Large-scale rural-to-urban economic migration in developing countries leaves millions of rural-origin children growing up separated from their migrant parents. Due to the limited parent-child interaction, parental migration poses developmental challenges for left-behind children. This dissertation consists two chapters that study children's cognitive development and parental migration in developing countries.

In the first chapter, I develop a structural model of household migration to understand the effects of parental migration decisions on the dynamics of children's cognitive skill formation from birth until the end of the developmental stage at age 14. I estimate the model using data from the Indonesian Family Life Survey via Simulated Maximum Likelihood. I find that children's cognitive skill formation is sensitive to the duration of parental migration: being left behind for one year during childhood reduces cognitive skill by 0.02 standard deviations per year. Using the estimated model, I estimate a 0.30 standard deviations increase in left-behind children's skills at age 14 had their parents not left them behind. I also simulate a series of counterfactual migration policies that are aimed at improving children's cognitive development. I show that migration policies that incentivize family migration with their children to urban destinations are effective in fostering children's cognitive development: an annual migration subsidy of $150 lifts children's cognitive skills by 0.14 standard deviations at the end of the developmental stage.

In the second chapter, I turn the phenomenon of left-behind children in China. I propose a different set of migration policies due to the unique institutional migration constraint in China. To quantify the impacts of these policies prior to implementation, I specify a static household migration model embedding a cognitive skill production function of their child. I use the model solution to construct a nonparametric matching estimator to directly evaluate the impacts of counterfactual migration policies. The proposed estimation strategy is computationally inexpensive because it does not require to estimate the full structural model, as described in the first chapter. By exploiting income variation using data from the China Family Panel Studies, I find that a non-migration subsidy is most effective in improving children's cognitive achievements when it targets low-income families and younger children. When associated with middle school graduation, the policy-induced change in cognitive achievements translates into an 8.6 percentage points increase in graduation rate for children from low-income households.