A Joint Model for the Term Structure of Interest Rates and Realized Volatility.

This article develops a no-arbitrage term structure model that jointly characterizes bond yield levels and realized bond market volatility by linking realized yield curve covariation to a multivariate generalized autoregressive conditional heteroskedasticity (GARCH)-type conditional covariation. The model is tractable, estimated via an exact Kalman-type filtering algorithm, and incorporates measurement equations for both yield levels and realized covariation constructed from high-frequency data. Empirical analysis using U.S. Treasury bond data from 1990 to 2019 shows that conditional yield curve covariation is priced in long-term yields, improves multi-step ahead forecasting of realized covariation, and aids in quantifying interest-rate risk and risk compensation. The model outperforms benchmarks in forecasting realized covariation and facilitates the construction of optimal mean-variance bond portfolios, highlighting the economic relevance of time-varying conditional covariation in bond pricing and risk management.