Daily Prediction of the Foreign Exchange Rate Between the US Dollar and the German Mark Using Neural

In this paper we study the problem of predicting the daily foreign exchange rate (FX) between the US Dollar and the German Mark. The study can be seen as a typical example of predicting noisy time series. Our approach based on neural networks consists of searching for the optimal representation of the underlying market by selecting the optimal training patterns. Moreover we develop a trading system which is tested in several out-of-sample test periods from January 1990 to June 1995. After the encouraging results on the test sets the neural network prediction system has been embedded in a real trading system. Since July 1995 this trading system is successfully used to make online predictions at the FX market in New York.