Term Structure Models Can Predict Interest Rate Volatility. But How?

This paper attempts to predict the volatility of interest rates through dynamic term structure models. For this attempt, the models are improved, based on the three-factor Gaussian model, to have level-dependent volatilities supported by data. The empirical results show that the predictive power of the proposed models is higher than that of the affine models. Compared with time-series models, it is low for the four-week forecasting horizon but can be comparable for middle to long term rates by extending the horizon up to 32 weeks. The combination of these two different types of forecasts can lead to higher predictive power.