Forecasting the Australian yield curve

Abstract

We apply a number of forecasting models to Australian Government Bond yields. All methods rely solely on the history of yields. Consistent with findings from US Treasury data, we show that the simplest forecasting models across all maturities and forecasting horizons are also generally the best: the forward yield (when avail- able) and the random walk model. Models with more structure-e.g. principal components and Bayesian vector autoregression-can help forecast overnight yields at very short horizons, but provide little or no improvement in other cases.