Perspectives for the use of vector autoregressions in economic forecasting

Abstract

Subject: one of the new and promising areas of modern economic short- term forecasting is the use of vector autoregressions. The experience of successful application of such models in practice is quite small, although all researchers note that the models are the future of economic forecasting. The article deals with the problems of constructing and using vector autoregressions in short-term economic forecasting. Purpose: the purpose of this study is to identify problems that hinder the introduction of vector autoregressions into the practice of economic forecasting and to find the ways to deal with them. Research design: the article discusses the directions of using vector autoregressions in modern scientific research and in practice; shows the discrepancy between these results and the rich possibilities of vector autoregressions; reveals the main problem – a non-linear growth of a problem complexity with the increase in the dimensionality of the autoregression vector; it is proposed as an alternative to use the complex form of vector autoregressions; a practical example demonstrates the advantage of vector autoregressions in vector form compared to vector autoregressions of real variables. Results: the article shows how to calculate the non-linear increase in the complexity of the problem with an increase in the dimensionality of the modeled indicators vector. It also shows how much the dimensionality of the problem decreases if it is represented in the form of autoregression of complex variables. The decrease of the dimension of the presented in a complex form model is demonstrated on practical and hypothetical examples. To confirm the significance of the obtained results, an example of modeling and short-term forecasting of eight main indices of the Moscow Exchange in the period from February 24, 2022 to May 13, 2022 is given. Vector autoregression models and complex-valued vector autoregression models were built on this dataset. The constructed models were analyzed from the point of view of approximation accuracy and the accuracy of short-term economic forecasting. The results confirm the scientific significance of the proposed group of autoregressive models in a complex form.