This paper proposes a generalized method for modelling the conditional variance of term structured series. We estimate a model in which the term structure is summarized using three time-varying latent factors - as proposed in Nelson-Siegel (1987) and re-interpreted in Diebold and Li (2000) - which can, in the case of daily returns, display GARCH behavior. Given the substantially dierent structure that characterizes daily log returns with respect to interest rates along the time dimension, we have to reject the possibility to use in full the specication designed for the analysis of the yield curve conrming the fundamental unpredictability of nancial returns. On the other hand, to address the issue of modelling the conditional variance of such series, we augment the model with a GARCH specication showing that there is room for an appreciable reduction of the dimensionality of problems related to conditional variance behavior. In the empirical section we show that this approach allows to handle computations such as volatility impulse response functions- in a multivariate setting.