Forecasting Exchange Rates: A VAR Analysis

This thesis aims to out-of-sample forecast the USD/EUR exchange rate using four macroeconomic variables, namely ination, interest rate, unemployment rate and industrial production index. The model applied is the vector autoregressive model. We use monthly data for a period of 2002-2011 and use the data from 2012 in order to compare the forecast accuracy with the random walk, which is believed to outperform many models when forecasting for a short-time horizon, such as one year. We found out that the vector autogressive model beat the random walk in the period of one and three months, which was surprising. In the longer horizon of six, nine and twelve months, random walk, as expected, heavily outperformed vector autogressive model. The reasoning behind this could be that there was no clear trend in the USD/EUR exchange rate during this period