Twenty-five years ago William Dewald and Harry Johnson (1963) published their path-breaking Federal Reserve reaction function. By regressing an indicator of Fed policy on the Fed's goal variables, such as the unemployment rate and the price index, they tried to replace vague talk about the Fed's response to economic conditions with more rigorous econometric procedures. Not surprisingly, their work gave rise to an extensive research effort. Has that effort been successful? One way to approach that question is to see if Fed reaction functions generally have reached similar conclusions, so that one can say that certain results have been well established. Another way is to ask if the results reached by the use of Fed reaction functions are robust with respect to more or less arbitrary differences in specifications. Unfortunately, as this chapter will show, neither of those conditions has been met.

This is unfortunate, because a reliable Fed reaction function is needed for at least three purposes. One obviously is to predict Fed actions. Another is to evaluate Fed behavior. This is relevant for the debates about monetary rules and the appropriate degree of Fed independence. A third purpose is to aid in estimating the policy multipliers for econometric models. As Stephen Goldfeld and Allan Blinder (1972), among others, have shown, omitting the central bank's reaction function in an econometric model can lead to serious estimation errors.

There exist two types of reaction functions that answer quite distinct questions. One, which may be called an “intentions function,” asks how the Fed wants to change aggregate demand when, say, the unemployment rate changes.