In this paper we examine variance bound tests of the joint hypothesis
that (1) bond markets are efficient and (2) the term structure is dete
rmined by the expectations hypothesis. Both the Singleton and Shiller
tests are shown to be seriously biased toward rejecting the joint hypo
thesis in finite samples. Flavin's test is unbiased but has a very hig
h variance leading to many false rejections of the joint hypothesis. W
hen corrected as suggested by Flavin, Shiller's test is unbiased and h
as a relatively low variance. Unfortunately, it is also sensitive to m
easurement error.