On forecasting economic indexes by means of neuroevolutionary models

Time series prediction is quite complex and interesting problem. Usually a forecast is based on information from the past, i.e. on previous values offeatures. The economic situation in the country depends on different factors and events and influences many situations. The innovation projects in aerospace field require a large amount of investments and take a long time for development and implementation. These kinds of investments require relevant and reliable information about possible future economic situation to minimize risks and correct the plans in right time. In particular, to get such information one may use the time series of the various indexes of economic development. In this paper, the neural network predictors are used as the main technology offorecasting. They are generated automatically by means of genetic algorithms that allow increasing the quality of the prediction. Besides, the advantages of the use of selfconfiguring genetic algorithms in comparison with standard genetic algorithms are demonstrated when adjusting the weighting coefficients of the neural network. The quality of the solution provided by the neural network depends also on its structure. In this paper, the genetic programming is used for the neural network structure design due to its ability to provide effective and flexible models without a human expert. This allows automating the neural network based predictors configuring. The combination of these intellectual information technologies is tested on the financial time series forecasting task using the values of the prices and volumes along with the technical indicators as the inputs. The developed system allows automated designing the neural network based predictors and getting a high quality forecast of economic indexes changes. The developed system shows better results in the comparison with the other forecasting technologies.