Exchange rate forecasting with an artificial neural network model : can we beat a random walk model?

Abstract

Developing an understanding of exchange rate movements has long been an extremely important task because an ability to produce accurate forecasts of exchange rates has practical as well as theoretical value. The practical value lies in the ability of good forecasts to provide useful information for investors in asset allocation, business firms in risk hedging, and governments in policy making. On the theoretical side, whether a currency price is predictable or not has important implications for the efficient market hypothesis in the foreign exchange market and for theoretical modelling in international finance.

Owing to the importance of the movements of exchange rates in our real life, such as financial hedges and investment abroad, this research investigates the possibility of an accurate pattern of the exchange rate movement. The purpose of this research is to carry out an empirical investigation into the extent to which nonlinear econometric models can improve upon the predictability of foreign exchange rates compared to a standard regression model. We will use the artificial neural network approach and employ macroeconomic fundamental variables, including relative money supply, relative income, interest rate differential and inflation rate differential to examine whether or not the artificial neural network model could significantly improve the accuracy of describing the movement of exchange rates and the predictability of exchange rates, especially out-of-sample. The empirical research will focus on the New Zealand exchange rate with the currencies of its major trading partners (Australia and the United States).