Essays on Human Capital and Structural Models

Abstract

Human capital is accumulated over a lifetime, but the opportunities and constraints that shape those investments vary substantially from early childhood through adulthood. This dissertation studies three situations in that life-cycle process in different context: paying for college in the U. S. , measuring and modeling skill formation, and evaluating an early-childhood intervention in Ghana. To that end, I combine applied causal-inference techniques with structural models of individual behavior. Taken together, the chapters provide a unified picture of how finance, measurement and policy design interact to determine the returns to human-capital investment and the distribution of those returns across individuals. The first chapter quantifies how student-loan repayment plans shape college enrollment and field-of-study choices. I build a rich dynamic human capital investment model in which risk-averse high-school graduates decide whether to enrol, what major to pursue, how much to borrow each year, and how much to work while in school, all under uncertainty about tuition support, graduation probabilities and future earnings. The model is estimated with the National Longitudinal Survey of Youth 1997 using an algorithm that allows to accommodate unobserved heterogeneity and a continuous borrowing choice. Counterfactual simulations show that replacing the standard 10-year repayment plan with the 2023 "Saving on a Valuable Education'' (SAVE) income-driven repayment plan raises college graduation by two percentage points overall, and by 18 percent for low-income students, principally because SAVE insures borrowers against uncertain future earnings risk. The policy also changes labour-supply while enrolled, increases average debt at graduation and induces 15 percent of students (disproportionately from low-income backgrounds) to switch majors toward fields such as Education and Health that offer higher non-pecuniary returns or longer completion times. In the second chapter I address an identification problem when estimating technologies of skill formation. Reliable evaluation of early-life policies requires credible estimates of the technology that maps investments and existing skills into future skills. Because both skills and investments are unobserved, researchers rely on multiple noisy proxies of the latent factors. I show analytically and with Monte-Carlo experiments that the standard practice of fixing one measure per factor to estimate those measurement systems leaves key production-function parameters unidentified. Building on recent work, I propose a new identification strategy for a two-skill Cobb-Douglas production function: combine the "age-invariance'" assumption with a variance-standardization rule that re-scales all latent factors into contemporaneous standard-deviation units. This approach recovers all productivity parameters and total-factor productivities up to an interpretable scale that is comparable across studies. Finally, the last chapter evaluates the Quality Preschool for Ghana program, an early-childhood intervention aimed at boosting classroom quality and children's skills. Using the identification strategy developed in Chapter 2, I estimate a latent-factor skill-formation model. The analysis shows that, once noise is removed and the focus shifts to the latent factors of interest, the program produces lasting gains in both cognitive and socio-emotional skills, even though raw scores appear to fade-out. It also shows that the treatment operates by providing an initial productivity boost rather than by changing the way skills develop over time.

Date of Award	10 Sept 2025
Original language	English
Supervisor	Joan Llull Cabrer (Director)