A consensus has recently emerged that a number of variables in addition to the level, slope, and curvature of the term structure can help predict interest rates and excess bond returns. We demonstrate that the statistical tests that have been used to support this conclusion are subject to very large size distortions from a previously unrecognized problem arising from highly persistent regressors and correlation between the true predictors and lags of the dependent variable. We revisit the evidence using tests that are robust to this problem and conclude that the current consensus is wrong. Only the level and the slope of the yield curve are robust predictors of excess bond returns, and there is no robust and convincing evidence for unspanned macro risk.
yield curve spanning bond returns small-sample bias robust inference