FAS Theses and Dissertations

“Is credit given to where credit is due?” This paper examines how economic exposure to COVID affected household credit behavior in Thailand using loan-level panel data from the National Credit Bureau, combined with remotely-sensed night-time light intensity as a proxy for local economic disruption. I find that more COVID-exposed postal codes experienced significantly worse repayment outcomes, with delinquency rising especially strongly for loans originated during the pandemic. At the same time, total outstanding debt often declined in more exposed areas. However, this decline does not appear to reflect healthier household balance sheets. Rather, it is more consistent with tighter effective access to credit, as more exposed borrowers became more likely to make loan inquiries, while also facing a higher probability that those inquiries did not result in new loans. Further product-level results show that these dynamics were not uniform across loan categories — unsecured products such as credit cards and personal loans exhibit clearer signs of repayment distress, while agricultural credit appears more shaped by policy intervention, with compositional shifts toward smaller-amount loans. In addition, borrowers who entered the pandemic with only unsecured debt appear to be systematically more vulnerable to COVID exposure. These findings suggest that COVID-19 affected household credit markets in Thailand through a combination of various channels, including worsening repayment capacity, tighter effective credit supply, and changes in the composition of lending, with effects that react to the end of government debt relief policy and persisting well beyond the initial COVID period. This paper advances the use of remotely-sensed variables in development economics to proxy economic shock at granular geographic scales and holds implications for financial health monitoring and targeted government policy.

This dissertation is composed of three essays on finance in low-income and emerging market economies. The essays are unified by a central concern with how external finance shapes economic outcomes, examined through sovereign, cross-country, and firm-level settings. The first develops a general equilibrium model of sovereign default with heterogeneous creditors. Creditor composition affects restructuring dynamics and, through them, equilibrium default probabilities and bond prices. The model provides a consistent account of the observed shift in creditor shares, the rise in default duration and the persistence of modest default frequencies. The second uses loan-level cash-flow data from the World Bank’s private-sector investing arm, the International Finance Corporation, to provide a comprehensive assessment of realized returns to private debt investment in emerging and developing economies. It shows that this portfolio performs at near parity relative to public-market equivalents, though that headline result masks substantial underlying heterogeneity. The third revisits the determinants of foreign-currency borrowing by non-financial firms in emerging markets. In a setting where the domestic supply-side channel emphasized in recent work is absent, it provides detailed loan-level evidence that carry-trade incentives are an important driver of such borrowing and documents a credit contraction following adverse shocks to foreign-currency credit supply.

Economic mechanism design shapes our economic and social systems, silently influencing outcomes for billions worldwide—from auctions powering global online advertising to matching algorithms determining hospital placements for medical residents. Yet, despite decades of significant theoretical advances, classical economic approaches have encountered analytical bottlenecks, leaving fundamental questions unresolved. Inspired by transformative breakthroughs in the application of AI and machine learning to the natural and physical sciences, this thesis pioneers and extends Differentiable Economics, a computational framework that integrates economic theory with deep learning to systematically design optimal economic mechanisms.

Differentiable Economics reformulates economic design as an end-to-end optimization problem, representing auctions, data markets, and matching mechanisms through differentiable neural architectures. This allows previously intractable economic problems to be solved systematically and with practical computational and statistical efficiency. This thesis is organized into three main parts:

In Part 1, Auctions, I propose two neural architectures: RegretNet, a characterization-free method that flexibly addresses the design of complex multi-item auctions, and RochetNet, a characterization-based architecture that explicitly enforces incentive compatibility for single buyer settings. These architectures successfully replicate known optimal solutions, validate important conjectures, and discover entirely new  mechanisms in scenarios where analytical solutions remain elusive. Additionally, I extend the RochetNet framework to a sequential setting by developing a new reinforcement learning approach that outperforms traditional RL methods and analytical baselines.

