Predictability of Short-Term Interest Rates: A Multifactor Model for the Term Structure
This study examines whether information contained in the term structure of interest rates can be used to improve forecasts of short-term interest rates when compared to forecasts using only past values of the interest rate being forecasted. Lagged values of the factors representing the term structure are used in a multifactor forecasting model. The multifactor forecasting model generates out-of-sample forecast errors which grow as the forecast horizon increases. A simple autoregressive forecasting model which uses the lagged value of the interest rate being forecasted as the explanatory variable is used to make out-of-sample forecasts. The comparison of forecast errors indicates that the multifactor model is more accurate than the autoregressive model, that the difference in accuracy is greater for longer forecast horizons, and that the difference in accuracy is larger for shorter term interest rates. The fundamental conclusion is that using information reflecting the entire term structure extends the length of the forecasting horizon for a given level of accuracy beyond that attainable using just information from lagged values of interest rate being forecasted.