This paper considers the optimal approach to reducing inflation when t
he cost of inflation is its conditional variability. Inflation is stoc
hastically related to money growth, with unobservable time-varying aut
onomous and induced components. A sharp reduction in money growth prov
ides information about the responsiveness of inflation to money, but a
lso induces variability as the economy heads into unknown territory. G
radual policy is always optimal and the model explains why moderate-in
flation countries adopt a much more gradual money growth reduction tha
n high-inflation countries. Additionally, the analysis sheds light on
the more general problem of learning with two unobservable parameters.