In Part 2, Data Markets, I design neural network architectures capable of learning signaling schemes for selling information. This framework can effectively handle new economic constraints arising from obedience conditions, modeling the action that an agent will take upon receiving information, and incentive compatibility. This approach not only replicates established theoretical benchmarks but also identifies and validates novel optimal signaling strategies previously unknown in economic theory.

In Part 3, Matching Markets, I introduce permutation-equivariant convolutional neural networks alongside novel differentiable surrogate metrics for stability and incentive compatibility in the design of two-sided matching markets. This combination enables the computational exploration and characterization of previously unknown trade-offs between stability and incentive compatibility.

Taken together, these contributions highlight the potential of Differentiable Economics as a flexible and powerful methodology for economic design. This thesis takes foundational steps towards democratizing economic mechanism design, providing economists and computer scientists alike with new, accessible tools to address long-standing theoretical challenges in optimal economic design.

This dissertation contains three essays in econometrics and international trade. A common theme is the development of econometric tools to support informed policy decisions, with a focus on quantitative trade and spatial models. The first chapter considers sampling uncertainty in quantitative trade and spatial models. Economists use these models to make counterfactual predictions. Because such predictions aim to inform policy decisions, it is important to communicate the uncertainty surrounding them. Three key challenges arise in this setting: the data are dyadic and exhibit complex dependence; the number of interacting units is typically small; and counterfactual predictions depend on the data in two distinct ways—through the estimation of structural parameters and through the description as the status quo. I propose a new Bayesian bootstrap procedure that is tailored to this setting and that addresses all these challenges. The procedure is simple to implement and provides both finite-sample Bayesian and asymptotic frequentist guarantees. I illustrate the practical advantages of this approach by revisiting the applications in Waugh (2010), Caliendo and Parro (2015), and Artuç et al. (2010).

The second chapter considers measurement error in quantitative trade and spatial models. Counterfactuals in these models are functions of the current state of the world and the model parameters. Common practice treats the current state of the world as perfectly observed, but there is good reason to believe that it is measured with error. This chapter provides tools for quantifying uncertainty about counterfactuals when the current state of the world is measured with error. I recommend an empirical Bayes approach to uncertainty quantification, and show that it is both practical and theoretically justified. I apply the proposed method to the settings in Adao et al. (2017) and Allen and Arkolakis (2022) and find non-trivial uncertainty about counterfactuals.

The third chapter, which is co-authored with Isaiah Andrews and Raj Chetty, considers policy choice when both experimental and observational evidence are available. We characterize when and how these sources of evidence should be combined to guide treatment adoption at a new site. We show that the optimal linear predictor for the site-specific treatment effect is a weighted average of the cross-site experimental ATE and the local observational estimate, with weights determined by the covariance matrix of site effects and observational estimands. We provide unbiased estimators for this covariance in settings with both large and small sites, quantify the effect of mismatch between experimental and target sites, and derive easy-to-interpret breakdown points.

This dissertation examines how dominant parties sustain power by entrenching themselves through patronage, clientelism, and institutional manipulation. Despite repeated reform efforts, state weakness persists when political loyalty rather than merit drives bureaucratic appointments, when electoral institutions are distorted by partisan actors, and when ruling coalitions are managed through the strategic allocation of rents. I study these dynamics in Latin America, focusing on Paraguay and Venezuela, to understand how parties tilt the electoral playing field and maintain elite cohesion during crises. Together, these cases reveal the organizational and institutional mechanisms through which dominant parties consolidate power at the expense of state capacity, accountability, and democratic competition.

The first chapter of the dissertation asks: why are developing country bureaucracies ineffective despite repeated modernization efforts? Patronage (hiring on partisanship rather than merit) is often thought to undermine public sector performance, yet it is difficult to measure and usually studied either among frontline agents or supervisors separately, rather than across hierarchical layers. I emphasize patronage hierarchies, instances where layers of the bureaucracy have patronage hires, and measure how they affect state performance. I study this problem using comprehensive data on around 300,000 shipments inspected by Paraguayan customs and develop a method for identifying patronage hires based on political appointment cycles. My findings suggest that patronage inspectors monitored by a patronage port administrator (patronage pairs) exhibit compounded underperformance. They detect less customs fraud and more often fail to detect fraud subsequently identified by headquarter audits, while also deviating more frequently from prescribed random assignment of inspectors to shipments. Additionally, their fraud detection decreases further when dealing with high shipment volumes at their port, as high workloads can mask lower inspection effort. Finally, patronage pairs make smaller tax adjustments than non-patronage pairs, even among comparable products with similar tax evasion risks. Shipments handled by patronage pairs generate around 11% less tax revenue, undermining state capacity.

