This paper studies the performance of a methodology that can be used to evaluate the impact of new policies that radically depart from existing ones. It uses data gathered from a randomized schooling subsidy experiment in Mexico (i) to estimate and validate a dynamic behavioral model of parental decisions about fertility and child schooling, (ii) to forecast long-term program impacts that extend beyond the life of the experiment, and (iii) to assess the impact of a variety of counterfactual policies. The behavioral model is estimated using data on families in the randomized-out control group and in the treatment group prior to the experiment, both of which did not receive any subsidy. Child wages provide a valuable source of variation in the data for identifying subsidy effects. Using the estimated model, we predict the effects of school subsidies according to the schedule that was implemented under the Mexican PROGRESA program. We compare the predicted impacts to the experimental benchmarks and find that the model's predictions track the experimental results closely. The model is also used to simulate the effects of counterfactual programs and to find an alternative subsidy schedule that provides greater impact on schooling achievement at similar cost to the existing program.