Papers accepted for presentation

This paper develops a search-and-matching model of the labor market with affirmative action in public-sector hiring, and estimates it using Brazilian data to study the labor-market effects of racial hiring quotas. Workers face a discriminatory wedge in the private sector, while the public sector — split across municipal, state, and federal government — may reserve a share of its vacancies for the discriminated group. The direct effects are characterized in closed form. A reservation works by lowering a hiring standard on the reserved share, not by transferring positions between groups. When a quota's fiscal cost is financed through a payroll tax, no configuration of quotas raises aggregate worker welfare above the no-quota baseline. However, if the planner places weight on the distribution of gains across race and education, a large quota becomes optimal. This optimal quota is concentrated entirely at the federal level — 41% to 62% of federal vacancies, well beyond anything Brazil has legislated. Federal government's advantage lies in its wage schedule, since the transfer a reservation delivers scales with the public wage while its cost does not. Since the three levels of government draw on the same economy-wide outside option, quotas on different spheres act as substitutes rather than complements. The welfare gain itself reflects correcting private-sector discrimination: absent it, no quota would be optimal.

Racial minorities are underrepresented among firm owners in Brazil. One potential explanation for this gap is unequal access to high-quality college education, which may limit the acquisition of skills, credentials and networks necessary to start and sustain a business. If owners’ identities and networks are relevant for hiring and compensation decisions, this under-representation can further contribute to persistent racial inequality within firms. This paper provides quasi-experimental evidence on whether expanding non-white access to higher education affects the racial composition of entrepreneurs and the characteristics of the firms they create. We exploit Brazil's Law of Quotas, which mandated race-based admissions policies in federal public universities, and compare programs where quotas were binding in increasing non-white enrollment to programs where they were non-binding, as well as to institutions unaffected by the law, using an event-study design. We find that exposure to binding quotas increases the number of non-white alumni who become firm owners, reduces racial wage gaps within firms owned by exposed alumni and leads to firm creation in census tracts with larger non-white population shares. We also examine municipality-level impacts among alumni from both private and public colleges. The evidence points to an overall increase in entrepreneurship, consistent with expanded university access for non-white students without a corresponding displacement of white students from higher education.

How well can organizations allocate tasks when individual performance is only imperfectly observed? We develop a simple framework that exploits quasi-random personnel exits to infer how much organizations attend to differences in worker quality when assigning tasks. We apply this framework to health care, a setting with life-and-death stakes where quality is particularly difficult to assess. Using rich administrative data from Brazil, we show that hospitals react to the unexpected death of higher-performing physicians by shifting patients toward higher-quality physicians. This adjustment is sharply asymmetric: losses from high-performer exits are largely offset, whereas exits of low performers lead to nearly random reassignment. Quantitatively, targeted reassignment prevents a 20 percent increase in mortality that would otherwise arise under random reassignment after the exit of a top physician. The pattern is driven almost entirely by shifts among incumbent physicians rather than external hiring, becomes stronger with the de-ceased physician’s tenure, and is most pronounced in hospitals with formal oversight structures and labor slack. These findings suggest that organizations attend to differences in quality, but that organizational capacity is crucial for responding effectively to such differences.

We study Brazil’s 2023 National Nursing Wage Floor, an occupation-specific mandate directly enforceable in public and eligible non-profit providers but not in private for-profit firms. Using matched employer-employee data covering the universe of formal employment, we combine a market-level GAP design with a wage-bin design. We find large wage gains in targeted sectors, including spillovers to workers already earning above the floor, with no evidence of employment losses. Wages also increased for nursing workers in non-targeted private firms. These cross-sector spillovers were substantial but remained bounded by occupation, with no detectable effects on other health professionals in the same establishments. Spillovers were stronger in more exposed labor markets and among firms covered by collective bargaining agreements that referenced the wage floor. We find little evidence that they varied systematically with firm market share or were driven by worker reallocation toward targeted sectors. Beyond the labor market, we find no effects on inpatient mortality over the short post-reform horizon, but private health insurance prices increased more in highly exposed markets. Overall, wage shocks can cross firm and sector boundaries while remaining concentrated within occupational labor markets.

Can laws designed to protect democracy inadvertently compromise public safety? We examine whether temporary restrictions on arrest enforcement during elections increaseintimate partner violence. We exploit a distinctive feature of the Brazilian Electoral Code that restricts the arrest of eligible voters during a statutory period surroundingelection day. Because only municipalities holding a runoff election are exposed to an additional restriction period, the law generates plausibly exogenous variation in arrestenforcement across municipalities. Using municipality-day administrative data covering all Brazilian mayoral elections between 2012 and 2024, we estimate the effect of thisadditional exposure on reported intimate partner violence. We find that municipalities exposed to this additional restriction window experience a 17.5 percent increase inreported intimate partner violence, while broader measures of violence against women remain unaffected. The effect is concentrated in the days preceding the election, beforevoting takes place, and is absent in placebo windows, supporting a deterrence-based mechanism rather than one driven by election-day events or electoral outcomes. Ourfindings uncover an unintended trade-off between protecting electoral integrity and preserving public safety, highlighting how temporary legal constraints on law enforcementcan disproportionately affect vulnerable populations.

This paper investigates how reforms aimed at fiscal sustainability change the targeting and the effects of student financing programs. We exploit two reforms implemented in Brazil's Student Financing Fund (FIES) and use a regression discontinuity design generated by the admission process cutoff scores to estimate the causal effects of financing on access, completion, and earnings. Student financing substantially increases access to higher education, but in the long run its effects on completion and earnings vanish, except for the most financially constrained applicants. Overall, non-recipients find alternative ways to enroll and eventually obtain their degrees. We also show that the reforms began to channel financing toward programs with lower default rates, changing the composition of beneficiaries. The results reveal a potential conflict between the fiscal sustainability of student loan programs and their role in easing financial constraints on access to higher education.

We study how dominant private employers reshape small towns and their surrounding regions. Using administrative records for Brazilian municipalities from 2013 to 2022, we identify 582 anchor-employer events in which a private establishment becomes both large and locally dominant. With a staggered difference-in-differences design, we find that formal employment rises by 10.8 percent and output by 7.0 percent in host towns. Those gains remain concentrated in the anchor establishment, with null effects on employment elsewhere in that town and the number of local firms. In this rural setting, agriculture is an alternative margin of adjustment, and we also find local agricultural productivity rising by 7.7 percent, through physical yield rather than crop prices. This intensification happens without a detectable expansion of the agricultural frontier. Outside the host town, formal employment falls by 1.9 percent in exposed neighbors that never receive an anchor, yet rises by 1.45 percent for the microregion as a whole. At the same level, agricultural productivity also increases by 2.85 percent, indicating a broader regional production response, rather than zero-sum reallocation or the conventional local employment multipliers associated with large plants.

This paper studies how electoral rules shape candidates’ digital campaign strategies, exploiting Brazil’s 200,000-voter runoff threshold in a regression discontinuity design. We test whether candidates who anticipate a potential second round reallocate their digital spending intertemporally, preserving resources for the runoff at the expense of first-round exposure. We combine official campaign finance records from the 2024 municipal elections with highfrequency Instagram advertising data from Meta’s Marketing API — 57,677 ads matched to 540 candidates. We show three patterns of digital campaigning: ads are launched within hours of being created, candidates favor broad demographic reach over micro-targeting, and advertising intensity escalates sharply as election day approaches — serving as a tool for rapid and broad-reach voter outreach rather than for sustained or targeted communication. Runoff-eligible candidates reduce their digital boosting budget share by approximately 4 p.p. in the first round — roughly USD 4,200 less in digital expenditure — with the effect concentrated in the final weeks and strongest among runoff-viable candidates, for whom a secondround contingency is most relevant.No other expenditure category responds to the threshold. The reallocation operates through scale rather than content — candidates reach fewer voters without altering the format, duration, or text of individual ads — consistent with digital advertising serving as a strategically flexible instrument whose near-zero marginal cost enables intertemporal allocation decisions that traditional media cannot accommodate.

Broadband Internet has been shown to transform employment through skill- and task-biased mechanisms. Much less is known, however, about whether this technology alters the internal organization of firms, particularly their managerial span and hierarchical structure. Using novel Brazilian geocoded data on plants and telecommunications infrastructure, we study the effects of broadband Internet availability on firm reorganization. To establish causality, we exploit a physical constraint on internet connectivity and the staggered rollout of the technology across locations in a difference-in-differences framework. In contrast to the effects documented for robots and other digital technologies---often associated with stable or increasing managerial span---we find that broadband Internet leads to a decline in span of control. This reorganization is accompanied by a leaner employment structure, characterized by a relative contraction in lower-level nonmanagerial employment compared with control plants, no significant changes in average wages across nonmanagerial layers, with managerial wage growth primarily concentrated in smaller firms. Although treated plants do not increase total employment, larger firms expand through the creation of new affiliated plants across locations, which partly dampens the decline in span of control observed in incumbent plants when measured at the firm level. On net, however, employment destruction in incumbent plants largely offsets job creation in newly established units. Our findings are broadly consistent with broadband Internet reducing communication costs and facilitating the upward escalation of exceptions to managers, as predicted by knowledge-based theories of hierarchies.