The second chapter studies how politicians rely on political brokers to buy votes, a phenomenon often found throughout much of the developing world. We investigate how social networks facilitate these vote-buying exchanges. Our conceptual framework suggests brokers should be particularly well-placed within the network to learn about non-copartisans' reciprocity in order to target transfers effectively. As a result, parties should recruit brokers who are central among non-copartisans. We combine village network data from brokers and citizens with broker reports of vote buying, allowing us to use broker and citizen fixed effects. We show that networks diffuse information about citizens to brokers who leverage it to target transfers. In particular, among those citizens who are not registered to their party, brokers target reciprocal citizens about whom they can learn more through their network, and these citizens are more likely to support the brokers' party. Moreover, recruited brokers are significantly more central than other citizens among non-copartisans, but not among copartisans. These results highlight the importance of information diffusion through social networks for vote buying, broker recruitment, and ultimately for political outcomes.

The third chapter inquires whether poll workers' partisanship affect electoral outcomes. Many countries use partisan and adversarial vote-counting systems where poll workers are party representatives and mutual control is expected to provide fairness. Yet in countries with multiple parties, these often have de facto unequal capacities to send representatives to all booths. Exploiting quasi-random assignment of voters to booths in Paraguay's 2018 general elections, we estimate that partisan poll workers decrease an opposing party's vote share by up to 1.1 percentage points (pp) and increase theirs by up to 0.7 pp. Our analyses also expose differential effects according to the electoral system. In proportional representation races, established parties have more opportunities to increase their support at the expense of smaller parties. In contrast, single-winner plurality races dampen this effect due to the winner-take-all aspect of these races. Our results inform the institutional design of vote-counting systems and highlight that adversarial election administration can fail when parties differ in operational capacity.

Finally, the fourth chapter investigates whether leaders court or cut the entourage of sidelined elites. We examine the case of Rafael Ramirez, Venezuela's former Oil Czar, who was purged from the Cabinet in late 2014. We find that Ramirez-affiliated individuals and firms became discretely less likely to receive government appointments and contracts upon his purge. Effects on appointments are greatest for high-spending agencies, and firms affiliated with the military and with Nicolas Maduro gained access to government contracts. Downstream agents share the fortunes of their patrons.

