A SVM approach to foreign exchange forecasting

This paper presents a new approach based on Support Vector Machine (SVM) for forecasting the currency exchange rate between the U.S. dollar and the euro. The variables considered to determine exchange rate movements are: the level of inflation, the interest rate differential, the GDP, the level of productivity, the current account balance, the equity flow, and the current fiscal position among others. Out of these variables, the five variables having the greatest influence on the exchange rate were selected for both zones. The data for these variables was collected quarterly from the beginning of 1999, which was the year of the introduction of the euro, until the third quarter of 2007. Different sources of databases were used to collect this information: the U.S. Federal Reserve Board, the Eurostat, the European Central Bank statistical warehouse, and the U.S. Bureau of Economic Analysis, among others. Support Vector Machines (SVM) are a type of learning machine based on the statistical learning theory, which implements the structural risk minimization inductive principle in order to obtain a good generalization from data sets of limited size. There has been a great deal of research interest in these methods over the last years, because they provide good generalization on the data, and they are well suited for sparse data; i.e. a reduced set of data is enough to obtain good estimations. In addition, they exhibit independence of the results from the input space dimension; the number of variables chosen to describe the patterns and is not critical. The results obtained show SVM's forecasting power as a sound alternative to other conventional forecasting tools such as neural networks and ARIMA models in the international finance area.