Rationality and Subjective Bond Risk Premia

This paper documents large micro-heterogeneity and forecasting skill in the cross-section of survey based bond risk premia. We reject informationally constrained rational expectations but show a learning model distorted by sentiment is consistent with the data. Aggregating, we propose a belief measure for the marginal agent that is consistent with Friedman's market selection hypothesis. This measure is available in real-time and compares favourably to popular statistical models. Moreover, forecast errors from this measure, while predictable, are not easily corrected in real-time. Finally, we re-assess structural models and and support for both sentiment and time-varying quantity of risk channels