INTRODUCTION

This paper describes experiments with human subjects in an environment that provokes the time-consistency problem of Kydland and Prescott (1977). There is an expectational Phillips curve, a single policymaker, who sets inflation up to a random error term, and members of the public, who forecast the inflation rate. The policymaker knows the model. Kydland and Prescott consider a one-period model and describe how the inability to commit to an inflation policy causes the policymaker to set inflation to a Nash (that is, time-consistent) level that is higher than it would be if it could commit. With repetition (see Barro and Gordon 1983), the availability of history-dependent strategies multiplies the range of equilibrium outcomes. Some are better than the one-period, time-consistent one; others are worse.

Some commentators, including Blinder (1998) and McCallum (1995), assert that in practice, the time-consistency problem can be solved through an unspecified process that lets the monetary authority “just do it,” in the terminology of an American sports shoe advertisement. Here, “it” is to choose the optimal or Ramsey target inflation rate. Although reputational macroeconomics provides no support for “just do it” as a piece of policy advice, the range of outcomes predicted by that theory is big enough to rationalize such behavior. The large set of outcomes motivated us to put human subjects inside a Kydland–Prescott environment.

We paid undergraduate students to perform as policymakers and private forecasters in a repeated version of the Kydland–Prescott economy.