Matching Daily Yields of All Maturities and Their Volatility

This paper presents affine term structure models with Garch heteroscedastic latent factors driving the short rate, and an extension of the Kalman Filter to estimate such models. With six or more factors, these models that can match daily US Treasury yields of all maturities, from one year to thirty years, with almost zero observation errors, and some such models can also closely match yield volatilities. The preferable models feature a single Garch volatility factor unspanned by bond yields, or multiple Garch factors spanned by yields, but some homoscedastic affine models can on average match yield volatilities almost as well as, while stochastic volatility affine models fare much worse