I investigate the contribution of higher education to the development process. I leverage a 1996 education reform in Brazil that reduced entry costs to private colleges. The reform's impact on educational attainment varied across local labor markets. I estimate this heterogeneous effect through an IV strategy exploiting the idea that private colleges targeted locations under-served before the reform. Rising college attainment causes income growth, a decline in agricultural employment, and an increase in the prevalence of larger firms in the local economy. To quantify the macroeconomic effects of rising educational attainment, I propose a structural model featuring an educational choice and a sectoral specialization of educated workers in non-agricultural activities. A key implication of the theory is that increasing college attainment facilitates firm expansion and improves productivity in the non-agricultural sector. I estimate the model's structural parameters and develop an identification strategy that enables me to disentangle changes in demand for skills from changes in access costs to college. I show that the model can quantitatively replicate the 2SLS coefficients from the empirical analysis. Finally, I use a counterfactual analysis to show the reduction in college costs accounts for 18% of Brazil's GDP per capita growth and 7% of the decline in agricultural employment between 2000 and 2010. Labor markets with faster growth in skill-bias technology benefit relatively more from reducing college costs.

In this paper, we study how demand-side subsidies interact with the equilibrium price and quality level in Brazil's higher education sector. More precisely, we consider two policies that assist low-income students in attending private institutions: scholarships and subsidized student loans. First, we develop a quality measure for undergraduate programs using value-added models, where a student’s post-graduation outcome is determined by their pre-enrollment characteristics. To do so, we link multiple administrative datasets to track individual students before enrollment, during college, and after college. We consider two post-graduation outcomes: a standardized “exit” exam, which tests students’ major-specific knowledge, and income from a matched employer-employee database. We document key patterns and correlations of our quality measure and extensively validate it. Next, we develop a static equilibrium model of demand, pricing, and quality provision. We consider two counterfactuals: decreasing the supply of loans by 10% and decreasing the supply of scholarships by 10%. We find that decreasing the supply of scholarships by 10% reduces the enrollment among subsidized students by 8% and decreases the quality provided by private colleges, with a median change of -5% in value-added. There is substantial heterogeneity in the results. The reduction of subsidies decreases the college enrollment of relatively high-ability students who come from poor families. For-profit programs are the big losers from the policy experiment. Importantly, reducing subsidies indirectly affects all remaining college students because colleges invest less in quality.

In many developing countries, corruption is a pervasive phenomenon, widespread across districts and local officials. This paper studies the impact of corruption on the spatial distribution of economic activity and its dynamic effects on local and aggregate growth. Our investigation focuses on a federal policy in Brazil that randomly selected local governments for audits on the use of public funds received through transfers. While evidence suggests this program effectively reduced corruption and enhanced political accountability, its implications for firms remain less understood. For example, diminishing corruption could optimize the allocation of procurement contracts by prioritizing efficiency over political connections, fostering competition. Building upon Colonnelli and Prem (2022), we use a difference-in-differences analysis to reveal the positive impact of corruption reduction on local economic activity. As all eligible municipalities were aware of the policy, this approach captures the relative effects of audits on firm outcomes. To discern the policy's aggregate effects, we develop a spatial model wherein firms' entry decisions and choice of production locations are endogenously determined. Variations in corruption levels influence relative productivity and potentially lead to misallocation. In our model, audited municipalities witness a more significant decrease in corruption, creating favorable conditions for business initiation. We derive equations from the model that directly correspond to the empirical difference-in-differences coefficient. This relationship between the model's structural parameters and empirical findings enables us to estimate the upper and lower bounds of the policy's aggregate impact. 

I explore the benefits of forecasting technologies in a setting where risk sharing, as a response to uninsured risk, is prevalent for consumption smoothing. I add information technologies into a limited commitment model where households participate in cross-village transfers. Forecasting technologies provide households with a new assessment of the probability distributions of the future realization of weather, which is relevant to production. Forecasting technologies correctly predict the actual weather realization more frequently than the conventional methods used by rural households. Access to accurate weather forecasts allows households to increase the average returns of their production investments (i.e. crop choice and fertilizers, among others). This reduces the value of participation in the risk-sharing agreement. On the other hand, the investment decisions might expose rural households to more considerable losses when the forecast is wrong. The second channel increases the value of remaining in the risk-sharing agreement. I explore the interaction between long-run weather forecasting technologies and consumption mechanisms in the data. I focus on the case of rural villages in India. I use seasonal weather forecasts of the monsoon season, detailed monthly information on agricultural investments, and rainfall data on a set of rural villages collected in the ICRISAT database. The forecast predicts the monsoon weather for four aggregate regions in India. There is important heterogeneity across villages in the accuracy of the forecast. Rosenzweig and Udry (2019) find that this forecasting technology is uncorrelated with important determinants of agricultural investments once we focus on the sample of villages for whom the correlation between the forecast and the observed rainfall realization is positive. Because the seasonal weather forecast accuracy differs across villages, so do the benefits of using the forecasting technology. I use panel data of farmers in India, where I observe, for many years, different forecasts and rainfall realizations to assess various dimensions of household decisions. First, I observe households inform themselves about the weather in those villages with accurate weather forecasts suggesting the use of the forecasting technology. Second, when the forecast is optimistic, I investigate whether farmers in locations with high accuracy of the forecasts increase investments in more profitable but riskier crops. Third, I consider the implications for consumption smoothing. Preliminary evidence suggests that income shocks, driven by rainfall realization, affect consumption more in locations with more accurate weather forecasts. Conditional on the accuracy of the forecast and the rainfall realization, preliminary evidence suggests that households participate less in risk-sharing, meaning households send or receive fewer transfers as a share of their income. This is a work in progress.