Empirical tests of the deterrence hypothesis - the idea that crime can be deterred through changes in the costs or benefits derived from committing crime - typically focus on estimation of the relationship between current crime rates and contemporaneous measures of economic conditions, demographics, and enforcement levels. We argue this approach is misguided because both past behavior and future conditions should impact decisions to commit a crime in the present. Accordingly, we develop an econometric model of aggregate crime rates that includes past behavior and expectations of future conditions. Model estimates suggest that expectations of future conditions are important in determining current crime rates. One implication of this finding is that the long run elasticity of crime rates with respect to policy variables may be orders of magnitude larger than short run elasticities, and depend upon whether the change is permanent or transitory. A second implication of this finding is that credibility and the ability of policy makers to commit are important in thinking about policies designed to deter crime.