Motherhood can reshape women’s daily mobility well beyond their journeys to work. Using data from the São Paulo Origin-Destination Survey and a pseudo-event-study design following Kleven (2025), this paper examines the child penalty in mobility in Brazil. We show that care-related trips rise substantially after the birth of the first child, while work-related trips decline (a 38% long-run penalty in work-trip duration and a 30.8% decline in the number of work trips). We find no evidence that motherhood reshapes trip chaining or work-trip distance and speed, but care-related and overall trips become shorter and slower after childbirth, indicating that the adjustment to caregiving operates through both the volume of trips and, for care-related travel, a shift toward more localized destinations. These findings extend the child penalty literature beyond labor market outcomes and work commuting, showing that motherhood reshapes the broader structure of women’s daily lives.

Climate change can weaken political voice not only by changing voters' preferences, but also by changing who is able to vote. Using georeferenced polling-station records from eight Brazilian elections (2010–2024), matched to daily river discharge, we show that historically low river discharge on election day increases voter abstention in communities that depend on river transportation. The participation loss is concentrated in sections with higher illiteracy rates and among married voters, thereby shifting the electorate's composition. These shocks reduce the vote share of parties whose bases overlap with affected populations, redistributing political representation along the lines of climate exposure.

On weakly governed frontiers, land is often illegally occupied. This paper asks whether securing land against this process can reduce modern slavery. We study the Brazilian Amazon, where these frontier activities rely on geographically isolated workers with few outside options. We argue that these activities can sustain demand for vulnerable workers who are kept in slavery conditions. By removing land from the pool of areas available for private appropriation, formal protection can weaken this coercive frontier economy. We test this hypothesis by exploiting the staggered expansion of indigenous lands and ecological protected areas across municipalities of the Legal Amazon from 1997 to 2010. Our empirical strategy combines the timing of protected-area expansion with federal slave-labor inspection records and individual-level data of rescued workers. We find that protection strongly reduces the prevalence of slave-labor and the number of workers released. The effect is concentrated in cattle ranching and extractive activities, the sectors most closely tied to land occupation and forest conversion. Additionally, land-cover regressions show less pasture expansion and greater forest retention in newly protected municipalities, consistent with property rights curbing those frontier activities.

This paper studies how affirmative action shapes disadvantaged students' higher education choices. The Quota Law, gradually introduced in Brazil's Centralized Admission System (SISU), reserved an increasing share of federal university seats for public high school students. I exploit this variation to estimate an event-study difference-in-differences, using linked SISU and ENEM microdata covering 10 million student-year observations. Newly eligible students become significantly more likely to apply to competitive majors and STEM fields, and shift toward higher-quality universities, confirming that constrained opportunity sets, not just preferences, shaped their prior choices. Yet this upgrading is driven almost entirely by higher-ability and higher-income students within the eligible group; lower-ability and lower-income students see smaller, less persistent application gains, as academic preparation remains a binding constraint even once eligibility expands. Expanding access thus relaxes one constraint on disadvantaged students' choices, but it does not, on its own, level the playing field.

Does the size of a jurisdiction influence democratic participation in developing countries? While most quasi-experimental studies focus on developed nations with consolidated municipalities, where mergers usually lead to lower turnout, our research examines territorial fragmentation in Brazil, where the number of municipalities increased by 24% from 1990 to 2013 under compulsory voting. Using a new dataset of 621 fragmented areas and 605 failed emancipation requests from regional electoral courts and legislative bodies across 19 states, we employ a difference-in-differences approach with inverse probability weighting. Results show voter turnout rose by 1.95 percentage points in municipal elections and 2.19 percentage points in national elections, with the increase driven mainly by valid votes rather than blank or null ballots, consistent with engagement beyond rule-following, though the design cannot rule out candidate-driven mobilization. The effects are stronger in areas with more fragmented jurisdictions, especially in poorer, rural, and remote regions, suggesting fragmentation mobilizes voters where state presence is weaker. Because turnout increased similarly in national elections, where local splits do not affect voting weights, a simple pivotal-probability explanation is insufficient. The findings imply that multiple factors influence voting behavior, which our method cannot isolate. Overall, the study contributes to debates on decentralization in developing democracies: while municipal consolidations tend to lower turnout, creating smaller jurisdictions can boost participation, even in compulsory-voting systems with already high baseline turnout.

Distinguishing taste-based fromstatistical discrimination in elections is challenging because voters may oppose female candidates either because of their gender per se or because of what they believe about their ability and policies. We combine a flexiblemodel of voting, featuring vertical differentiation in ability and horizontal differentiation in gender and policy, with a large-scale electoral experiment. Applying the approach to Brazilian elections, we find evidence of both forms of discrimination. We also show that conventional reducedformestimates fail to recover treatment effects in this setting. Finally, we use the estimated model to evaluate counterfactual campaigns designed to reduce discrimination.

Infrastructure rollouts are increasingly used as sources of quasi-experimental variation to study the effects of digital connectivity, transportation, energy, and other large-scale investments. However, treatment assignment in these settings is generated by institutional processes rather than random allocation, raising important questions about the credibility of causal claims. This paper proposes the \emph{Causal Credibility Framework} (CCF), which integrates institutional analysis with econometric evidence to assess the credibility of causal inference from policy-driven infrastructure deployments. We illustrate the framework using Brazil's nationwide 4G rollout, exploiting the staggered expansion of mobile broadband to examine political polarization. We find a robust empirical association between 4G adoption and increased political polarization, particularly when polarization is measured using the Foster--Wolfson index. However, systematic pre-treatment differences across rollout cohorts and sensitivity analyses indicate that the strength of the causal interpretation remains conditional on identifying assumptions. The paper's principal contribution is methodological: infrastructure rollouts should not be evaluated solely by the sophistication of the econometric estimator but also by the institutional mechanisms generating treatment assignment.

Brazil has one of the world's highest cesarean rates, and a large share of them lacks clinical indication. This paper asks whether the Projeto Parto Adequado, a vaginal-birth incentive program run by the Brazilian National Supplementary Health Agency, lowered cesarean rates among adopters. An annual hospital panel from SINASC linked to CNES, covering 2014 to 2022 and restricted to lower-risk pregnancies, identifies the effect with the interaction-weighted estimator of Sun and Abraham, robust to heterogeneity under staggered adoption in three waves. Adoption lowers the cesarean rate by 5.56 percentage points on the spontaneous-labor margin and by 3.66 points in the clinically eligible population with a fixed denominator, two estimands read as a pair. The magnitude repositions what was known: over all births the causal effect is 2.74 points, about one fifth of the improvement the program reports in before-and-after terms, so most of that gain is sector-wide trend and selection. At comparable exposure, the 2015 and 2017 cohorts respond similarly on the spontaneous margin (a 1.59-point difference, p = 0.48), whereas in the fixed-denominator population the effect concentrates in the pilot cohort (a 4.74-point difference, p = 0.002), consistent with weaker implementation intensity as the program scaled. The finding survives treatment and outcome placebos, nine control-eligibility scenarios and a panel extended to 2012, in which the pilot cohort's parallel trends become testable and are not rejected. Under Honest DiD the breakdown value is 0.98 in the pooled reading and 2.94 when calibrated on the informative cohort.

Non-fatal strangulation (NFS) is a form of intimate partner violence (IPV) and a strong predictor of intimate partner homicide (IPH). We compile NFS statute data, link them to homicide reports, 1990–2019, and estimate effects on IPH using a two-stage difference-in-differences estimator for staggered adoption and heterogeneous effects. NFS laws reduce IPH for ages 18–49, by 0.169 female-victim homicides per 100,000 women and 0.090 male-victim homicides per 100,000 men. We find no effects on IPH for ages 50–70 or on stranger-perpetrated homicides. Evidence from reported IPV incidents is suggestive of changes in reported incident classification and arrest patterns.

Do politicians' personal experiences shape the policies they pursue in office? We study how electing a candidate who had experienced a child's hospitalization affects mayoral decisions and local health outcomes. Linking parents and children across millions of Brazilian families over more than two decades to administrative hospitalization records, we use a regression discontinuity design on close elections to compare municipalities where a mayor who had experienced such a shock was narrowly elected with municipalities where a candidate who had experienced it narrowly lost. We find that mayors who had experienced a child’s higher-severity health shock invest more in public health, leading to improvements in local health outcomes. We also find evidence that they are less likely to appoint political supporters to high-ranking positions in the health sector. For lower-severity cases, we find no significant effects. Our findings suggest that personally impactful life events—particularly a child's severe hospitalization—can shape political priorities and generate measurable improvements in public health.

This paper investigates the effects of set-aside auctions for small and medium enterprises (SMEs) in Brazilian public procurement. We exploit a legal reform that made set-aside auctions mandatory for contracts below BRL 80,000. We show that restricting competition to SMEs increases their probability of winning public contracts by 17.9 percentage points, but also raises procurement costs by 12.1%. Given the total procurement spending affected by the policy, this represents an annual fiscal cost of approximately BRL 73 million. This result likely reflects the exclusion of larger and potentially more efficient firms. However, the policy also generates meaningful economic gains for the firms it targets. Using a difference-in-differences approach, we find positive effects on employment and earnings among SMEs, including an average increase of 0.93 jobs per firm. Our cost-benefit analysis suggests that, despite the additional procurement expenditure, set-aside auctions may be a viable policy instrument in certain contexts. The estimated cost per job created is approximately USD 2,800.

