Deconstructing the Yield Curve.

This article introduces a novel nonparametric bootstrap method for the yield curve that does not rely on specifying the true factor structure of interest rates, addressing challenges posed by strong cross-sectional and time-series dependence in bond data. The method resamples primitive objects—excess returns and a single far-future forward rate—allowing for internally consistent reconstruction of the entire yield curve and valid inference without assuming a parametric model. Simulation evidence demonstrates that this bootstrap controls size well and maintains power in predictive regressions, outperforming existing parametric bootstrap approaches, especially under model misspecification. Empirically, the method revisits bond return predictability, finding that trend inflation has significant predictive power beyond the yield curve, whereas the equilibrium real interest rate does not, contrasting with prior findings based on other bootstrap methods. Additionally, the bootstrap is applied to regression-based tests of the expectations hypothesis and to bias-correct probit models forecasting recessions, illustrating its broad applicability and robustness in financial econometrics.