Estimation of Factors for Term Structures with Dependence Clusters

This paper studies the importance of accounting for term structure maturity clusters while estimating latent factors, for the purpose of forecasting yield curves. The maturity clusters are identified using a hierarchical clustering algorithm. We then propose a new block dynamic Nelson-Siegel model to account for the separate dynamics driving the different clusters.For the US zero coupon bond term structure, we find two (short and long) maturity clusters. The level, slope, and curvature factors driving the two clusters are loosely dependent on each other and further, this dependence is also time-varying. The block dynamic model fits the clustered bond term structure well and produces superior out-of-sample forecasts for the short and medium term maturities, suggesting that clustering is a vital feature that cannot be ignored when estimating common factors in term structures. We further extend the empirical application to the case of Libor-Swap term structure and draw similar conclusions