We use a novel data set to study return predictability in debt markets. The data are collected from J.P. Morgan’s periodic surveys on its clients’ outlook for changes in US Treasury yields and corporate credit spreads. We document that simple signals constructed from such surveys predict excess returns on debt portfolios formed on the basis of duration (2-years minus zero) or credit quality (BBB minus AAA). A linear trading strategy placing equal weight on Treasury and Credit signals has an annualized Information Ratio equal to 1.18, before transaction costs. We also show that predictability is likely to stem from private information possessed by survey respondents rather than from risk premia.