Smart grid technologies are viewed as a key component of electricity sector modernization, yet causal evidence on their impacts in low and middle-income countries remains limited. This paper studies the effects of smart meter adoption in Brazil using administrative data from five major electricity distributors between 2021 and 2023. Exploiting the staggered rollout of smart meters across households and distribution networks, we estimate the impacts of smart meters on electricity consumption, bills, non-technical losses, and service reliability. We find that smart meter adoption increases measured electricity consumption by 9.01\% and total bills by 8.43\%, with substantially larger effects among low-income households enrolled in the social tariff program. Higher smart meter penetration also significantly reduces non-technical losses, while the estimated effects on outage frequency and duration are negative but statistically imprecise. To investigate the mechanisms behind these results, we estimate a household electricity demand model that allows price responsiveness to vary by meter type. We find near-zero price responsiveness among households with smart meters, suggesting that the increase in measured consumption primarily reflects improved measurement and formalization of previously unrecorded consumption rather than changes in underlying electricity demand.

This paper examines how an often-overlooked source of pay transparency—the public disclosure of tax information—affects gender wage gaps. We exploit a 2001 change in Norway that made individual tax returns searchable online. Using matched employer–employee data and a difference-in-differences design, we find that within-firm gender wage gaps fell by 2.2 percentage points (8.7 percent), driven by rising female wages. Effects are strongest in private-sector firms, industries with initially larger gaps, and municipalities that previously lacked easy access to printed tax lists. Wage gains are concentrated among job-changing women, suggesting that broad-based transparency mainly operates through improved information for job search.

Hydropower-dependent electricity systems are increasingly exposed to drought risk. This paper studies one consequence of that exposure in Brazil: when droughts reduce the availability of hydropower, the system operators increase electricity generation from fossil fuels, and the resulting air pollution exposure raises urgent respiratory morbidity. We combine data on plant-level generation, hydrological conditions, and ERA5 wind fields with hospitalization records from the public health system SUS. We first show that droughts reduce hydro inflows and reservoir storage and that lower hydropower generation increases generation from coal, oil, and natural gas. We construct a drought-predicted transported fossil generation exposure measure by combining predicted thermal dispatch with source-receptor weights based on distance and wind alignment. We then estimate how increases in power generation from fossil fuels affect urgent hospitalizations in nearby areas. We find that exposure to drought-induced power generation from fossil fuels is associated with higher urgent respiratory morbidity. Perhaps surprisingly, we find that even power generation from natural gas is associated with large and significant increases in hospitalization rates due to respiratory conditions. Finally, after the expansion of wind and solar generation in the previous decade, the fossil dispatch response to hydropower scarcity became weaker; a counterfactual analysis suggests that this weaker response reduced drought-induced ramp-up of generation from fossil fuels and may have averted several hundred urgent respiratory admissions.

Can entry discipline a merger when establishing a new plant requires substantial sunk investment? Evidence from the Brazilian ready-mix concrete industry indicates that it may not, and that the resulting delay can be economically significant for antitrust policy. Competition in this industry occurs in narrow local markets, where acquisitions of small producers may fall below Brazil’s mandatory merger-notification thresholds. Using an annual panel of 454 geographically isolated municipalities from 2008 to 2018, we estimate a dynamic entry-and-exit model with separate demand thresholds for entry and continuation and a serially correlated unobserved demand process. The preferred specification implies that approximately 24,700 formal construction workers are required to induce a second plant to enter a market, while only 5,900 workers are required to sustain an existing duopoly. This wedge generates substantial hysteresis in local market structure. In a counterfactual duopoly-to-monopoly merger, it translates into approximately 4.90 additional discounted years of monopoly. The results show that the mere possibility of eventual entry is not evidence that entry will be timely: in markets with sunk costs, a one-time merger can reshape local competition for years, including when the transaction falls below national notification thresholds

How does generative AI affect labor markets in a developing economy? Using ChatGPT's November 2022 release as a timing shock, we combine Brazil's nationally representative household survey (5.7 million worker-quarter observations, 2017Q1–2024Q4) with the universe of formal-sector hiring and separation records (13 million events). In a difference-in-differences design comparing occupations above and below the employment-weighted median AI exposure score, workers aged 16–21 in high-exposure occupations earn 4.7% less than those in low-exposure occupations. The coefficient passes the linear pre-trends test, date placebos, and randomization inference that reshuffles AI exposure scores across occupations. Two supporting estimates partially explain this result: young workers in high-exposure occupations are 1.5 percentage points less likely to be formally employed, and administrative hiring wages for new hires aged 22–29 in high-exposure occupations fall 2.6% more than in low-exposure occupations. In contrast, prime-age (30–54) and older (55+) workers show no differential movement on either wages or formality, consistent with firms lowering entry-wage terms for new hires. Unlike US and UK evidence, where AI effects concentrate on employment quantities for 22–25 year-olds, adjustment in Brazil runs through entry wages and contract terms rather than aggregate hiring levels, consistent with prior evidence on automation in Latin America. To our knowledge, we provide the first nationally representative estimates of early generative AI labor market effects from a developing economy.

Many studies compute correlations between variables aggregated at different levels. Here we show that ecological fallacies can arise not only from inferring individual-level correlations from group-level ones, but also, from analyses using data with mismatched levels of aggregation – documenting the conditions under which such mismatch can even reverse the sign of underlying population parameters. We illustrate the phenomenon by leveraging unique school-level data on school staff’s implicit racial preferences (based on implicit association tests, IATs); often only available at higher levels of aggregation. In the data, (1) variance in commuting zones (CZ)-average IATs is ?70% lower than in school-level data; (2) aggregation induces non-classical measurement error in racial preferences; (3) race gaps in disciplinary measures are not systematically associated with CZ-average racial preferences; but (4) a significant association between race gaps and racial preferences – with the opposite sign – emerges when using school-level IATs instead.

We study whether extending the school day in Brazilian public high schools affects students' transition to higher education. Linking individual records from the School Census, the ENEM, and the Higher Education Census through students' taxpayer identifiers, we follow six cohorts of students who entered high school between 2012 and 2017, from the 9th grade of elementary school to college enrollment observed three years after expected high school completion. Exploiting the staggered adoption of full-time classes across schools in a difference-in-differences design, we find that completing high school in full-time classes increases on-time progression through high school by 6.2 percentage points, ENEM participation by 13.4 percentage points, and higher education enrollment by 3.3 percentage points. Enrollment gains are concentrated in channels particularly relevant for public-school students: enrollment in public institutions rises by 36% and enrollment through affirmative action quotas more than doubles relative to the control mean. Effects on ENEM scores are positive but modest. The evidence is consistent with full-time education improving college access primarily by keeping students on track and bringing them to the ENEM, rather than through large performance gains. Effects grow with exposure and are strongest for students with a longer history of full-time education during elementary school.

This paper estimates the causal effect of fare-free public transit on crime, exploiting the staggered adoption by eleven São Paulo municipalities in 2023. Using a Stacked Triple Differences design estimated by Poisson QMLE over a panel of 172 municipalities and H3 resolution-9 hexagons, we find that street theft near bus stops falls by 24.0% after adoption, whereas street robbery, a confrontational, mostly nighttime offense, does not respond. The asymmetry fits the premise that passive surveillance mainly deters opportunistic crime. The effect concentrates on weekdays and outside peak hours, when the policy adds foot traffic to previously underused spaces. The decline also reaches the 200-to-600-meter transition zone (?19.7%), ruling out spatial displacement. The estimates imply roughly 740 street thefts prevented per year, nearly a quarter of those that would occur around the stops absent the policy. These results provide the first causal evidence that fare-free transit reduces crime through informal surveillance.

In Brazil, the contraceptive method used by more than four in ten married women was tubal ligation during the 1990s. In 1997, the country restricted access to female sterilization for those aged 25 or older or aged at least 21 with two living children. In this paper, we estimate the effect of the reform on fertility using administrative birth records from 1994 to 2004 and an event-study design comparing adjacent cohorts around each eligibility threshold. Births rise by 0.40 per 1,000 women per quarter at the 21-year threshold and by 0.33 at the 25-year threshold, peaking within four years of the reform before gradually receding. The response is concentrated in regions where sterilization was most prevalent before the reform, where it is more than twice the pooled estimate. Additionally, using two nationally representative surveys on fertility, we show that the restricted cohorts eventually rebuilt contraceptive coverage through short-acting methods, with no increase in the use of long-acting reversible contraception. Together, these findings suggest that the fertility response reflected a gradual transition toward alternative contraceptive methods rather than a permanent reduction in contraceptive use.

This paper examines how a large-scale urban disaster in Brazil, which triggered the mandatory evacuation and forced displacement of thousands of low-income households, affected household welfare. We exploit the fact that the evacuation zone's boundary was drawn on the basis of independent geological risk assessments, and estimate the effects of forced displacement using difference-in-differences combined with propensity score matching on administrative panel data, linked at the individual level to matched employer-employee records to trace formal labor market outcomes. Displaced households relocated significantly farther from the city's employment concentrations---distance to the CBD rose by 47\% and to employment subcenters by 33\%---and faced higher commuting costs, including a 12\% increase in per capita transport expenditure. We find that commuting distance to their own workplace also rose, while average formal employment and dismissal were unaffected on average. This average, however, masks sharp heterogeneity by pre-disaster renting status: non-renting households relocated farthest and captured the housing quality gains, most visibly a 7 percentage point rise in the probability of adequate flooring, consistent with compensated relocation, whereas renting households absorbed the shock through higher commuting costs and a significant increase in the probability of formal job dismissal.

