Approaches to forecasting stock

Among the factors that characterize the dynamics of the market and the impact on it, there's a fair amount of data is not a numerical nature, which should be known only to a certain degree of confidence. We can distinguish different types of uncertainties of which are important for the financial analysis are as follows: associated with ignorance or inaccurate knowledge of certain factors

or processes affecting the development of the situation, related to the mathematical incommensurability numerical estimates, variables that characterize the dynamics of the system, associated with the non-linearity and the presence of a system of several equilibrium states or attractors, associated with the lack or inadequacy of the conceptual apparatus and the impossibility of identifying the facts.

In order to understand what benefits gives further suggests that new methods of data analysis and forecasting, it is necessary to point out three fundamental problems associated with the creation of financial markets analysis systems and the development of predictive models.

First - this is the definition of necessary and sufficient parameters to assess the state of the market, as well as objective functions, ie deystviy.Formalizatsiya choice of performance criteria, ie modeling the behavior of a system consisting of heterogeneous components, it requires the use of a single metric for their description. The second problem - a problem of dimension. The desire to take into account in the model as much as possible indicators and evaluation criteria can lead to a virtually unworkable terms of computational complexity. In other words, the essence of the problem is reduced to a limitation on the speed and size of computer system depending on the amount of information processed per unit time.
The third problem arises from the existence of super-system feature. It is known that the interacting system form Supersystem higher level has its own (super-system) the properties of which has none of the constituents of systems. The problem is fundamental impossibility to reveal system-spanning said display means displays comprising the vzaimodeystvuschih systems.
Which replaced the classic new approaches to prognozirovazirovaniyu have appeared in order to overcome some of these problem.Eti approaches are based on the use of such branches of modern mathematics as neurocomputers, stochastic modeling theory (chaos theory), the theory of catastrophes, synergetic and theory of self-organizing systems, including genetic algorithms, fractal theory and a clear logic. It is believed that these methods will increase the depth of the forecast in the financial markets by identifying hidden patterns inherent in these rynkam.Takim way, due to the fact that within the framework of the classical approach is not possible to obtain a significant improvement in the quality of the securities exchange prediction in the stock market, the actual It is to improve prediction techniques, combining the advantages of chaos theory, cellular automata, and the theory of fuzzy sets.

1. Methods based on the construction of multivariate correlation and regression models.

2. autoregression methods, taking into account the relationship of the members of the time series.

3. Methods based on the decomposition of the time series into components: trend, seasonal variation, the cyclic component and the random component.

4. Methods allowing to consider nonequivalence original data.

5. direct interpolation methods using different trend models.

To date, of the groups listed above prediction methods and the most widely used in actual calculations receiving method of the third group. Most often, the real economic and mathematical modeling focuses on the analysis of trends and seasonality. In this construct the predictive model is implemented under consideration BP through its conversion into a basic model of the time series. Similarly, each element, i.e. each number in the base model of the time series is obtained by multiplying five components: "Data = trend? seasonality? cyclical? regularity? eventfulness ".Soderzhatelnoe definition of these five components in the event of economic forecasting is as follows:

1. Long-term trend (trend) indicates truly long-term behavior of the time series, as a rule, in the form of a straight line or exponential, rarely, a power curve. This is useful if you want to see the whole picture.

2. Similarly recurring seasonal component determines the effect of time goda.Kazhdy period of time during the year is characterized by its seasonal index, which indicates how much higher or lower than the corresponding figure in the given period of time as compared to other periods.

3. Medium cyclic component consists of successive elevations and depressions which are not repeated regularly, for example every year and are therefore excluded from seasonal components. Because these alternate raising and lowering, can not be considered random enough and seen as part of an independent random error (irregular

hydrochloric components). Cyclic variation is particularly difficult to predict beyond the immediate future. Nevertheless, it can be very important, since the main effects of the economic cycle (such as the economic downturn) are treated as part of the cyclical variation in economic performance.

4. Short-term irregular (random) component is the residual variation, which can not be explained. It manifests the action of the one-time events that occur over time by chance, but not systematically. The most that can be done with this irregular component to assess its value, using, for example, a standard deviation, to determine whether it changes over time, and to recognize that even in ideal conditions forecast may not be accurate (on average) than a typical value irregular variation.

5. An event component or briefly "event driven component» (unusual events) occurs in the dynamics of the time series, to levels which in some way influenced by the current event global or local character.

These five basic components of the time series (trend, seasonality, cyclical, random and event driven components) can be assessed in different ways. Below is a brief overview of the methods, which are based on the moving average. The basis of these methods is the division number of elements on the ordinate value moving average.

1. The moving average is used to eliminate seasonal effects averaging over the year, as well as to reduce the irregular component and obtaining a combination of trend and cyclical components.

2. The division number of elements starting at the value corresponding to the ordinate number of the smoothed moving average, giving a "related to the moving average", which gives us seasonal and irregular value. Performing grouping by seasonal periods, such as the time of year and then averaged in the groups obtained, a seasonal index for each season. Performing dividing each value of the series on the appropriate seasonal index for the appropriate time of the year, we find the value of seasonally adjusted.

3. Regression seasonally adjusted series (Y) with respect to time (X) serves for evaluation of long-term trend of the straight line as a function of time, i.e., This time the variable X may consist of the numbers 1,2,3, .... This trend (trend) does not reflect seasonal fluctuations and enables a forecast seasonally adjusted.

4. Forecasting performed taking into account the seasonal trend. Receiving from the regression equation predicted values ​​(trend) for future periods of time, and then multiply them by the appropriate seasonal index, you can get forecasts which reflect both long-term trend and seasonal behavior.

The main error of the existing classification schemes is a violation of classification principles. Among these basic principles are: sufficient comprehensiveness prognostic methods unity classification attribute for each division level in the multi-level classification, disjointness classification sections openness classification scheme, i.e. the possibility of new methods of addition.