In his monograph The Conquest of American Inflation, Sargent (1999) points out the perils of econometric policy evaluation of the Theil-Tinbergen tradition wherein one estimates a reduced form econometric model of the economy and subjects it to control. If the model is misspecified, as is usually the case, the resulting economic performance can be poor. A substantial literature adopting the Bayesian approach to parameter uncertainty has devel- oped which treats estimated model parameters as being correct up to a well-defined ran- dom error. Policy is then designed to take account of that error. In most instances, the policy prescription that arises from such an exercise is one of attenuation; that is, policy is less aggressive than the certainty equivalent case, or in the words of Blinder (1998) needs to "get it right, then do less." As is well known from Lucas (1976) however, the `random error' may not be random at all but rather could be a function of the conduct of monetary policy. In contrast to the Bayesian approach, the Knightian response to uncertainty leads, in most cases, to policy that is more aggressive than the certainty equivalent case. This paper examines a variant of Sargent's problem where a monetary authority is recursively estimating and controlling a misspecified model, getting wrong the persistence of infla- tion. We compare the Bayesian and Knightian (robust) approaches to model uncertainty. We show that in at least some circumstances, the authority is better off utilizing the robust approach and ignoring Blinder's advice. The robust policymaker protects against statisti- cally unlikely inflation instability--indirectly acknowledging the Lucas critique--and par- tially avoids the pitfalls Lucas warned of.
