In this paper we examine the information on inflation contained in the term spread of the Australian term structure in a model in which we allow the expected real term spread to vary with time. Previously, Mishkin (1990) assumed a constant expected real term spread in a similar inflation forecasting model. We further extend the model by allowing the coefficient of the nominal yield spread also to vary with time. Results show that the model based on the time-varying expected real rate, estimated with the Kalman filter, is more suitable than the model based on the constant real rate. Also, the term spread lagged one period has more information on future inflation than the contemporaneous term. Finally, the forecasting power of a model with a randomly time-varying yield spread is inferior to the other versions examined.