This dissertation presents three essays in development economics that address the relationship between social exclusion and poverty. Chapter one asks: What are the causal effects of new social relationships, and do returns to link formation vary within versus across social class? I conduct a field experiment among women in rural Malawi who, due to patrilocality, relocate to their husband’s home village, thereby disrupting their preexisting social ties. I randomly assign low socioeconomic status (SES) female migrants to receive a list of neighboring women willing to receive meal invitations and the opportunity to send them invitations. I experimentally vary whether the names appearing on the list are all low-SES, all high-SES, or both. I cross-randomize the invitation-sending arms with an in-kind meal subsidy, which weakens financial constraints to meal-sharing. I find a substantial response to the invitation list: 81% of women send invitations across all treatment arms, suggesting that information and effort frictions are important barriers to building networks in this setting. The meal subsidy encourages sending invitations to both high- and low-SES women among participants who can send invitations to both, instead of only inviting low-SES women. I show that there are substantial but differential returns to cross- and same-SES linkages: participants with the opportunity to invite higher-SES guests report being 50% more likely to earn income from self-employment and a 0.21 SD improvement in food security, while women with the opportunity to invite guests within-SES experience a 0.23 SD (30%) reduction in depression one year later. This highlights a trade-off in network formation, where social ties that offer economic advancement may come at the expense of relationships that improve psychological well-being. Chapter two studies experiential and social learning in the context of health technology adoption. Behavioral change can arise from learning through personal experience or learning from others, but how do these forms of learning interact? We conduct a field experiment on household water chlorination in Pakistan, where a randomized group learns through experience by tracking their children’s diarrhea before and after chlorine distribution. Learning-arm households with learning arm neighbors chlorinate their water at a significantly higher rate for one year after the learning intervention. Their children’s health improves by 0.10 SD relative to all other households receiving chlorine. Neither learning households without learning-arm neighbors, nor non-learning households with learning-arm neighbors, exhibit sustained behavioral change. Chapter three asks: Why are people willing to pay for things that are symbolic? We analyze reactions to the renaming of U.S. geographic landmarks to argue that people lose utility through the way that symbolic change is enacted, even when they are indifferent to the symbols themselves. Descriptively, people living in zip-codes that experience geographic landmark name changes initiated by a Democratic administration are more likely to donate to Republican political candidates and less likely to donate to Democratic political candidates after name removals are announced, but not when the replacement names themselves are announced. In a 2024 survey experiment, all respondents express much more conservative policy preferences when we frame the 2015 (Democratic administration) renaming of Mt. McKinley to Mt. Denali as a process enacted by and benefiting groups of people, rather than a passive occurrence. Emphasizing the Indigenous origins of the name “Denali” does not generate the same response. In a 2025 survey experiment, White respondents behave similarly, reporting weakly more conservative policy preferences when we frame the 2025 (Republican administration) renaming of Mt. Denali to Mt. McKinley as an active rather than passive process, while non-White respondents express much more liberal policy preferences in 2025 than 2024, flipping the sign of the “active voice” treatment effect. These results suggest that how people interpret the process of symbolic change depends on the interaction between personal identity and the identity represented by the symbolic change.

Despite major advances in healthcare access worldwide, ensuring delivery of high-quality care remains a central challenge across all health system contexts. In both low- and middle-income and high-income settings, systemic challenges, ranging from inadequate resources and training support to time pressures, burnout, and fragmented care processes, undermine health workers’ ability to deliver timely, evidence-based care. These structural barriers contribute to gaps in delivering high quality care, care that is effective, safe, and people centered. Addressing these structural barriers and achieving better health outcomes requires strengthening the capacity, organization, and working conditions of the health workforce. This dissertation investigates how the characteristics and organization of the health workforce shape the quality of healthcare delivery across diverse contexts. The first two papers focus on maternal health in Kenya, a setting where facility-based deliveries have expanded rapidly but quality of care remains poor. Using a unique dataset combining direct clinical observations, provider interviews, and facility assessments, Paper 1 examines how factors such as facility resources, provider attributes, and workload affect the quality of labor and delivery care. Paper 2 then evaluates whether strengthening obstetric and newborn care skills through a large-scale nurse mentorship program can improve the quality of care. Leveraging data from a cluster-randomized trial, it provides evidence on how mentorship impacts provider knowledge, delivery care quality, and maternal and neonatal outcomes. Paper 3 shifts to a high-income setting, focusing on a different, yet pressing set of challenges. Although high patient volumes and limited time per encounter are challenges common to many health systems, primary care in the United States has undergone changes over recent decades such as increasing administrative demands and provider shortages that have substantially constrained the time available for each visit. These pressures potentially shape both the quality and comprehensiveness of care. Using electronic health record data and quasi-experimental variation in visit scheduling, this paper examines how reductions in visit time pressure influence the content and quality of primary care, including the delivery of preventive services. This dissertation provides evidence on how the organization, capacity, and environment of the health workforce shape the delivery of care. The findings contribute to a deeper understanding of how future initiatives might be designed to strengthen quality of care through the health workforce.

