Short-term Currency Exchange Rate Forecasting with Econometric Models

Currency exchange rate forecasting is a subject that concerns many parties that participate in foreign exchange market. Being the largest market in the world in terms of trading volume, foreign exchange market is a complex system with many participants, different currency instruments traded, and a 24 hour timeframe. Statistical approaches such as ARIMA and GARCH models are widely used for analyzing time series financial data. The main problem of this thesis is to find which of these models produces more accurate forecast. In this thesis daily data of USD/EUR and GBP/JPY currency pairs were analyzed throughout the period of 10 years, from January 2000 to March 2010. Analysis was carried out based on Box–Jenkins methodology with Eviews 6.0 statistical package. 8 ARIMA and 8 GARCH variants were estimated for each currency pair. Models with at least one statistically significant variance equation coefficient and the lowest Schwartz Criterion value were selected for forecasting. One-step ahead, out-of-sample forecasting was carried out for USD/EUR and GBP/JPY currency pairs. Forecasting results show that GARCH model outperforms ARIMA model and Naïve Forecasting in terms of Mean Absolute Percentage Error. ARIMA forecasts for USD/EUR currency pair was also outperformed by Naïve method.