Exchange Rate Modelling using Artificial Neural Networks

William Verkooijen
Tilburg University
W.J.H.Verkooijen@kub.nl

Joseph Plasmans
University of Antwerp
fte.Plasmans.J@alpha.ufsia.ac.be

Hennie Daniëls
Tilburg University
Daniels@kub.nl

Abstract

In the economics literature on exchange rate determination no theory has yet been found that performs well in prediction experiments. Only very seldom the simple random walk model has been significantly outperformed. The aim of this paper is to investigate whether neural network (nonlinear) model specification improves prediction performance of identified structural exchange rate models, which are traditionally estimated by (linear) regression methods or by (transfer function) time series methods. The empirical experiences for the dollar-deutsche mark, dollar-guilder, dollar-pound, and dollar-yen exchange rates, indicate that neglected nonlinearities in mean are not a likely cause for the generally bad prediction performance of the structural exchange rate models.