All traders are doing it

How to improve the effectiveness of various technologies used by the trader practitioner? Vice President of Business Development begins a series of articles designed to help understand this delicate matter.
Control theory What strategy speculation nowhere, all traders, one way or another, decide to very similar problem. If you try to classify them, it turns out that the main types of problems is not so much - all three.
1. Choosing stocks that are best suited to your game strategy. 2. Forecast future price movement in the interesting period. 3. Issuing specific orders.
These tasks correspond to the three main stages, well known in control theory: information collection and analysis, forecast of the situation, management decisions for correction in case of deviation from the projected dynamics. At this similarity could not draw the attention of the producers of analytical software. For example, to select the purchased shares have gained widespread popularity stock skrinery, or common filters - from the database they choose only to have the specified parameters shares. For the price forecast used by a wide set of tools. These include the traditional methods of extrapolation, and rather sophisticated (and expensive!) Solutions based on neural network algorithms.
For order placement on the market, there are many trading systems. Most of them allow you to program the rules of decision for the automatic placement of orders. But the rules themselves should ask the user! Work to draw up such rules will automatically take over the mechanical trading system, but most of them turn out to be inadequate.
What emerges is a picture? Producers of software products and online services are based on management theory, but also an impressive effect that accompanies its use in traditional areas, with investment management is clearly not observed. What's the matter?
Simplicity is worse than theft A deeper analysis of the proposed instruments shows that in most cases based on analytical programs laid unduly simple algorithms for the solution. This simplicity does not match the sophistication of the object - the market. Judge for yourself. What, for example, represent a drain-skrinery? This is just one query to the database with the correct . For market participants, who know exactly what they want, such a tool, of course, will be useful.
But far more relevant, in my opinion, would be a tool to determine the allowable ranges of values for each indicator, which characterizes the action, for specific purposes. For example, the user indicates that he is interested in weekly growth companies. In what range should be based on the value of its interest rates, the probability of such growth was the maximum? Now on the market there is no tool to answer this question.
Or this aspect. To set the drain-skrinera, you need to know what at the moment the value of a particular group of companies, for example, drug manufacturers, in general there. How many companies fall into a particular range of values of a particular indicator? Unfortunately, and this issue is resolved satisfactorily. Recently, attempts to build such a And what do you do when you need to track multiple indicators?
How to predict? With regard to forecasting methods, there is observed a marked polarization of approaches: either catastrophically simple methods or are so complex and demanding that their practical use for the trader is difficult. The former includes the usual methods of extrapolation, and to the second - models that are based on neural networks or fuzzy logic. These two large areas on their sebe have many subtle nuances and settings are set to the task adequately only by a narrow specialists. Thus, the important role played by the method of forming a training set. There is a problem with the ideal moment, when you want to stop learning. A special feature of adaptive systems is self-taught, or the ability to adjust their internal options under the dynamics of the projected series. Training can be provided and : In the first case, change the parameters of the model is in accordance with the internal algorithms used in the model, while in the latter case, you need a clear indication of what change is better or worse.
Often, as a The set of data, which is the minimization, called a training or learning set. With this method of teaching there is one very serious problem - overfitting. This phenomenon is associated with a random selection of the training set. First, when the first steps of learning, the model begins to capture the desired relationship, which leads to a decrease in error - the target function. However, with further training in an effort to reduce the error, the parameters adjusted to the features of the observed training set. This model has no law describes the dynamics of values of the series, and features specific subsets selected as a training set. Naturally, with a decrease in the accuracy of real prediction (outside training set).
The proposed system does not provide an answer and questions about the number of channels used for training, the amount of data required for each channel, as well as the principle of predictability of a price series. Is it in the context of predictability, to communicate with a particular action or not?
The efficiency of complex systems prediction is determined by the level of all problems settings, and this, in turn, is determined by the skilled user.
Last step However, the most difficult point is the debugging process of analysis and forecasting to take specific steps. To justify the decisions necessary statistical information about how likely performed found a rule or system of rules. No matter how much input catchy acronyms, such as money management, the basis for any such newfangled practical success must be reliable theory, not a dubious myth of the miracle of leading indicators of the market. And every time - new.
In subsequent articles we will explain what issues the predictability of the market, the means of analyzing the market situation and support management decisions when playing the stock markets of the modern concepts of money management, risk and try to dispel the veil above the theoretical basis of these new-fangled paradigms.
Vladimir Cherkashenko