In this paper, we study a large-scale staggered deregulation reform implemented across more than two thousand Brazilian municipalities. The reform exempts low-risk economic activities from government licensing and other public authorizations required to start and operate. We first document the reform’s direct effect on bureaucracy using new data on the time required to open a firm, showing that adoption significantly shortens the municipal approval stage targeted by the reform. Consistent with the literature on entry deregulation, we find that the reform increases the number of formal establishments and formal employment. The results are robust to a specification that exploits within-municipality variation in exposure to the reform arising from municipality-specific definitions of low-risk activities. We further show that these gains translate into greater fiscal capacity: the reform increases municipal own-source revenue, driven by the tax on services, by 20%, and leads to higher municipal spending, including on health and education.

This paper estimates the health effects of long-range air pollution from agricultural fires in Brazil. I combine satellite-based measures of fire intensity with high-resolution atmospheric trajectories to construct monthly municipality-level exposure to distant upwind fires. The empirical strategy exploits variation in smoke transport generated by the South American Low-Level Jet, a major atmospheric current that carries emissions from the Amazon and Cerrado toward Brazil’s South and Southeast regions. To isolate plausibly exogenous variation in exposure, I instrument realized upwind fires with predicted exposure based on fixed baseline fire locations and time-varying wind trajectories. A one-standard-deviation increase in distant upwind fire exposure raises monthly PM2.5 concentrations by 0.13 \mu g/m^3, validating the pollution-transport mechanism. The same shock increases respiratory hospitalizations by 3.6 percent and all-cause mortality by 1.0 percent. Mortality effects are concentrated among vulnerable individuals and in respiratory causes: respiratory mortality rises by 3.8 percent in the full population and by more than 4 percent among individuals aged 70 and above. Placebo tests using hernia outcomes and future fire exposure show no systematic effects. These findings demonstrate that agricultural burning generates economically meaningful health externalities far from fire-origin regions, highlighting the importance of coordinated environmental policy across jurisdictions.

We study how loan guarantees, which expand credit access for borrowers without suitable collateral, transmit repayment shocks within a financial cooperative serving dairy farmers in Kenya. Monthly records for about 25,000 members reveal an institution-spanning network in which nearly all participants belong to one connected component and more than half are both borrowers and guarantors. Exploiting variation in the timing of milk production shocks, we find that a shock raises nonperformance by 17.4 percentage points for the affected borrower and 1.4 percentage points for borrowers connected by a direct guarantee, even though their milk production is unchanged. Propagation accounts for 43% of additional nonperformance over the next six months. Calibrated simulations separating contagion from correlated shocks show that contagion accounts for about one-third of average nonperformance and nearly all of the risk that nonperformance exceeds 20%, twice the regulatory threshold for serious deterioration. Loan guarantees expand credit access, but can also expose financial institutions to contagion risk

This paper studies how the effects of minimum wage increases on wages and employment vary across similar workers employed at different firms. Using matched employer-employee data from Brazil (RAIS, 1996--2012), I exploit pre-policy exposure to the minimum wage — measured by the share of earnings falling below the incoming minimum wage at the worker, firm, and commuting-zone level — in a difference-in-differences design applied at each level of aggregation. The main specification explores worker-level heterogeneity, jointly exploiting individual and firm-level exposure to test whether the effects vary with the exposure of the employing firm. Bound workers employed at more exposed firms experience smaller wage gains and are significantly more likely to separate and transition to new employers than comparable bound workers at less exposed firms. Commuting-zone-level estimates show only modest employment effects, whereas firm-level estimates reveal substantial employment contractions among highly exposed firms. Employment losses at the firm level are more than five times larger than those estimated at the commuting-zone level, suggesting that the main adjustment margin is within-market reallocation: workers displaced from highly exposed firms are largely reabsorbed by less exposed firms rather than exiting formal employment.

This paper examines the effects of a climate adaptation policy on production and environmental outcomes in the context of Brazil's semiarid region, the country's poorest and most drought-prone region. The large-scale, low-cost water policy builds rainwater reservoirs designed to boost production and strengthen rural producers' resilience. Using a difference-in-differences approach and linking property-level administrative data to high-resolution satellite data, we find that cistern construction reallocates land toward higher-productivity uses. Results indicate an increase in cropland area by 7.6% and higher-quality pasture area by 14.5%, while lower-quality pasture area decreases by 3.2%. Forest cover increases by 1.1%, consistent with a land-saving effect driven by a reduction of lower-quality pasture. Effects hold across property sizes and are slightly larger in magnitude for small-sized properties. Our cost-benefit analysis reveals a positive aggregate return with each invested monetary unit generating 1.76 units of benefits, indicating that adaptation policies can also advance mitigation goals via forest preservation.

Why do so many informal workers earn the minimum wage? We argue that this phenomenon, known as the lighthouse effect, exists because the minimum wage acts as a benchmark and lowers wage-setting costs for firms. We develop a dual-sector model with monopsonistic competition and minimum wage-anchoring features that generates equilibrium wage distributions featuring bunching at the minimum wage and missing masses around it in both the formal and informal sectors. We estimate the model using Brazilian firm- and worker-level microdata. Modeling lighthouse effects matters: ignoring the informal spike overstates the effect of minimum wage increases on informal wage inequality and meaningfully alters estimated structural parameters.

This paper examines how European immigration during the Age of Mass Migration (1886-1920) shaped long-run political preferences in Brazil. While existing research links the emergence of labor-based politics in Latin America to industrialization and urban working-class growth, we argue that earlier demographic changes induced by mass migration also played an important role. Using an instrumental-variable strategy that exploits the interaction between aggregate immigration inflows and the expansion of São Paulo's railway network, we show that municipalities with a larger share of European immigrants were more likely to support labor-oriented candidates during Brazil's democratic interregnum (1945-1964). We also show suggestive evidence that the political effects of immigration persisted after Brazil's return to democracy in 1985. The findings suggest that the consequences of mass migration extended beyond economic development and contributed to the long-run formation of local political preferences.

This paper studies procurement auctions in which the government acts simultaneously as the market designer and as a participant through state-owned enterprises. Using Brazil’s electricity transmission auctions, I examine whether the more aggressive bidding of public firms reflects productive efficiency or objectives that differ from profit maximization. Reduced-form evidence shows that public bidders submit lower bids than the average private bidder and bid particularly aggressively in politically aligned locations. I develop a first-price procurement model in which public firms maximize expected profits together with an additional objective that depends on the awarded procurement price. Because differences in bids can arise from both cost heterogeneity and objective-function heterogeneity, these components are not separately identified from bidding data alone. I therefore partially identify the marginal public objective using restrictions that allow public firms to be more efficient than the average private bidder, while ruling out the stronger explanation that they are systematically more efficient than the most efficient private bidder in the same auction. Moment-inequality estimates imply that public bidders behave as if they place additional value on reducing procurement payments. Under the preferred specification, this objective lowers their optimal markups without requiring below-cost bidding. The results show that state participation can discipline procurement prices, but may also generate a trade-off between fiscal savings and allocative efficiency.

Developing countries are characterized by substantial gender inequality and large labor market informality. The latter is particularly pronounced among women and increases substantially after childbirth. While informality may help mothers remain attached to the labor market by o?ering flexibility, informality is also associated with lower earnings, limited human capital accumulation, and no social insurance—potentially exacerbating career penalties over the life cycle. This paper studies motherhood penalties over women’s life cycle, how they are shaped by informality, and how public policies could reduce them. We build a novel dataset linking over 40 million births in Brazil to both parents’ employment outcomes to document large formal employment penalties of around 40%, half of which materializes before childbirth. We develop and estimate a partial equilibrium life cycle model of women’s fertility and labor supply choices across self-employment, formal and informal wage employment. The counterfactual exercises indicate that motherhood penalties are long-lived, persisting for more than 30 years. Policies that increase formal employment flexibility could substantially reduce lifetime employment penalties, eliminate the shift to informal employment after motherhood, and reduce lifetime consumption losses by more than half.

This paper studies how trade-induced demand shocks affect firms' internal organizational structure. We combine matched employer–employee data from Brazil with a shift-share measure of foreign demand shocks generated by the Great Recession of 2008. We measure organizational structure using hierarchical occupational layers and estimate the causal effect of demand shocks in a difference-in-differences framework with continuous treatment. We find that firms exposed to adverse demand shocks reduce their organizational depth, while favorable shocks generate comparatively limited organizational expansion. This asymmetry suggests that organizational contraction is the main driver of the average effect. We further show that trade shocks increase employment and within-firm wage inequality, with little impact on average wages. Finally, among firms that maintain a fixed organizational structure, positive demand shocks are associated with higher wages within existing layers, consistent with firms adjusting the knowledge embodied in incumbent workers rather than expanding their hierarchy. Overall, our findings provide new causal evidence that international trade reshapes firms' internal organization.

