Abstract:

An important issue in predicting the processes of the development of economic systems, including regional economic and industrial ones, is choosing a prediction model that would most adequately reflect the trends in the development of these systems. The most common forecasting models, i.e., the linear, the quadratic, the exponential, the auto-regression, and the Holt-Winters model are based on extrapolation, which is to say carrying a trend observed in the past or the present to extend into the future. In practice, economic systems develop under the conditions of uncertainty and incomplete observability of the processes of their functioning, that is, under the conditions of non-linear processes. This circumstance allowed the authors to prove that it is expedient to make predictions based on neural networks. A comparative assessment of the efficiency of different prediction models forecasting was carried out using as an example the assessment of the development of the Astrakhan region in 2016-2017, with the following input data: GRP, volume of industrial output, total agricultural output, volume of investments into fixed capital, volume of construction works, average monthly wages, consumer price index, the unemployment rate for 2001-2014. The obtained values of prediction errors for the economic development of the region based on neural networks indicate a higher degree of objectivity for the neural network-based prediction results compared with other models.