This paper discusses statistical and political-economic methods for evaluating different macro-econometric models and suggests three optimization approaches, based on maximizing expected utility of economic welfare, and a design approach to the formulation of economic policy rules under model uncertainty. When one compromise model can be found, an objective risk-averse policy optimization may be appropriate, but otherwise a political approach is required where the government attaches subjective probabilities to each model being the correct description of the economy. A mini-max frequency-domain design of the policy feedback rules is suggested, which ensures suitable correction for exogeneous shocks, satisfactory decoupled dynamic responses of the regulated economy and robustness under all views of the economy. The desired paths, used in the rules, are obtained by maximizing the expected utility of welfare subject to the constraints of the regulated economy. Applications to the derivation of New Keynesian policy rules from reduced forms of the LBS, NIESR and a condensed version of the HMT models of the U.K. economy are provided.