People often make important choices in pursuit of higher income, but do they place too much weight on income when deciding what path to take? We propose a simple model of misspecified learning in which individuals overestimate the marginal satisfaction from income---the expected effect of an income increase on overall life satisfaction---and therefore give income too much importance in their decisions. Guided by this model, we designed a pre-registered experiment that elicits beliefs about the marginal satisfaction from income and randomly assigns scientific evidence about it. Respondents report beliefs about the marginal satisfaction from income that are substantially above the scientific-evidence benchmark, both for themselves and for others, with a larger gap for themselves. These gaps shrink when respondents are exposed to scientific evidence, and the effects persist one month later. To study whether these beliefs matter for behavior, we measure income-versus-non-income trade-offs using job-choice scenarios tailored to each respondent and a real-world decision the respondent is facing. To elicit the latter, we make a methodological contribution: an AI-led interview method that moves beyond static, predefined survey instruments by combining the flexibility of qualitative interviewing with the discipline of closed-ended survey measurement. We find that these beliefs are consequential: after learning that income matters less for life satisfaction than they initially thought, respondents place less weight on income in their decisions.

We study the labor-market effects of a major flood disaster in Brazil and whether a temporary labor-retention scheme — Programa Emergencial de Apoio Financeiro (PEAF) — mitigated these effects by providing wage subsidies while requiring firms to retain covered workers. Drawing on administrative data, we use a matched event-study design that compares workers in flood-affected establishments with and without PEAF coverage to matched workers in non-affected establishments. Direct flood exposure reduced formal employment among unprotected workers by about 2 percentage points and increased job switching, indicating rapid reallocation to other formal employers. PEAF increased employment by 2.4 percentage points relative to the counterfactual of flood exposure without program protection. PEAF also reduced reliance on unemployment insurance, consistent with its effect on employment retention. Earnings effects are more nuanced: PEAF mitigated unconditional earnings losses through employment reservation, but earnings among retained workers declined, consistent with downward compensation adjustment within continuing jobs. Effects are concentrated in very small establishments and in sectors more exposed to disruption. The results show that labor-retention schemes can preserve employment after climate disasters, but employment preservation need not imply full income protection, a distinction that is central to the design of policies aimed at mitigating the economic consequences of climate shocks.

This paper examines firms’ and workers’ responses to a rapid e-commerce expansion. I leverage regulation-induced variation in shipping costs and comprehensive matched employer-employee data from Brazil. Using difference-in-differences designs that compare retail segments within local labor markets, I find that e-commerce growth reduced employment in exposed retail sectors by 11-14% and the number of establishments by 10%. Employment effects on incumbent workers were modest: workers who moved to other sectors or regions experienced no earnings losses, while those who remained in the exposed sector saw wages 6% lower by 2016, reflecting forgone wage growth rather than nominal cuts. The aggregate impacts were primarily driven by reduced entry (of new establishments and of first-time workers into formal employment), with only limited absorption of displaced entrants into other formal sectors. These findings underscore the importance of entry margins in capturing labor market responses to sector-specific shocks.

Reforms that deregulate labor markets seek to reduce frictions in the relationship between firms and workers, but their effects may depend on the institutional capacity to implement them. We study the 2017 Brazilian Labor Reform, which lowered the costs of litigating in the labor courts, to investigate how pre-existing differences in the capacity of labor courts shaped its effects on the formal labor market. Using an instrument based on the potential extra jurisdiction of seat municipalities of labor courts, we show that municipalities that were more congested before the reform had relatively larger growth in formal employment (about 0.7 percentage points, for a 10\% increase in congestion), as well as in involuntary dismissal and resignation rates (0.1 to 0.2 percentage points). The effects are heterogeneous across demographic groups: non-white women respond less than the other groups in involuntary dismissals but more in resignations, an asymmetric pattern that suggests institutional capacity shapes not only the intensity but also the distribution of the reform's effects across workers.

Land inequality may shape economic development as it is an important resource to food production. In Latin America there are large, often underused, farms despite thousands of families who want to farm, which could be evidence of misallocation of land. This paper studies a production-based criteria for expropriation and redistribution implemented in Brazil over four decades. I combine administrative records on the geolocated boundaries of redistributed farms with satellite-derived land-use data and municipality-level production data in an event-study design that relies on the variation in timing and location of expropriations. I show that redistributed farms increase their farming activity by 7% on average ten years after redistribution. I also document that neighbouring farmers increase their farming activity by 1% on average after redistribution, consistent with the idea that the production-based criteria induced farmers to increase their output to prevent expropriation.

We estimate the causal effects of a Brazilian full-day schooling program that extends instructional time from approximately four to seven hours per day on students' academic outcomes. In 2018, the Ministry of Education randomly assigned 204 public high schools to treatment or control through a public lottery stratified by state and prior performance, providing direct experimental identification in a literature that otherwise relies on quasi-experimental variation. We find that full-day schooling raises high school approval rates by 2.4 percentage points and lowers grade retention 1.8 percentage points in the first year post-treatment, with dropout effects negative but imprecisely estimated. These average effects mask substantial heterogeneity, where gains are concentrated in schools serving students with higher shares of working-age enrollment, higher nighttime enrollment, and located in the poorer regions. To distinguish between student dropout and reallocation to neighboring schools, we construct 2~km school markets using GPS coordinates and test whether treatment affected total market-level enrollment. Preliminary evidence favors student migration over true dropout, with market-level enrollment remaining stable after the transition to full-day schooling. Our findings suggest that full-day schooling is most effective precisely where educational disadvantage is greatest, without evidence of large-scale student dropout.

This study investigates the potential of peer mentoring programs to enhance organizational efficiency and outcomes by pairing high-performing managers with underperforming peers. Using a state-level peer mentoring program for school principals in Ceará, Brazil, we employ a regression discontinuity design (RDD) to identify causal effects. Our findings show that the peer mentoring program increased the performance of mentored principals, reflected in a 0.247 standard deviation increase in literacy achievement among 2nd-grade students, without harming the performance of mentor principals. These gains are reflected not only in average performance but also in the reduction of literacy disparities among students within schools. Heterogeneity analysis indicates that effects are stronger when mentors and mentees are geographically close or when mentors are of higher quality. These findings suggest that peer mentoring is an effective and scalable alternative to leadership turnover for improving the performance of low-performing managers.

We estimate the marginal treatment effects—specifically, of enrollment in formal daycare—on fifth-grade math and Portuguese test scores, using an administrative panel of 11.1 million students from Brazil's public school system, covering the 2003 to 2016 birth cohorts. The identification strategy exploits the uneven expansion of municipal daycare supply across cohorts, using as an instrumental variable the net public-system enrollment rate when the cohort's children were three years old. The marginal treatment effect curve is U-shaped: returns peak near 1.9 standard deviations for children with low unobserved resistance to enrollment, fall to 0.33 standard deviation at intermediate levels of resistance, and rise again at the upper end of the distribution. The local average treatment effect for compliers is 0.67 standard deviation in math and 0.65 in Portuguese, although this figure is provisional: it is concentrated among students who are behind in school (grade-for-age lag), and this heterogeneity has not yet been separated from a possible mediation effect arising from grade retention. The ordering ATT > ATE > ATUT indicates positive selection on gains, with no negative returns at any point of the distribution.

Does employer market power breed modern slavery? Market power and forced labor are two routes to the same rent, one legal and one criminal, and an employer who holds the first may have less use for the second. We define labor markets by municipality and farm activity in Brazil, measure the concentration of each from the Agricultural Census, and match them to the universe of federal anti-slavery inspections between 1995 and 2019. Within the same municipality and period, the activity with more concentrated land has fewer operations finding slavery, fewer rescued workers, and a lower probability of any rescue. Instrumenting a market’s concentration with the structure of the same activity elsewhere in the country and absorbing shocks common to an activity and period give the same answer. The pattern does not come from where enforcement looks, since operations respond to complaints and concentration lowers their arrival and their findings together. In our model of recruitment by deception and retention by confinement, the gain from coercion is the wage bill it avoids, a bill the markdown of a dominant employer has already collected, while concentration may also cheapen the scheme, so the model leaves the sign to the data. The two ways an employer can hold a wage down are substitutes, and the legal one crowds out the criminal one.

Governments are increasingly relying on private firms to deliver publicly funded goods and services as a cheaper option than direct public provision. Strategic responses from firms in the targeted industry might generate unexpected effects from such policies. This paper leverages a supplier-targeted subsidy policy in Brazil to study how market structure is affected by this practice. This subsidy policy created a reimbursement system through which targeted retail pharmacies could dispense subsidized prescription drugs. Exploiting the exogeneity created by the timing each store accesses the subsidy, I find the subsidy benefits any targeted pharmacy by increasing its survival probability from 54% to 80%. At the market level, I find that the presence of subsidized incumbents decreases the rate at which new stores enter policy-exposed markets, and the effect is concentrated among smaller stores. I then use a dynamic entry model to understand counterfactual effects of the removal of the subsidy, and find that without such policy, 7% less markets would have at least one single store. Results indicate that the subsidy helped bring stores to more markets than what would have happened otherwise, at the same time it did not differently favor any particular kind of store, although it have pushed smaller stores to arguably less attractive markets.

This paper studies how insurer market structure shapes the payment contracts negotiated between insurers and healthcare providers. Using the universe of private inpatient claims in Brazil from 2017 to 2024, I exploit staggered insurer exits as a positive shock to the bargaining power of surviving insurers, testing whether they use this leverage to shift financial risk to providers. In this setting, risk shifting is measured by the adoption of payment through bundles of procedures. Using a difference-in-differences design that allows for non-binary treatment, accounting for the multiple exits observed in many treated markets, I find that exits increase the share of bundled procedures by about 32 percent relative to the pre-treatment mean. Consistent with the predictions of the prospective payment literature, within-episode care intensity falls, with shorter hospital stays. I find no detectable effects on in-hospital mortality.

