Non-linear interest rate dynamics and forecasting: evidence for US and Australian interest rates

Recent empirical finance research has suggested the potential for interest rate series to exhibit non-linear adjustment to equilibrium. This paper examines a variety of models designed to capture these effects and compares both their in-sample and out-of-sample performance with a linear alternative. Using short- and long-term interest rates we report evidence that a logistic smooth-transition error-correction model is able to best characterize the data and provide superior out-of-sample forecasts, especially for the short rate, over both linear and non-linear alternatives. This model suggests that market dynamics differ depending on whether the deviations from long-run equilibrium are above or below the threshold value