A closely related result is that rules involving model-based forecasts generally do not outperform first-difference rules based on the current output gap and inflation rate, and quite often generate higher variability of output and inflation. Finally, the class of first-difference rules is robust to model uncertainty, in the sense that a first-difference rule taken from the policy frontier of one model is very close to the policy frontier of each of the other three models. In contrast, we find that more complicated rules are somewhat less robust to model uncertainty: rules with a larger number of free parameters can be fine-tuned to the dynamics of a specific model, but often perform poorly in other models compared with the best simple rules.

The approach of evaluating policy rules used in this paper follows the long and distinguished tradition dating to Phillips (1954).3 As is standard in this literature, we assume the objective of policy is to minimize the weighted sum of the unconditional variances of the inflation rate and the output gap (the percent deviation of GDP from its potential level). Source

In addition, we allow that interest rate volatility may enter into the policymakers’ optimization problem, either through preferences or constraints on policy actions. The funds rate is set according to a time-invariant policy rule. For a given class of policy rules, the policy frontier traces out the best obtainable outcomes in terms of inflation, output, and funds rate volatility.

We refer to the policy rules underlying such a frontier as “optimal” in the sense that these rules represent solutions to the specified constrained optimization problem.

A closely related result is that rules involving model-based forecasts generally do not outperform first-difference rules based on the current output gap and inflation rate, and quite often generate higher variability of output and inflation. Finally, the class of first-difference rules is robust to model uncertainty, in the sense that a first-difference rule taken from the policy frontier of one model is very close to the policy frontier of each of the other three models. In contrast, we find that more complicated rules are somewhat less robust to model uncertainty: rules with a larger number of free parameters can be fine-tuned to the dynamics of a specific model, but often perform poorly in other models compared with the best simple