This paper studies the extent to which macro-finance term structure models are susceptible to predictive uncertainty. We propose a general form of arbitrage-free models and quantify the relative importance of unpredictable priced risk variance, as well as macro-finance model uncertainty and learning uncertainty in predictability. Predictive performance and relative contributions of uncertainty sources are dynamically measured based on Bayesian methods, revealing dominating priced risk variance and other important uncertainty sources at different points in time. Macro-finance model uncertainty is high for near-term forward spread forecasts and contributes up to 87% of predictive uncertainty prior to recessions, implying strong dispersion in the information content of macro variables when forming near-term monetary policy expectations.

The decoupling of US short and long interest rates has been a distinctive feature of the 2000s. We employ recent advances in panel econometrics to document this disconnect for industrial countries and link it to a global latent factor in long term rates. We investigate whether international forces, such as global inflation, global output, or the global savings glut may be behind this global latent factor. The savings glut is the most likely contender, suggesting that reserve accumulation and a search for yield from emerging markets has lowered long rates internationally, driving a wedge between domestic short and long rates.