Rethinking cointegration and the expectation hypothesis of the term structure

Empirical investigations of the expectations hypothesis of the term structure often test the stationarity on the yield spread. We show that as the term difference increases this stationarity breaks down even under the most favorable assumption with regard to the risk premium. As a result, these cointegration tests are inappropriate. We conduct Monte Carlo simulations and provide empirical evidences that the frequency with which the data fail to reject no cointegration increases as the term difference increases. This finding remains robust after we account for a proxy for risk premium, asymmetries, and employ a new augmented error correction model that allows for time-varying error correction terms.