computed as the unexpected change in the long term interest rate, which is also the unanticipated
change in the yield spread:
n-m
tp
n,m
t
m
∑Eim
t t + mq
q=0
(8)
For instance, in early 1980s, when the economy was suffering from soaring inflation due to effects
of the oil shocks, term premia were unusually large; moreover, GDP growth was weak and the rate
of growth of industrial production became negative. Hence, the increased uncertainty affecting the
economy was also reflected in financial markets affecting the risk-averse attitude of both investors
and consumers. In general, large negative values of the output gap, due to substantial deviations of
the actual GDP from its potential, induce the monetary authority to change preferences in
conducting monetary policy, since it becomes socially optimal to include unemployment among the
final targets. The enhanced complexity of the macro scenario contributes to the changing behaviour
of the monetary authority which, in turn, affects the ability of agents to anticipate accurately the
future dynamics of the term structure. Large term premia reflect both the market inability of
forecasting future monetary policy stances and market participants’ difficulty in anticipating the
future dynamics of the term structure slope.
We estimate the Campbell-Shiller Equation (9) by means of a rolling procedure to highlight the
time-varying pattern of the slope coefficient14 (β).
n-m
∑m m nm
Etit+mq -it = α+ β(it -it )+εt
q=0
(9)
Each regression is estimated by OLS (with Newey-West corrected standard errors) on samples with
60 monthly observations, i.e. five years. Figure 5 shows the rolling estimation of the slope
coefficient and the associate probability values of the t-test (null hypothesis β = 0 ) for the pairs of
maturities (60, 3) and (120, 3). Results for other combination (n, m) confirm the variable path of the
slope coefficient. We recall that the expectation hypothesis holds if β = 1 and α= 0; nevertheless,
following standard practice, we mainly focus on the slope coefficient β. The rolling estimated
slope tends to fluctuate around one; deviations from the value implied by the EH are substantial
though, that is, the variance of the estimated coefficient is quite large over time.
14 Equation (9) here is Equation (5) in Section 4.
16