Dependency Analysis and Neural Network Modeling of Currency Exchange Rates

2 : In this paper we explore function dependences in multivariate currency exchange rate data using the-test. Despite large noise levels and the high dimensional nature of the data set, the-test captures the dependency structures in the data quite eeciently. Dependences of a currency on its own history, interdependencies between currencies , and dependences of a currency on the Dow Jones Industrial Index are established from the data by the-test. When the variables identiied in this way are used as inputs to a feedforward neural network to model the exchange rate series, a prediction performance better than status quo is obtained. The results connrm that exchange rates series are not random walk processes and that a certain degree of short term predictability is possible if the dependency structure is properly modeled.