We use firm-level data to estimate markup responses to uncertainty shocks. We measure markups using a production function approach in a quarterly-frequency dataset of US publicly listed firms. We recover structural shocks from macroeconomic data using a Bayesian VAR with stochastic volatility. We find that markups are procyclical in response to uncertainty shocks: a one-standard-deviation increase in macroeconomic uncertainty lowers markups by about 0.30 percentage points. The effect is long-lived, lasting up to three years. Using a workhorse New Keynesian general equilibrium model with nominal price and wage stickiness, we quantitatively match the estimated empirical response under a calibration that matches the average duration of prices and wages. Because wages are stickier than prices, uncertainty raises the real wage and marginal costs, leading to a decline in markups.

This paper asks how public transport shapes the spatial organization of informal settlements (slums). Studying the São Paulo Metropolitan Region, the largest in Latin America, I show that new transit does not simply expand or contract informality but relocates it: opening a station lowers the share of informal residents near the city center while raising it in the periphery. To uncover the mechanisms behind this pattern and measure its welfare consequences, I build a quantitative spatial model with high- and low-skill workers and separate markets for formal and informal housing. A closed-city counterfactual shows that transit expansion on its own barely changes aggregate informality, rigid formal housing supply converts improved access into higher rents and increasing informality. Pairing the same expansion with a targeted relaxation of formal supply around new stations instead cuts informality by 12\% and raises welfare by roughly 11\%, more than the two policies deliver in isolation, showing that transit access and housing policy are complements. A fourth counterfactual re-estimates the model as an open city and is able to reproduce the empirical pattern: once migration is allowed, transit expansion raises informality through an inflow of low-skill workers, showing that peripheral slum growth reflects migrants drawn by better access rather than residents relocating within the city. Thus dealling with informality requires pairing infrastructure investment with land-use deregulation.

Racial inequality can vary over time because the relative position of racial groups in the earnings distribution changes, or because race-neutral shifts in the distribution affect groups differently since they hold different positions in it. One key race-neutral policy shown to shape the earnings distribution is the minimum wage. In this paper, I extend the Bayer and Charles (2018, QJE) decomposition approach in a way that accounts for the role of the minimum wage in driving changes in racial differences due to shifts in the distribution. I then apply this framework to understand the forces behind the evolution of earnings differences between Black and White Brazilian men from 1960 to 2025. A race-neutral compression of the earnings distribution was the main factor reducing the gap at the median and the bottom quartile, while changes in the relative position of Black men played a key role in the convergence at the top of the distribution. The minimum wage's effect on the racial gap is strongly bottom-concentrated and close to zero from the median up. Its erosion through the 1980s widened the bottom gap, while its recovery afterwards compressed it. However, because Black workers are the majority of those earning below the minimum wage, the same recovery widened the gap below the 10th percentile. On net, the rising minimum wage was a first-order factor behind the racial convergence in mean earnings over the period.

In many institutional settings, standardized scores and classifications determine eligibility for treatment and access to care. This paper studies how measurement technologies affect healthcare allocation when embedded in threshold-based clinical rules, focusing on Sweden’s transition from last-menstrual-period dating to routine ultrasound dating of pregnancy. Ultrasound dating standardized the clinical clock used to classify pregnancies as preterm, term, or post-term and to trigger obstetric management. Using linked Swedish register data from 1973 to 2017 and an instrumental-variables design based on lagged hospital ultrasound-dating intensity, I show that ultrasound dating shifts recorded gestational age and reclassifies pregnancies across clinically relevant thresholds. The effects are strongest where the classification rule is most consequential: at the post-term threshold. Because ultrasound dating maps fetal size into gestational age using sex-neutral charts, it may systematically understate gestational age for female fetuses, who are smaller on average, thereby changing their measured eligibility for post-term care. Downstream effects are concentrated in neonatal conditions and postnatal healthcare use, while longer-run effects on school grades appear only in selected specifications. In municipalities where routine dating occurred earlier in pregnancy, when fetal growth is more homogeneous across sexes, the adverse reclassification pattern is attenuated, and some outcome estimates are more favorable. The findings show that standardization can reallocate care when measurement rules interact with discrete treatment thresholds, making measurement error policy-relevant once the rule becomes operational.

We study whether unverified but credible narratives can shape political behavior by activating collective memories of past economic shocks. A 1989 fungal outbreak devastatedcocoa production in Northeast Brazil, displacing nearly 250,000 workers. Seventeen years later, a widely read newsmagazine article attributed the outbreak to deliber-ate sabotage by left-wing Labor Party affiliates — an allegation drawn from a real but ultimately inconclusive Federal Police investigation. Exploiting variation in outbreak ex-posure across municipalities, we estimate that the 2006 news release reduced left-wing presidential vote shares by 12.3 percentage points and increased right-wing vote sharesby 7.8 percentage points in mayoral elections, respectively. Effects persisted across four presidential elections through 2022. We also show strong effects in municipalities withat least one radio station. The findings show how unverified narratives can reactivate collective memories of economic trauma and durably alter political alignment

The relationship between schooling and crime is well documented in the economics literature: more schooling means less crime. Yet the mechanisms behind this relationship remain an open question, especially in developing countries. We ask whether full-time schooling lowers crime by building human capital or simply by keeping students supervised, exploiting the staggered rollout of São Paulo's Full-Time Education Program (\emph{Programa de Ensino Integral}) across public schools between 2012 and 2024. Using \citet{callaway2021difference}'s doubly-robust difference-in-differences estimator applied to geocoded crime records, we find that adopting the full-time model reduces property crime in the surrounding area by about 7\% (18.8 fewer offenses per hexagon-year), an effect driven by robbery. The reductions are sharply concentrated in time and by offender: theft falls by 26\% during school hours but is, if anything, slightly higher outside them; effects are strongest on school days; and crimes committed by juveniles fall disproportionately — roughly 65\% for theft and 23\% for robbery. Because the effects emerge precisely when, and for whom, school supervision is most binding, they are hard to reconcile with longer-run human-capital or opportunity-cost channels and instead point to incapacitation. Full-time schooling thus yields immediate public-safety gains in a middle-income setting, which is relevant to both education and crime-prevention policy.

This paper provides the first systematic evidence that organizations contribute to gender inequality by designing or sustaining work environments with unequal opportunities to perform productive tasks. We explore this question in the context of surgical specialists, a setting where task output is readily measurable, determines career success, and is comparable across firms. Using rich administrative data from Brazil, we document a gender gap in surgical output of about 27 percent that emerges early in careers, cannot be explained by surgeons' background characteristics or work schedules, and accounts for most of the gap in payments directly tied to medical procedures. Exploiting variation generated by surgeons moving across hospitals in a triple-difference design, we show that workplace environments causally account for a large share of the gap. Estimates of hospital-specific work environment effects indicate that more male-centric hospitals retain fewer women and exhibit lower scores on other revealed-preference measures of firm value, suggesting that women are more dissatisfied at these organizations. Female managers mitigate these disparities, but only when baseline gaps are especially pronounced. These findings recast part of gender inequality as a problem of production, not only compensation, and highlight the importance of organizational design for inequality in high-skill labor markets.

A firm's compensation shapes the outside options of workers at competing employers. We study how this channel produces spillovers in the setting of collective bargaining, exploiting a legal change in Brazil that raised union bargaining power at a subset of firms. To capture indirect exposure at competing employers, we construct a network of firms connected by workers' job-to-job transitions. Firms directly affected by the reform raise wages and provide more collectively bargained amenities. Workers at closely competing firms also see wage gains, about one-fifth of the direct effect, but no change in amenities. We find no evidence that unions actively propagate these spillovers, supporting the competition-driven interpretation. Employers' wage responses leave employment and workforce composition broadly unchanged, and are larger where competition runs through longer-tenured workers, who are costlier to replace. Within firms, we detect no differential wage response to outside-option exposure across workers, suggesting internal wage-setting constraints.

A growing literature documents how religious institutions shape behavior through social influence, but less is known about what happens when religious movements gain political power and use the tools of government to advance their agenda. We use a regression discontinuity design on close mayoral elections in Brazil to show that mayors from parties institutionally tied to Pentecostal denominations increase teenage fertility 3 per 1,000 higher (a 40% increase). This effect appears for cohorts exposed to middle school during the administration. Consistent with a school-based mechanism, we find that the likelihood that municipal schools offer sexual education programs falls by 12.5 percentage points, with no changes in state schools outside mayoral control. We also find elevated STD rates, and higher middle school dropout rates, while slightly older cohorts show no effects. Results are not explained by changes in contraceptive availability in public clinics, pointing to sexual education as the primary mechanism. We also find no effects from other right-wing parties, indicating the importance of institutional links to Pentecostal parties. }

Empirical studies of racial disparities often treat race as fixed, though scholars across disciplines view it as fluid, context-dependent, and observer-specific. We study the determinants of racial perception in the labor market and its consequences for measured wage disparities. Using linked administrative data on 330,000 workers in Brazil (2003-2015), we observe three racial measures: employer-recorded race from payroll filings, self-identified race from electoral records, and phenotype from standardized photographs. Self-identified and employer-ascribed race diverge in over 20\% of cases, and employers disagree in their classification of the same worker. Estimating a 'race function' that captures how employers map worker characteristics into racial categories, we find that phenotype matters, but so do productivity-relevant characteristics. University-educated workers are 10 pp more likely to be classified as White than workers with the same skin tone but no high school degree. Education 'whitens' workers even conditional on self-declared race, all visual cues in the photographs, and firm-by-occupation fixed effects. Conventional wage-gap estimates based on recorded race or skin tone shrink by 80-90\% under a comprehensive set of controls, partly because those controls also shape racial classification. We introduce a perception-normalized disparity measure that attenuates far less, suggesting that conventional estimates overstate the extent to which productivity differences explain racial wage gaps.

