Job Market Paper
The timing of human capital specialization varies considerably across the world. While early specialization leads to depth in field-specific knowledge, late specialization may help students find which field they like by maintaining exposure to a greater breadth of options. This paper quantifies the change in student welfare when switching from a depth-oriented to a breadth-oriented system. To do so, I leverage a policy experiment in the country of Georgia. The policy imposed generalist high school (HS) graduation exams and forced the students to generalize more than they used to. Compared to the pilot cohort, the cohort that began HS with advance notice of the policy demonstrated broader subject knowledge, applied to more diverse fields in college, and was subject to changing competition for college programs across fields. To quantify the welfare impact of the policy, I develop and estimate a dynamic structural model of gradual specialization in HS. The key ingredients of the model are endogenous skill formation, field match uncertainty, and learning by experimentation. Policy simulations with estimated parameters and observed changes in competition across fields indicate that the policy increased average HS graduate welfare by at most an amount equivalent to 0.6% of the present discounted value of average wages. However, policy welfare effects are not uniform: students with higher general academic ability enjoy moderate utility gains but students with lower general academic ability experience significant losses.
I investigate whether learning by college admission exam retakers reduces the initial socio-economic skill gap between applicants. I estimate learning for urban and rural retakers in the country of Georgia using rich administrative panel data and a novel semi-parametric estimator. The proposed three-step estimator is robust to self-selection on latent skills and test measurement error into retaking. Furthermore, the estimator attains much greater precision than conventional two-step correction methods by using data on multiple test scores per applicant. I find no evidence of catch-up, emphasizing the context- and method-dependence of previous conclusions about the background-neutralizing nature of retaking.