The Expectations Hypothesis of the Term Structure and Time Varying Risk Premia : A Panel Data Approach

The expectations hypothesis of the term structure of interest rates implies that the spread between short and long bond yields should forecast next period's change in the long yield. Regression based tests have systematically rejected the expectations hypothesis, with estimated coefficients far from their hypothesised values. One explanation of this rejection is that regression tests fail to account for time varying risk premia that are correlated with the spread, causing a downward bias in the estimated regression parameters. This paper uses panel data in order to test the expectations hypothesis in the presence of time varying risk premia. It is assumed that risk premia are driven by single factor, and that they are linear in bond maturity. This allows the unobserved time varying risk premia to be captured by time-specific fixed effects in a panel data regression. The hypothesis that risk premia are linear in maturity is tested by the inclusion of bond specific fixed effects in the regression. The bond specific effects are not significant, implying that risk premia are well approximated by a linear single factor model. When risk premia are handled in this way, the estimated coefficient in the regression remains significantly less than unity but is insignificantly different from zero, implying that in contrast with existing studies yield spreads have no predictive power for changes in the long yield