Limitations of using rules in forecasting currency exchange rates

Fuzzy neural networks have a growing importance for problems where one wants to avoid pure block box solutions or when hints-i.e. known dependencies (rules) -are used. In this paper an application to the field of currency forecasting is examined. It is shown that a classical set of rules does not lead to on acceptable solution and a classical neural network does apply better. The latter is only basing on the time series data which is fed into the network without rules. The example used here is a medium term forecasting of a one month interval. The subject is the Deutschmark/US-Dollar exchange rate over a period of several years