Dynamic Term Structure Modelling: A Forecasting Perspective

This paper examines the predictability of the term structure of US treasuries. We consider a range of term structure models in which we include richer dynamics. The starting point of our term structure models is the Nelson-Siegel framework. We extend this framework in three ways: (i) allow interac­tions with the macro-economy, (ii) include a time-varying unconditional mean, and (iii) incorporate Markov-switching dynamics. Moreover, we explore the merits of combining forecasts of individual term structure models by considering two types of combination schemes: (a) simple weighting schemes, and (b) time-varying weighting schemes. We find that no individual model consistently outperforms a no-change forecast. However, we do obtain more accurate and more robust forecasts by combining individual forecasts. Furthermore, we find the performance of both individual models and forecast combinations to be highly dependent on the forecasting period. Next, we barely find predictability over short horizons whereas longer horizons show promising results. All in all, we demonstrate how combining forecasts can improve the overall performance.