short term policy rate has largely affected the predictability of short term rates. Mankiw and Miron
admit they cannot fully explain the failure of the EH though; however, they believe that time-
varying risk premia, change in risk perception, adjustments in relative asset supplies, measurement
errors, and, finally, near rational, rather than rational, expectations can play a role in explaining the
empirical rejection of EH. Kool and Thornton (2004) find that, if financial panics in 1907 - 1908
are properly taken into account, the evidence in favour of EH is not significantly different before
and after the foundation of the Federal Reserve. They show that the apparent support for the EH
theory before 1915 is merely due to some extreme observations. In line with Mankiw and Miron
(1986), Campbell (1995) points out that, in the regression for predicting short term rate over the life
of the long term bond, changing rational expectations about long bond returns act like a
measurement error; this kind of measurement error biases the coefficient downward toward zero.
The magnitude of the bias depends on the relative variances of the two components of the spread. In
particular, the larger the variance of the theoretical spread relative to the variance of the term
premium, the smaller the size of the bias, i.e. the closer the estimated slope coefficient to one. In
such a situation, agents are supposed to be well informed about future interest rate changes, since
the variance of the term premium, which is the unexpected component of the yield spread, is
relatively small.
Rudebusch (1995) finds that the ability of the term structure to predict future changes in short rates
is quite good at very short horizons, i.e. shorter than about one month. However, he argues that at
horizons longer than two years there is some evidence that the predictive power increases. The EH
has also been examined in different monetary regimes. Hardouvelis (1988) using weekly data shows
that the spread carries substantial predictive power between October 1979 and October 1992. In the
same vein, Simon (1990) finds that the slope of the term structure significantly anticipates future
changes in interest rates during the non-borrowed reserve operating procedure. Roberds, Runkle,
and Whiteman (1996) provide evidence regarding the informative content of the yield curve using
daily data for settlement Wednesdays. Thornton (2005) believes these results are contrary to
common wisdom. He rationalizes these tricky empirical findings and concludes: “these results are
anomalous in that they suggest that the funds rate is more predictable (1) during periods when the
Fed is targeting monetary aggregates than when it is explicitly targeting the federal funds rate and
(2) on days when there are large idiosyncratic shocks to the federal funds rate. I argue that the
funds rate should be more predictable when the Fed is explicitly targeting it.... In addition, I show
that settlement Wednesday changes in the funds rate can, at best, account for a very modest
improvement in the market’s ability to predict the funds rate”.