The Lucas critique holds that policy evaluations based on historical correlations can fail because policy changes alter expectation formation. We develop a new approach to monetary policy evaluation that addresses this concern: we elicit expectations under alternative policy scenarios from household surveys and feed these measured expectations into a heterogeneous agent model. The surveys reveal that the response of income and inflation expectations to interest rate changes is state-dependent. Incorporating these expectation differences into the model yields estimates of the effects of policy on aggregate consumption that are statedependent, varying with economic conditions at the time of the policy change.

