Document Type

Working Paper

Publication Date



This paper presents a multi-period, dynamic programming model of household choices on savings, consumption, having children and helping to fund children's education. Data from the National Longitudinal Survey young women cohort are used to estimate the parameters of the model. The full structural model is estimated using a simulated maximum likelihood procedure utilizing the dynamic programming model solution to create simulated data samples from which nonparametric kernel estimators are used to construct the densities in the likelihood. The estimated model is able to match the general trends in the NLS data, particularly as related to the interaction between children, savings and spending on education. The life-cycle paths of these choices suggest that parents do save to help make sizeable transfers to their children, and that making such choices endogenous is important. Furthermore, the parameter estimates indicate that the amount that parents choose to contribute to a child 19s education has a strong impact on the probability that a child attains a college degree, as does the level of education of the parents. Using the estimated model, policy experiments are performed to look at the impact of additional government grants for college education, tax credits for college spending and the creation of tax-free education savings accounts on parental savings, contributions toward education, and the education attainment of children. While all of the policies increase net contributions to children and increase the probability that a child attains a college degree, the grants and education savings accounts are found to be the most effective. In addition, both policies are actually found to have a greater impact on children with less educated parents.