principal cost is the excess noise in estimated perceptions that occurs in the absence of
structural shifts.
The constant-gain learning algorithm that was chosen differs from the inflation-based
algorithm examined in Kozicki and Tinsley (2001b). Implicitly, inflation-based constant
gain learning assumes considerable ignorance on the part of the private sector about the
economic structure and requires policy actions to first influence aggregate demand, which
in turn affects inflation, before the perceived target adjusts. However, additional lags in the
learning process are by assumption. Given the structure of the model in the current paper,
learning is based on unexpected movements in the funds rate as these are direct, but noisy,
measurements of shifts in the inflation target. A funds rate that is higher than expected
may reflect a permanent reduction in the inflation target or a transient positive policy
shock to the funds rate, and vice versa. With policy response-based learning, movements
in the perceived target may (but need not) lead inflation. By contrast, with inflation-based
constant gain learning, inflation necessarily leads the perceived target. Effects of shocks
will dissipate more slowly because inflation affects the perceived target and shifts in the
perceived target feed back into inflation.
To isolate the effects of imperfect policy credibility, private sector agents are assumed
to know the structure and parameters of the economy with the exception of the level and
evolution of the inflation target and the transitory policy shock, ur,t . Since they know the
general form of the policy rule—including policy response coefficients to the output gap,
Yy, to inflation deviations, γπ, and to the policy smoothing parameter, p—the deviation of
the federal funds rate, rt , from private sector expectations of the federal funds rate, rte,
provides noisy information on the deviation of the perceived target, πP (t), from the true
target, πT (t),
rt - rte = γπ (πP (t) - πT (t)) + ur,t. (3)
Private sector learning about the target is modeled by assuming agents use a perpetual
13