How to predict?

As to how the forecast, there is a pronounced polarization of approaches: either the methods catastrophically simple or so complex and resource intensive, their practical use is difficult for the trader. The former are the usual methods of extrapolation, and the second - the model that are based on neural networks or fuzzy logic. The two large areas on their sebe are many subtle nuances and settings set to adequately address the problem only by specialists. Thus, the important role played by the method of formation of the training set. There is a problem with the choice of the ideal moment when you have to stop learning. Feature is a self-adaptive systems, or the ability to adjust its internal parameters for the projected dynamics of the series. Education can be <without a teacher "and" the teacher>: in the first case, the change of the model parameters is in accordance with the internal algorithms used in the model, and in the second case, you need a clear indication of what change is better or worse.

Often as <opinions teachers> acts value of the prediction error, which is called the objective function and the purpose of education - to adjust the settings so that it would be minimal. Data set on which the minimization, called training or learning set. With this method of learning there is one very serious problem - overfitting. This phenomenon is associated with a random selection of the training set. First, the first steps of learning, the model begins to capture the desired dependence, which leads to a decrease in error - the target function. However, with further training, aiming to reduce the error, the parameters adapt to the features of the observed training set. This model is not law describes the dynamics of the series, and it features a specific subset selected as the training set. Naturally, with a decrease in the accuracy of the forecast of the real (outside the training set).

The proposed system does not provide an answer to the questions about the number of channels for learning, the amount of data required for each channel, as well as the principle of predictability of a price series. Whether, in the context of predictability associated with the specific action or not?

The effectiveness of prediction of complex systems is determined by the level of all the problems set, and this, in turn, is determined by the quality of the user.

The last step

However, the most difficult thing is finishing the process of analysis and forecast to take concrete steps. To justify the decision, the necessary statistical information on with what degree of probability satisfied found law or legal system. No matter how you enter catchy acronyms like money management, the basis of any such newfangled <Paradigm> for practical success should be <sewn> reliable theory, not a dubious myth about miracle leading indicator of the market. And every time - new.

In the following we will discuss the problems of the predictability of the market, the means of analysis of the market situation and the support of management decisions when playing the stock market, the modern concepts of money management, risk and try to dispel the veil <mystery> above the theoretical foundations of these newfangled paradigms.