Forecasting Australian Exchange Rate Volatility: A Comparative Study of Alternate Modelling Techniques and the Impact of Power Transformations.

The modelling and forecasting of exchange rates and their volatility has important implications for many issues in economics and finance. This paper compares the ability of Autoregressive Conditional Heteroscedasticity, Autoregressive and Mean models to forecast the magnitude of change in 19 Australian bilateral exchange rate series. Further, the impact of power transformations on the accuracy of the AR models is considered. The results of this paper suggest that ARCH models generate superior forecasting performance when considering the squared returns of an exchange rate series. However, when one considers power terms other than a squared term, the forecasting accuracy of the AR models often exceeds that of the ARCH models. All models performed reasonably well in a conventional test of forecasting efficiency.