MODELS OF SCENARIO FORECASTING OF ECONOMIC CRISES ON THE BASIS OF HYBRID APPROACH

Abstract

The present article studies a semi-parametrical method of macro-economic forecast in periods of sharp changes in economy put forward by the authors. Data blocks are chosen on the basis of clusterization methods, which are as close to the current economic conditions as possible. The key method of clusterization is the method of the closest neighbor. Time series is split into blocks and then the closest block is chosen for the last block of observation, which demonstrates the idea of coordination of directive series movements. As a basic model of forecasting the authors use ARIMA model. The authors show advantages of this approach for forecasting during the great recession – the economic slump of 2008 for such variables as inflation rate, unemployment and real private income. This method demonstrates its superiority in comparison with parametrical linear, non-linear, single-dimension and multidimension alternative methods for the period 2007–2019. The article provides calculation s obtained as a result of computer experiment using Python language for data on inflation rate and oil prices for the mentioned period. This approach in future can be used in intellectual methods of machine teaching, such as neuron networks.