfahrungen solution to the problem.
Two major breakthroughs in NN technology were non-linear capabilities, and optimization routines. Both breakthroughs are critical for using NNs in trading trend-tradercc erfahrungen. Many problems require non-linear capabilities for a solution.
Here is a simple NN trading imaginary relationships between gold, bonds, and stocks. The NN categorizes points on graphs of an intermarket strategy that looks at past gold and bond prices to decide whether to buy stocks. A linear network will separate the buy and sell signals into two regions trend-tradercc erfahrungen to a simple straight line and perform poorly.
A more trend-tradercc erfahrungen non-linear network, such as the NN shown, will separate the buy and sell signals according to a much more complicated formula. Of course, the intermarket relationship is not static and that must be allowed for when using a system like this for trading. A simple Neural Network using an intermarket strategy to buy or sell stocks. Neural networks can generate far more complex relationships with many more variables. Real world NN are generally far more complex than our simple network above.
Trend-tradercc erfahrungen can have many dozens or hundreds of inputs, and the NN can find extremely subtle relationships among these inputs that elude even astute market observers.
For example, you could use historical interest rates for many different bonds, their trend-tradercc erfahrungen and 60 day changes, and how far away the current prices are from several moving averages. You might want to use a time trend-tradercc erfahrungen extending back before interest rates started a secular bear market in Daily, weekly, or monthly data is then paired with a goal buy, sell, hold.
If there is a pattern hidden in all of the noise, eventually the NN will find it. Trend-tradercc erfahrungen NN can also find every possible spurious or temporary correlation and try to trade on them; experience and understanding of both markets and NN will help avoid these types of pitfalls.
Ants individually have but a few thousand brain cells and are quite dumb, but many ants together form complex societies; as a group they can hunt for food, establish complex homes, and wage war — the colony is collectively highly intelligent. Ant colony optimization is one category of what is called swarm intelligence. Swarm intelligence is exhibited by ants, termites, trend-tradercc erfahrungen and wasps, birds, and fish, but we will use ants in our study.
Ants secrete a chemical called a pheromone that they are extremely trend-tradercc erfahrungen to, as low as a few molecules. When ants from a colony bump trend-tradercc erfahrungen each other they trade pheromones. As ants repeatedly travel from trend-tradercc erfahrungen food source to the nest and back again, the trail of pheromones becomes reinforced, and more ants follow it.
One key to ant behavior is how fast the chemical signal weakens, a depleted food source will not attract foraging ants. A second key is not all ants will follow the dominant chemical trail; a few ants will always buck the system and try new paths.
As the main food source is consumed, the weaker pathways will begin to be used more frequently. How can this be applied trend-tradercc erfahrungen real problems? One area this type of AI has been applied to is traveling salesman problems, which attempt to optimize processes like mail delivery and garbage pickup.
Simulated ants will roam the pathways, and the fastest paths will display the strongest pheromone signals. Trend-tradercc erfahrungen addition to travelling salesman type problems, ACO has been applied to other routing and scheduling situations. ACO may be useful for optimizing the dates of periodic investments such as monthly trend-tradercc erfahrungen to a k, dollar cost averaging, or share buybacks.
SVM can categorize data as bullish or bearish, for example. SVM software is similar to NN and learning to use trend-tradercc erfahrungen type will be helpful in migrating to the other. Some of these software packages are specifically designed as trading systems. There are a few free packages that have been released that may be usable, but in general they are probably more trouble than the savings would justify. You trend-tradercc erfahrungen need data to train and evaluate your AI system.
There are plenty of free chart services but actual usable data is surprising hard to find. Usable data means in a trend-tradercc erfahrungen or similar list. OHLC data for most stocks and indices is available from Yahoo and other providers for nothing, but it is notoriously error-ridden and prone to survivorship trend-tradercc erfahrungen. Intraday data for stocks and indices is pretty much only available through subscriptions.
Plan to spend much more time finding and fixing errors than you will spend in downloading, especially if you rely on free data.