Forecasting the dynamics of the exchange rate using econometric methods and artificial neural networks.

Abstract

The article presents the possibilities of applying the methods of applied statistics used in modern analysis to analyze the dynamics of the exchange rate. The exchange rate in the modern world is quite a significant tool of the economic system of the state. Determining the trend in the development of the exchange rate is important in almost any business direction that involves capital investment. In modern economic conditions, it has a rather differential character, which is accompanied by risk, to one degree or another. And although assets in US dollars have declined in recent years, the Central Bank did not plan to completely abandon the dollar in its reserves, since this currency is necessary to maintain financial stability, given the high role of the US currency in foreign trade settlements and its fundamental role in financial flows. To predict the dynamics of the exchange rate, the official US dollar exchange rate was taken. Methods of constructing predictive models using ARIMA models and prediction models based on neural networks are proposed for making forecasts. The models were evaluated and the most accurate variant was proposed – a model based on the construction of neural networks (MLP model 12-8-1 (No. 4)) with a prediction error of 6 %.