This paper studies how changing upstream firm boundaries affects downstream logistics in illicit markets. Exploiting a reorganization of drug trafficking routes in Brazil following the expansion of control by the Primeiro Comando da Capital (PCC) over wholesale nodes in Paraguay in 2016, I examine how vertical integration reshapes shipment size and routing downstream. First, I develop a simple model of vertical integration to derive main implications and guide the empirical analysis. Then, using micro-level police seizure data from São Paulo between 2003 and 2022 and a triple-difference design, I show that cocaine shipments seized during transportation become significantly larger after 2016, while no comparable changes occur for crack cocaine, a locally produced drug. Geocoded evidence further suggests increased concentration of distribution near major transport infrastructure. Finally, I use additional, country-wide data to analyze potential mechanisms and I find evidence of foreclosure: after vertically integrating, PCC partially reduces drug sales to downstream competitors, claiming the residual volume. Overall, the results highlight the role of firm organization in shaping logistics within illegal markets.

Transportation is a major source of greenhouse gas and air pollution emissions worldwide, with most automobile pollution coming from vehicles older than 10 years. However, many policies discourage the removal of older vehicles from the fleet, with potential consequences for pollution and public health. This paper examines an anti-scrappage policy in Brazil using a border-pair design that exploits state variation inthe age thresholds for vehicle ownership tax exemptions. Using municipality-level vehicle registration data from 2013-2020, we find that lower exemption age thresholds increase the average fleet age by up to one year, as they encourage owners to retain vehicles until they qualify for tax exemptions. Combining fleet-based estimates with satellite data, we show that these policies increase on-road CO2 emissions per capita by up to 23% and PM2.5 concentrations by 4%. We also find that the incidence of low birth weight rises by 10% under a 10-year exemption policy and by 2.8% in areas with thresholds between 15 and 20 years. Despite these social costs, we present suggestive evidence that electoral incentives make such policies difficult to reverse.

Access to sewerage infrastructure is widely recognized as a key public health intervention, yet recent sanitation expansions in developing countries have often generated modest health gains. This paper studies a large-scale sewerage expansion in Campo Grande, Brazil. Unlike most previous studies, we isolate the health effects of sewerage expansion in a large urban setting where piped water and indoor toilets were already nearly universal, allowing us to identify the effects of replacing on-site sanitation systems with sewerage collection and treatment.We combine the universe of administrative birth records with georeferenced information on the timing of block-level sewerage expansion to compare mothers living in the same location but exposed to the new infrastructure at different stages of pregnancy. We find that each additional month of prenatal exposure increases birth weight by approximately 20 grams, primarily through improvements in fetal growth rather than longer gestation.Our administrative data also allow us to examine why sanitation investments do not benefit all households equally. Health gains are concentrated among older and more educated mothers, while younger and less educated mothers experience substantially smaller effects. We provide evidence that financial barriers to household connection help explain these heterogeneous impacts. Although sewerage infrastructure became available, connecting to the network required households to incur substantial upfront and recurring costs.Our findings suggest that maximizing the health returns to sanitation investments requires not only expanding infrastructure, but also overcoming last-mile financial barriers that prevent vulnerable households from connecting to the sewerage network.

We study the effect of greater exposure to Pix, Brazil’s instant digital payment system launched in November 2020, on firm formalization, legal form choice, and tax collection. We exploit differences in pre-existing mobile access infrastructure, which strongly predicts Pix adoption after rollout. We find that greater Pix exposure increased participation in the corporate tax system: in our preferred specification, municipalities with greater exposure experienced a 0.15 percentage point increase in the share of corporate taxpayers, equivalent to roughly 15 percent of the sample mean. By contrast, the share of individual income tax filers does not move---the estimate is a precisely estimated zero---suggesting that digital payments primarily encouraged formalization and legal-form switching among firms rather than broad household entry into the tax system. These effects are accompanied by higher tax revenues, including an increase in corporate income tax revenue per capita of about 6.3 percent relative to the sample mean. The results suggest that digital payment systems can strengthen fiscal capacity by increasing the traceability of transactions and encouraging participation in formal tax regimes. The gains are uneven across tax instruments, however, and track which authority observes the payment trail: they are largest for the federally administered income taxes, weaker for the municipally enforced service tax, and absent for a formula-based state transfer. That unevenness carries direct implications for economies in which informality is high and subnational tax administrations differ widely in capacity.

This paper studies how consumption opportunities vary within a city and how these differences shape urban inequality and residential sorting. Using novel barcode-level price data from all formal retailers in Alagoas, Brazil, I document large within-city disparities in prices, varieties, and product quality. Higher-income neighbourhoods offer greater variety and quality at systematically higher prices. To interpret these patterns, I develop a quantitative spatial model in which households value varieties and quality differently by income, and firms choose both locations and product assortments. In this environment, neighbourhood composition affects the market size for a given variety, creating consumption externalities: households benefit from living near others with similar preferences because it expands the set of varieties they can access. The model clarifies how these endogenous differences in product availability and pricing generate an amenity that attracts richer households and amplifies spatial inequality. A counterfactual that eliminates local shopping frictions substantially reduces income-gradient sorting, quantifying how endogenous retail supply amplifies spatial inequality through the consumption channel.

Place-based tax incentives are widely used to address regional inequality and are routinely justified on equity grounds. Yet, their consequences for the distribution of pay within targeted firms remain largely unexplored, despite the first-order relevance of firm pay policies in shaping aggregate earnings inequality. We explore the rollout of a large federal corporate income tax benefit granted to plants operating in the Brazilian Amazon to study how tax incentives reshape within-plant wage inequality, the skill composition of employment, and R&D activity. Using a staggered difference-in-differences estimator with not-yet-treated establishments as the comparison group, we document a decline of approximately 6.4% in within-plant wage dispersion in beneficiary establishments. This observed decline is explained by a leveling-down process, rather than gains at the bottom: average earnings fall by 7.5%, with a larger reduction among lower-skilled workers (8.9% reduction vs. 6.2% for higher-skilled workers). Treated plants hire approximately 27% more lower-skilled workers, while higher-skilled employment remains unchanged, a pattern consistent with the absence of any detectable effect on patenting activity we observe. A back-of-the-envelope cost-benefit analysis yields a Marginal Value of Public Funds of approximately 0.43, meaning that each real of public funds channeled through the program delivers about forty-three cents of additional labor income to workers, placing SUDAM among the least cost-effective place-based interventions evaluated in the literature.

How do different income taxation systems affect household decisions and welfare? We answer this question by first documenting the strong labor supply disincentives for secondary earners of the U.S. tax system and by using variations from the Bush Tax Cuts to assess their effects on intra-household specialization. We then develop a lifecycle model incorporating labor supply, marriage and divorce decisions with limited commitment, household production, human capital accumulation, and assortative mating. After estimating and validating the model with various datasets, we evaluate four tax systems: a U.S.-like income-splitting system, an individual taxation system, a flexible general joint system, and an income-splitting system with secondary-earner deductions. We find that the individual taxation system provides higher welfare than income splitting but increases inequality. The general joint system offers the highest welfare but is complex to implement. The income-splitting system with a secondary-earner deduction improves welfare and reduces inequality while maintaining simplicity.

Despite widespread expectations that economic development would eradicate slavery, modern forms of labor coercion persist. This paper examines how technological change can sustain or mitigate coercive labor relations in exposed localities. We study Brazil, where heterogeneous adoption of labor-saving genetically engineered soybean seeds and labor-intensive maize double-cropping can be linked to detailed municipal data on anti-slavery inspections and rescued workers. We find that greater suitability for GE seeds increased reported cases of forced labor, largely through displacement into other sectors such as cattle ranching, while suitability for double-cropping reduced forced labor. The displacement of at-risk workers emerges as the central mechanism. Both technologies shift the employment of workers who match the demographic profile of rescued slaves by two to four times their effect on the average worker, and the entire adjustment occurs in informal work. A simple model of coercion in local labor markets rationalizes these findings, with a larger pool of displaced workers exposing more people to enslavement even as it erodes the wage that coercion saves. The response of forced labor to job losses is roughly one and a half times as large as the response to job gains, suggesting an asymmetry in which coercion expands faster than it recedes.