This dissertation investigates how households acquire information, form expectations, and make financial decisions in the presence of limited attention and reasoning. The first chapter documents selective inattention to interest rates. A combination of existing and newly-designed surveys shows that households close to durables purchases actively acquire more information about interest rates and have more accurate expectations. These empirical findings motivate a heterogeneous-agent model with endogenous attention to interest rates. Selective inattention shifts the composition of spending responses to interest rate cuts, accelerates the impact of larger cuts, and generates dampened responses to changes in volatility that are closer to empirical evidence. The second chapter introduces a theory of shallow thinking, where agents reason through only a limited number of steps in the causal structure of the economy. New survey data show that households mentally process just 2.6 steps of economic propagation on average. When incorporated into New Keynesian and real business cycle models, shallow thinking accounts for several empirical puzzles, including the asymmetric response of long-term interest rates to different shocks, the predictive power of inflation expectations for bond returns, and persistent output responses to productivity shocks. The third chapter studies how and why households adjust their consumption, saving, and borrowing in response to transitory income shocks. Drawing on new large-scale survey data, the chapter quantifies intertemporal marginal propensities to consume and deleverage and finds significant heterogeneity across households. Standard socioeconomic characteristics explain only a fraction of this variation; psychological factors, past experiences, and expectations improve explanatory power substantially. A machine learning approach identifies four behavioral types, highlighting that similar financial decisions often arise from fundamentally different motivations.

Medicare home health care accounts for over $18 billion in annual spending and serves 3.5 million beneficiaries per year, with demand growing alongside population aging and a broader shift toward community-based care. Yet key features of the program remain poorly understood. It is unclear whether home health substitutes for more expensive post-acute care or instead adds to Medicare spending. And despite for-profit agencies accounting for roughly 83 percent of the market, the implications of ownership for utilization and program value—particularly in a setting with concerns about overuse and fraud—remain understudied. This dissertation examines how ownership structure, payment incentives, and provider supply shape utilization and spending in Medicare home health care, a program that serves a broader and more heterogeneous population than is typically recognized, with community-admitted patients comprising nearly half of all spending and episodes. In the first chapter, I study Arkansas's 2016 sale of its statewide public home health agency network to a national for-profit chain. The standard concern with privatization is underprovision: private firms cutting costs in ways that reduce quality or restrict access to unprofitable populations. I document a different mechanism. When reimbursement is endogenous to provider behavior, as under prospective payment systems that reward coding intensity and service mix decisions, privatization may instead induce overprovision, raising spending without improving care. Difference-in-differences estimates using Medicare administrative data from 2010–2021 show that the ownership transition increased statewide home health use by roughly 10 percent and effective reimbursement per episode at treated agencies by 27 percent, with no evidence of reduced access for rural or high-cost populations. Spending rose substantially while measurable quality and outcomes remained flat. These findings suggest that the consequences of privatization depend critically on the institutional environment: when payment systems allow providers to influence their own reimbursement, overprovision rather than underprovision may be the first-order concern. In the second chapter, I leverage variation in home health agency entry and exit across time and local areas to examine supply sensitivity and its downstream consequences. Distinguishing between post-acute and community-entry patients reveals stark differences in responsiveness: changes in agency supply have minimal effects on post-acute care use but substantially increase community-entry use. Beyond these direct effects, expanded access generates complex shifts in broader healthcare utilization operating through both substitution and complementarity channels—reducing hospitalizations and Medicaid-funded nursing home stays while increasing prescription drug use, physician and other Part B services, and hospice use. The net increase in total Medicare spending suggests that expanded home health access represents an addition to Medicare services rather than pure substitution, challenging the conventional view of home health as primarily a substitute for institutional post-acute care. In the third chapter, I directly characterize the community-admitted population that the first two chapters show to be central to understanding home health spending and supply responses. Using data from 2008–2021, I document that community-entry users are older, more often dual-eligible, and have greater cognitive impairment than post-acute users, and are more frequently served by for-profit agencies. Community-entry users also use home health more intensively than post-acute users, and community-entry prevalence varies widely across states, explaining 60 percent of cross-state variation in per-capita Medicare home health spending. Despite representing a distinct population with unique needs, Medicare takes a largely one-size-fits-all approach to home health policy, suggesting value in tailoring reimbursement, benefit design, and regulation accordingly.

