Can Learnability Save New-Keynesian Models?

Bennett McCallum (2009), applying Evans and Honkapohja's (2001) results, argues that "learnability" can save New-Keynesian models from their indeterminacies. He claims the unique bounded equilibrium is learnable, and the explosive equilibria are not. However, he assumes that agents can directly observe the monetary policy shock. Reversing this assumption, I find the opposite result: the bounded equilibrium is not learnable and the unbounded equilibria are learnable. More generally, I argue that a threat by the Fed to move to an "unlearnable" equilibrium for all but one value of inflation is a poor foundation for choosing the bounded equilibrium of a New-Keynesian model.