We show that uncertainty about parameters of the short rate model can account for the rejections of the expectations hypothesis for the term structure of interest rates. We assume that agents employ Bayes rule to learn parameter values in the context of a model that is subject to stochastic structural breaks. We show that parameter uncertainty also implies that the verdict on the expectations hypothesis varies systematically with the term of the long bond and the particular test employed, in the same way that is found in empirical tests.

The expectations hypothesis of the term structure has been decisively rejected in a large empirical literature that spans several decades. In this paper, using a newly constructed dataset of synthetic zero-coupon bond yields, we show that evidence against the expectations hypothesis is substantially weaker in data generated after the widespread publicity of its failure. These results are consistent with the idea that asset pricing anomalies tend to disappear once they are widely recognized.