Abstract: This paper analyzes the optimality of reactive feedback rules advocated by
neo-Keynesians, and constant money growth rules proposed by monetarists.
The basis for this controversy is not merely a disagreement concerning
sources and impacts of uncertainty in the economy, but also an apparent
fundamental difference in the attitude toward uncertainty about models. To
address these differences, this paper compares the relative reactiveness of
a monetary policy instrument to conditioning information for two starkly
differing versions of model uncertainty about the model and the data
driving it: Bayesian uncertainty that assumes known probability
distributions for a model's parameters and the data Knightian uncertainty
that does not. In the latter case, the policy maker copes with extreme
uncertainty by playing a mental game against "natuare," using minmax
strategies. Contrary to common intuition, extreme uncertainty about a
model's parameters does not necessarily imply less responsiveness to
conditioning information--here represented by the lagged gap between
nominal income growth and its trend--and it certainly does not justify
constancy of money growth except in an extreme version of Brainard's (1967)
result. A partial constant growth rule can be derived in only one special
case: if the conditioning variable in the feedback rule is also uncertain
in either Bayesian or Knightian senses and the authority used
Neyman-Pearson likelihood ratio tests to distinguish noise from information
with each new observation.
Keywords: Monetary policy. model uncertainty, minimax strategies, signal detection, Bayes' risk, Bayesian policy
Full paper (219 KB PDF)
Home | FEDS | List of 2001 FEDS papers
Accessibility
To comment on this site, please fill out our feedback form.
Last update: January 12, 2001
|