Conventional time series methods have been generally unsuccessful in foreca
sting interest rates, with fitted ARIMA models being close to random walks.
The method proposed here is to forecast the whole of the yield curve, from
which forecasts of individual rates may be extracted. The method is applie
d to Australian Government bond data and the forecasts derived from the pro
posed method appear to perform at least as well as the naive no-change fore
casts. (C) 1999 IMACS/Elsevier Science B.V. All rights reserved.