Integrating Exchange Rate Theory in Data Mining

This paper focuses on an integration of exchange rate theory in a data mining process for the purpose of forecasting. The applied approach is centred by a Genetic Algorithm (GA) and Neural Networks (ANN), which allows the identification of relationships that are not describable by economic theory. As experience showed, relationships derived by data mining are often not convincing as regards their correctness and effectiveness. Most data mining approaches do not contribute much to persuade otherwise. In this work, it is tried to remove parts of this limitation by combining economic theory with data mining. However, usually the role of economic theory in exchange rate forecasting is to identify a list of relevant variables to be included in the analysis, with possibly and plausible signs of their coefficients. Previous research documents the failure of this way of analysis and thus for exchange rate theory in forecasting exchange rates at frequencies up to one month. Consistent with these results, in this paper the role of economic theory is extended as it is implemented in data mining as a framework in which and among which the possibilities of data mining are exploited. Other findings include: (i) During the years 2000 and 2001 countries relative economic growth is most significant in forecasting exchange rates one period ahead, (ii) the fmancial market is of major concern when explaining exchange rate movements as it may be used to proxy real economic activity as well as it maps massive capital flows between countries, which in turn affect exchange rates. (iii) fundamental forecast ing is more effective on lower frequencies. The approach is illustrated in some detail for five exchange rates on a daily, weekly and monthly frequency.