Automated Forex Trading systems do have advantages over manual human trading. Automated trading system can monitor the Forex markets 24 hours a day; automated systems are completely disciplined to the set of system rules and never stray; automated trading systems are immune to greed and fear and emotion never influence their trading decisions; automated Forex systems always follow the money management rules defined by the user. However, it is apparently very ironic that these basic principles, that define the strengths of a system, are also many times its downfall. Forex robots can not 'analyze' the market price action like a human being. Therefore, Forex Robots Enter Every Trade that meets a defined set of conditions. Human Traders Most Often Do Not!
Prevailing sentiment contends that, out of all Forex traders, only a small percentage are successful long term. The referred figures vary depending on the source cited, but the percentages consistently average in the 5% to 8% range. In alignment with this figure, very few Forex robots survive the tests of live account Forex trading, with a mere 1% to 2% surviving more than a few months prior to their rule-sets becoming obsolese, and the losses begin piling up. The ideal solution is obvious. Combine the discipline and tireless availability of an automated Forex robot with the savvy and experience of a successful human trader.
It is in this vein that much of the groundbreaking research on algorithmic Forex trading lies. By utilizing machine learning to 'teach' an algorithm certain prevailing 'human' decisions that affect trade entry, existing systems for trading Forex automatically can be converted. Some research shows that training entry tactics with machine learning strategies (Genetic Programming and Neural Networks to name a few) do significantly improve the performance of systems on out-of-sample data. These conclusions lend some early credibility to the notion of Forex trading using machine learning.
The concept that we discussed here departs from this strategy in that we use the learning technologies to train sets of 'humanized' data as opposed to raw data prior to a condition. By utilizing these datasets, the learning becomes 'why did the human enter this trade?' vs. 'do the raw data support entering a trade right now?' When the learning begins to focus on more abstract data, the resulting systems tend to become more robust, or tend to work better in varying market conditions than those that simply attempt to identify winning Forex trades from raw indicator data. The concept is that basic indicator conditions trigger a trade Set-Up, for instance, a fast moving average crosses a slower moving average. The learning algorithm then works to filter these set-ups using the training it acquitted from human training datasets. The automated trading system says, "Based on what I've learned from my expert human teacher, does this set-up look like a good deal?" Instead of, "The computational result using all of the empirical data is greater than the defined variable, get in or out?"
In summary, applying machine learning strategies to teach 'human' tactics for automated Forex trading system design, can be much more effective in producing robust Forex systems than utilizing the technologies in an attempt to forecast market direction. In future articles I will expand on this method and provide information on applications and technologies available to employ these concepts.