After statistical Forex systems were introduced in my blog, Ive described the problems of timeframe selection and statistics gathering. Now its time to discuss the decision making problem of such trading systems.
When a completed strategy has enough statistical information and a sample from the current market situation it should have some methods of comparing the statistical information with the sample and make the decision regarding its further actions on the market. For the majority of the systems these decisions would be limited only to buy, sell, hold and close previous position actions, while more advanced systems may include position adjustment actions into their arsenal.
The most obvious way to make the decision for the statistical Forex system is to calculate the differences between the sample data and the data stored in the statistics and the lowest difference will point out the most probable recorded outcome. For example, if you recorded RSI indicator values and the current RSI reading is 75.2, while the lowest difference from your statistics is 0.1 and it suggests that the price goes down near that RSI level, then your system should probably generate a sell signal. This method looks simple, but its also flawed as the comparing multiple parameters of the two samples is impossible.
In general, quotes-derived parameters should be compared with some method similar to Euclidean distance (best distance, average distance, etc.) with possible weighing of the different parameters according to their importance. Meanwhile, the comparison of the time- and fact-based parameters should be rather strict — e.g. if you recorded some information specific for Fridays and its Monday today, then you should disregard this information.
Another noteworthy idea regarding decision making would also require a special statistics gathering method used in the system. Using self-organizing maps (or Kohonen maps) is a popular decision making method that is widely used in finance. Unfortunately, my own tests of the self-organizing maps within the statistical Forex systems (in a form of MetaTrader expert advisor) didnt bring any interesting results. There are many other ways of utilizing the self-organizing structures to store and compare the quote-derived statistical information, but their complexity doesnt look to be necessary in such systems.
Chart-to-chart comparison can be used if the statistics stored is a raw or normalized market data, which brings a lot of opportunities based on the graphical chart analysis and the difference calculations. Its also necessary to note that such comparison would require a lot more CPU power and time to complete. It would also produce a more long-term aimed result than the immediate decision that would be true for the next bar or candle.
In my opinion, its optimal strategy to store the statistics in three separate «containers», where statistics in the first container would correspond to the buy action, in the second — to the sell action and third — hold action. Finding the best Euclid distance for the current market sample among all three «containers» gives you a hint for your next action. In this case, its more important to collect the right data and to format it in a right way for further comparison.
