This paper reexamines a familiar but unsettling result in the foreign excha
nge literature: that the forward rate is not an unbiased predictor of the f
uture spot rate. The paper outlines why some frequently used tests of unbia
sedness are non-informative in the sense that they are incapable of correct
ly testing the hypothesis. Specifically, many of these tests are based on r
egressions that suffer from simultaneity bias, resulting in biased and inco
nsistent estimators. This is true whether the tests are conducted using sta
tionary or non-stationary data. We demonstrate this point both analytically
and with simulations. Tests of co-integration, which are not subject to th
e critique presented in the paper, generally fail to reject unbiasedness.