PhD in Economics, Lecturer at the Department of Accounting in Manufacturing
Ternopil National Economic University
ARTIFICIAL NEURAL NETWORKS IN FOREIGN EXCHANGE MARKET FORECAST
Background. Effective risk management incorporates reduction of firm’s sensitivity to changes in real exchange rates. Thus, author investigates methods of forecasting exchange rates for strategic planning. Research indicates that conventional statistical approaches suffer from unsatisfactory accuracy of forecasts. However, artificial neural networks have proven effective for difficult prediction problems in a variety of domains.
Analysis of recent publications in this realm of research has shown that despite significant scientific achievements modern economic conditions pose new challenges for international firms. Hence, tools for projection of operations in foreign currency need to be further investigated.
The aim of the article is an experimental justification of satisfactory neural networks accuracy for forecasting exchange rates (i.e. EUR / USD, GBP / USD, USD / JPY, USD / UAH) with daily, monthly and quarterly steps in order to be exploit by enterprises, central banks, and other end-users.
Materials and methods. The study employs following methods: comparison, approximation, abstraction as well as graphical and tabular tools.
Results. The choice of neural networks as a forecasting method is justified. They are capable to assess dynamics and nonlinearity of financial data better than other known methods. Presented experimental results bear these out. In particular, author uses a multilayered perceptron with a single hidden layer as a neural network model. Forecasting has been implemented with one-step approach where perceptron retrains on every step of experiment. The accuracy level achieved is acceptable for developing measures to manage currency risk.
Conclusion. Observed prediction accuracy of neural networks proves their advantages for firm’s management of foreign exchange transactions, selection of the hedging methods and evaluating hedging results, varying the dates of payments in foreign currency, development of the intervention policy of the central bank, etc. Their approximation abilities show the possibility to analyze and consider psychological boundaries and other behavioral factors affecting exchange rates.
Integration of fundamental (macroeconomic) data into the experimental model yet represents another interesting area of future research.
Keywords: strategic planning, forecasting of exchange rates, artificial intelligence, neural networks.
FREE FULL TEXT (PDF)