Stochastic Optimization and Worst-Case Analysis in Monetary Policy Design

In this paper we compare expected loss minimization to worst-case or minimax analysis in the design of simple Taylor-style rules for monetary policy using a small model estimated for the euro area by Orphanides and Wieland (2000). We find that rules optimized under a minimax objective in the presence of general parameter and shock uncertainty do not imply extreme policy activism. Such rules tend to obey the Brainard principle of cautionary policymaking in much the same way as rules derived by expected loss minimization. Rules derived by means of minimax analysis are effective insurance policies limiting maximum loss over ranges of parameter values to be set by the policy maker. In practice, we propose to set these ranges with an eye towards the cost of such insurance cover in terms of the implied increase in expected inflation variability