Abstract
This thesis examines if real-time macro-financial information observed on global financial markets are successful at predicting nominal exchange rates over short time-horizons. In the literature, exchange rate determination primarily focuses on macroeconomic data when testing the empirical success of the monetary and portfolio balance models. During the periods between macroeconomic releases, financial market participants substitute macroeconomic data with macro-financial equivalents. In the author’s experience working on foreign exchange dealing floors, these macro-financial variables provide real-time proxies to guide market participants when determining a preference to buy or sell currencies in order to make profits.
Applying an augmented sticky-price asset model, I find mixed results but the forecasting performance improves when the models are characterised in an error-correction framework or by applying first differences. There is evidence of time-varying coefficients, supporting the findings found in previous studies that fixed-coefficient exchange rate models are not successful at forecasting out-of-sample. The accuracy of the directional change one week ahead is statistically significantly greater than 50 percent for many exchange rates when the exchange rate equation is estimated in first differences. More generally, a lack of cointegration and evidence of structural breaks suggests financial market participants can overstate the application of macro-financial variables as a driver of future exchange rate movements.