Why Random Walk Models of the Term Structure Are Hard to Reject

Tests of random walk models of the term structure generally fail to reject the null hypothesis, whereas direct tests of the fair game-efficient markets (FGEM) hypothesis generally reject the null. Random walk tests can be interpreted as FGEM tests that add measurement errors to the forward-rate revisions used in direct tests. Our empirical application is consistent with the literature; direct tests strongly reject the null, but random walk tests do not, despite using the same data and numbers of observations. The random walk measurement error is shown to reduce local asymptotic test power, which may explain this empirical puzzle.