Creative Commons

Exchange rates uctuate due to the continuous changes in supply and demand for di erent currencies. Also important factors that determine the value of a currency are the GDP growth, the in ation, and the output growth of a country. Consequently forecasting exchange rates is important not only for a pro table FX trading strategy but also for macroeconomic reasons. Exchange rates contain information about the in ation, the level of income (population purchasing power), and the output growth of a speci c set of countries. There are two broad methods for determining future exchange rates, called \fundamental analysis" and \nonstructural analysis". Fundamental analysis is based in solid economic theories about growth, interest rates etc, hence it is a theoretical way of trying to predict future exchange rate movements. Nonstructural analysis is not based in economic theories, but instead uses the information hidden in the past values in order to make predictions for the future. Hence nonstructural analysis simply uses advanced statistics in order to predict the movements of exchange rates. Some of these models are the vector autoregression models, the various Markov models,the non-parametric models etc. However, an overview of the research suggests that while some successful predictions are possible, no one type of analysis can yield accurate forecasts for di erent currencies for long horizons. In this work structural (Purchasing Power Parity, Interest Rate Parity) and non- structural models were applied to study three very important exchange rates, euro, pound sterling, and yen versus the US dollar. Our work shows that Stochastic pro- cesses seem to outperform the other analysis models (Vector Auto-Regression pro- cess) during relatively long intervals (1 year). For time horizons (of the order of a month) VAR modelling can give more accurate results. A combination of both fundamental analysis and nonstructural analysis is more e ective for analyzing exchange rates time-series data.