Tests of the Expectations Hypothesis of the Term Structure in a Model with Bayesian Learning

Evidence that the term structure of interest rates does not satisfy the expectations hypothesis has been reported in a number of papers. However the nature and degree of this rejection is very sensitive to the exact specification of the tests. In this paper, we identify a source of small sample bias that is consistent with this empirical finding. The direction of the bias which we uncover implies that the expections hypothesis is more likely to be true than asymptotic test statistics would suggest. In financial markets, agents often receive information about the future in a form that cannot be explicityly incorporated in an econometric model. We formalise this idea by assuming that agenst supplement forecasts of the short yield obtained from the econometric model with noisy signals of the error term of the model. Using simulation experiments, we demonstrate that the bias is much more pronounced in the specification which has the more decisive rejection of the expectations hypothesis.