Application of Neural Networks for Forecasting and Automated Decision Making for Stock Market Investments

The class of mathematical models – artificial neural networks – was studied as a method of technical analysis of financial and stock exchange instruments in this paper. This method was chosen due to the widespread popularity of the neural network modelling, being used for solving various classes of problems. At the same time intuitive validity of this approach in terms of problem analysing and financial time series forecasting played its role (artificial neural networks are observed as simulation models of the principles of the brain work and, as it was previously mentioned, a stock market analysis is directly related to the analysis of aggregate expectations of the representative economic agents). The main aim of this study is to conduct our own analysis of the efficiency of the models, which are based on neural networks, in solving the problems of financial forecasting. The results and conclusions could also be extended to many other areas of neural networks due to their uniform approaches. It should also be noted that the analysis, described in this paper, is not limited by choosing only one specific model, such as architecture, main characteristics of the network and the establishment of relationships to the amount of effectiveness of prediction. An important step in the research was to form the relevant data from the general empirical array for neural network forecasting and to analyse the preliminary handling method. The shares issued by the Russian companies that are a part of RTS Index (Sberbank, VTB, Bashneft and Rosneft) were selected among the analysed securities. The analysis of the relationship between the main characteristics of a neural network and the value of the adaptive prediction accuracy was conducted for each of the given instruments. In conclusion, the structures of networks that were optimized in frames of the study period were defined for each of the exchange asset.