Forecasting Australian GDP Growth Using Coefficients Constrained by A Term Structure Model

Yield spread between long and short bonds has been used to forecast economic activity for a long time and has yielded some positive results, particularly for the U.S. data. Recently it has been shown that the forecast can be improved by incorporating the economic activity variable into a term structure model with observable factors. The idea is to constrain the parameters of the system by the term structure model and see whether the constrained model produces better forecasts for the economic activity. This has been done for the U.S. We test this model on Australian GDP growth. We find the forecast results using constrained parameters are quite poor compared to those for the unconstrained model. The reason is that in the traditional affine yield model, researchers normally assume a mean reverting process for factors such as short rate. When this is not supported by the data, the forecast results could be quite poor. To overcome this problem, one might want to twist the affine factor model so that it can accommodate non-mean reverting processes for factors such as the s