This thesis examines the implications of market frictions in international finance and macroeconomics in three contexts. The first chapter documents the effect of trading relationships on client trading outcomes in the over-the-counter (OTC) foreign exchange (FX) derivatives market. The second chapter documents the effect of nominal wage setting frictions on employment. The third chapter examines the behavior of non-U.S. central banks when firms engage in currency mismatch, borrowing more in dollars than given by their dollar operating exposures, emphasizing how imperfect regulation may affect U.S. dollar interest rates.

In the first chapter, joint with Gerardo Ferrara, I study whether clients that rely more heavily on a dealer in the OTC FX derivatives market have worse trading outcomes after the dealer is adversely shocked. Using granular transaction-level data, we document that trading relationships are persistent—in an active trading week, clients are more likely to trade with a dealer that they had a relationship with and relied on more heavily. Then, we exploit the March 2023 collapse of Credit Suisse as an exogenous shock to exposed clients’ set of trading alternatives when relationships are persistent. Using difference-in differences analyses, we find that, although Credit Suisse’s EURUSD notional traded and trade count declined, clients that relied less heavily on Credit Suisse did not differentially reduce their Credit Suisse-specific trading activity relative to more reliant clients. Instead, more reliant clients continued trading at the client level and increased activity with other existing dealer relationships without incurring additional costs, relative to less reliant clients. These findings suggest that search and bargaining frictions were not particularly costly for heavily reliant clients after the shock—relationship persistence did not differentially prevent them from reallocating activity to existing alternative dealers, or lead to relatively greater costs, when their relationship dealer came under stress.

In the second chapter, joint with Gert Bijnens, Hugo Monnery, and Laura Nicolae, I empirically document the effect of wage changes, driven by wage indexation to inflation, on firm-level employment growth. In Belgium, nearly all employees’ wages are indexed to inflation and firms are grouped into labor agreements that determine the exact timing and frequency at which wages are indexed, e.g. every year or every month. Using firm-level administrative data, we estimate two-stage least squares regressions of firm-level employment growth on wage growth, instrumented by the wage growth implied by the firm’s indexation policy. We find that employment contracts by 0.4% over four quarters for each 1% increase in wages. This result is robust to including NACE sector-date fixed effects and to using only variation in firms’ indexation timing, controlling for their chosen indexation frequency. About one-third of the response comes via anticipation of future wage increases. The elasticity is more than twice as large in magnitude in the post-pandemic period than before it, suggesting strong nonlinearities. Overall, these results show that, by preventing inflation from reducing real wages, inflation indexation reduces employment.

In the third chapter, joint with Mitali Das, Gita Gopinath, Taehoon Kim, and Jeremy Stein, I document an externality of central banks’ imperfect regulation of firms that engage in currency mismatch, which results from central banks’ dollar reserve accumulation decisions. We explore how foreign central banks behave when firms engage in currency mismatch. Using a panel of 56 countries, we document that central bank holdings of dollar reserves are correlated with the dollar-denominated bank borrowing of their non-financial corporate sectors. Then, we build a model in which the central bank can deal with private-sector mismatch, and the associated risk of a domestic financial crisis, by: (i) imposing ex ante financial regulations; or (ii) accumulating dollar reserves to serve as an ex post dollar lender of last resort. The model highlights a novel externality: individual central banks may over-accumulate dollar reserves, relative to what a global planner would choose. Under imperfect regulation of currency mismatch, individual central banks do not internalize that their hoarding of reserves exacerbates a global scarcity of dollar-denominated safe assets, which lowers dollar interest rates and encourages firms to further increase the currency mismatch of their liabilities. Relative to the decentralized outcome, a global planner may therefore prefer higher capital requirements and reduced holdings of dollar reserves.