What are the short- and long-run impacts of Black teachers on students' educational and labor market outcomes? To answer this question, we leverage linked nationwide administrative data from Brazil to track two cohorts of about one million students each—from middle (6th grade) and high (10th grade) school—over 12 years, following them into higher education and the formal labor market. Our empirical strategy compares students within the same school who are assigned to classrooms with varying shares of Black teachers while controlling for prior-year test scores and a rich set of covariates. For Black students in the 10th-grade cohort, we find that having a higher share of Black teachers increases their probability of high school graduation, college enrollment, and college graduation. These educational gains translate to the labor market, resulting in higher earnings by the age of 27. Our analysis of the 6th-grade cohort reveals similar patterns, including improved test scores for Black students and increased rates of on-time high school graduation and college enrollment. Across both cohorts, we find no effects on their White peers. Our results suggest that increasing the share of Black teachers to 50%, approximately the share of Black individuals in the Brazilian population, would reduce the covariate-adjusted within-school racial gap in education by at least 30% and in earnings by about 60%, compared to a scenario with no Black teachers.

This paper estimates the causal effects of Operação Verão (2023/24), a large-scale militarized police operation in coastal São Paulo, Brazil, using Difference-in-Differences and Synthetic Difference-in-Differences on a municipality-month panel. The operation sharply increased police killings, approximately 2.0 additional deaths per month, more than seven times the pre-intervention mean, while producing only limited, identification constrained crime reductions. Drug seizures at the Port of Santos declined significantly, consistent with criminal organizations suspending export activity in response to heightened enforcement. Health records reveal a concurrent rise in aggression-related injuries among young men. Exploiting the contrast with the preceding Operação Escudo (2023), which raised police lethality without generating deterrence, I show that sustained territorial deployment, rather than lethality per se, is the operative mechanism of deterrence. These findings raise important questions about the proportionality and strategic design of militarized policing under stable criminal governance.

Brazil cleared just 36 percent of homicides in 2023, far below comparable countries,with large consequences for the certainty of punishment. We study Portaria 580/2021,a police management reform in the state of Para´?ba that formalized accountabilitystructures around police precincts and specialized homicide investigation units. Usingcase-level administrative data on 11,283 homicides across 223 municipalities from2015 to 2023, we compare municipalities hosting an investigative unit with non-hostmunicipalities before and after the reform. Our preferred estimator implies a 76 percentcompression in the expected time to clearance, whereas the probability of clearanceincreased by 7.7 percentage points, a 16 percent improvement over the pre-reform mean.We leverage the richness of our dataset to explore potential mechanisms driving ourresults. Null results for flagrante arrests, arrest rates, and prosecutorial indictmentsconfirm that gains operate through the investigative channel, not through generalenforcement activity. The findings provide the first quasi-experimental evidence thatinvestigative accountability structures causally affect homicide clearance

The 2021--2023 European energy crisis caused by the war in Ukraine prompted broad policy interventions across energy markets. In contrast to the retail-side subsidies and public transfers implemented elsewhere, Spain and Portugal targeted the wholesale electricity market through the so-called Iberian solution. Combining wholesale market simulations with machine learning techniques, we quantify the distributional implications of the crisis and this intervention on Spanish electricity firms and consumers. The crisis shifted substantial wealth from consumers to generators: power plant profits rose by 29 Billion Euros. Even accounting for the Iberian solution and the accompanying energy-saving measures, the crisis remained highly regressive: the lowest income quintile lost 3.4% of disposable income, three times the relative burden of the highest quintile. Absent these interventions, the crisis would have been considerably more costly for all households: the policy package delivered proportionally larger relief to lower-income groups, at no net fiscal cost.

Why do politicians who switch parties closer to the election lose individual votes yet win seats at higher rates? We develop a formal model of switching timing under open-list proportional representation (OLPR) that resolves this paradox. Because OLPR electoral survival depends simultaneously on individual preference votes and pooled party quotients, the informational gain of joining a viable party late in the cycle can dominate the reputational costs of defection in low-information environments, generating what we call the Timing Paradox as a theoretical equilibrium result. We test these predictions using a natural experiment in Brazil, where a 2015 electoral reform introduced a 30-day party switching window six months before election day, exogenously shifting the permissible timing of defection closer to the election. Leveraging panel data on 12,108 city councilors across three election cycles (2012, 2016, and 2020), we estimate fixed-effects and conditional logistic regression models. The results support the Timing Paradox: switching under the post-reform regime reduces individual vote share by approximately 2%, while increasing the odds of reelection by 25%. We further show that these returns are amplified when the origin party suffers an acute reputational shock: PT-affiliated councilors who defected during the 2016 political crisis achieved reelection odds 47% higher than their loyalist peers. Together, the findings establish that the electoral consequences of party switching depend on the interaction between institutional timing rules and the information environment, with implications for partisan accountability and electoral reform in fragmented multiparty systems.

Centralized welfare programs limit local discretion over eligibility and payments, yet ac-cess often depends on municipal bureaucracies. We study whether local politicians obtainelectoral returns from a federally designed cash transfer when their remaining discretion liesin registration. In Brazil’s Bolsa Família, a 2009 reform changed the formula assigning munici-pal beneficiary quotas and generated differential authorized expansion for reasons unrelatedto local politics. In the 2,078 municipalities where the 2008 incumbent mayor sought reelec-tion in 2012, a one-standard-deviation quota shock created 171.50 additional authorized slotsper 10,000 residents and induced 66.16 additional beneficiary families per 10,000 residents.The IV estimate implies that 100 additional beneficiary families per 10,000 residents raise voteshare by 1.73 percentage points. Enrollment expanded both with and without a pre-existingCRAS welfare o?ice, but the electoral response appears only in CRAS municipalities. Withinthat group, the vote-share response is larger where o?ices report active outreach and moreintensive service to Bolsa Família families. In the manager comparison, the per-beneficiaryreturn can be pinned down only where the incoming mayor replaced the inherited front-linemanager. The quotashock predicts neither manager turnover nor new CRAS openings, and substitute managersdo not appear to be partisan loyalists. Among turnover municipalities, larger quota shocksare associated with a higher probability that the substitute later runs for o?ice. Centralizingeligibility and payment does not eliminate local political returns when program access stilldepends on municipal implementation.

To improve service delivery, central governments can tie intergovernmental transfers to local policy performance. While performance-based transfers incentivize local governments and generate efficiency gains, they also shift transfers from low- to high-capacity governments, creating welfare-reducing inequities. We study this equity-efficiency trade-off using a bundle of transfer reforms to Brazilian municipalities. When two states tied transfers to relative educational performance, student test scores rose substantially: moving from the 25th to the 75th percentile of per capita conditional transfers increased scores by 0.16 standard deviations. However, the reform also widened funding disparities across municipalities, which translated into disparities in municipal expenditures in multiple sectors. In contrast, contemporaneous reforms to unconditional transfers had negligible effects on student outcomes. We use a simple model of optimal transfers to interpret these findings. Our results suggest that the introduction of performance-based transfers delivered large efficiency gains, limited welfare losses from inequities, and was welfare-enhancing. We find minimal evidence of multitasking distortions or score manipulation. Instead, we document improvements in the quality of education inputs and find suggestive evidence of reduced corruption.

Do monetary policy surprises reflect exogenous shocks or central-bank information? We provide a direct test using data from Brazil. We examine whether surprises around monetary policy meetings predict subsequently released GDP and inflation surprises. Because the outcomes underlying these releases predate the policy decision, any predictive relationship reflects information rather than causation. We find that unexpected changes in the current policy rate do not predict subsequent GDP or inflation surprises, but a path factor capturing revisions in expected interest rates over the following months significantly predicts GDP surprises. The evidence for inflation is weaker, partly because the relevant sample is considerably smaller.

We examine a critical case of regulatory arbitrage in the Brazilian banking sector, where a tax asymmetry created incentives for overhedging foreign currency risk. We demonstrate how this tax distortion, which exempted foreign investments from taxation while taxing the financial hedge, led banks to take on excessive hedge positions. The overhedge is a form of carry trade. Therefore, it influenced foreign exchange (FX) market dynamics and posed a significant challenge to financial stability. Our empirical analysis reveals that banks exploited this regulatory loophole to generate near-risk-free profits, contributing to heightened systemic risk.To explore the broader macroeconomic consequences, we develop a two-country, open-economy model that integrates financial frictions and the observed tax distortion. The model shows that such overhedging activities amplify macroeconomic volatility, intensify the impact of financial shocks, and worsen exchange rate depreciations during crises.We also use the model to assess the effectiveness of policy responses, such as FX interventions and macroprudential measures. Our results indicate that central bank interventions in the FX market can mitigate the adverse effects of overhedging by stabilizing the exchange rate and reducing welfare losses. Therefore, we provide important insights for policymakers regarding the design of prudential regulations and the importance of a coherent tax framework to prevent regulatory arbitrage and safeguard financial stability. The paper contributes to the literature on banking regulation, systemic risk, and the role of policy (FX) interventions in emerging markets.

This paper investigates the aggregate and dynamic effects of exchange rateand cattle price fluctuations on deforestation in the Brazilian Amazon. Wepropose two novel external instruments and use them in a Structural VectorAutoregression (SVAR) framework. We uncover two new channels throughwhich exchange rate depreciations increase deforestation. Through a revenuechannel, depreciations raise cattle prices in domestic currency persistently, in-creasing incentives for cattle expansion. By contrast, cattle price shocks trig-ger exchange rate appreciation, offsetting price increases in local currency.Through an enforcement channel, depreciations do not lead to stronger eco-nomic activity or environmental enforcement. Cattle price shocks, in turn,boost economic activity, which is followed by stronger environmental enforce-ment. This underscores the importance of macroeconomic conditions, beyondglobal commodity markets alone, in shaping deforestation dynamics.

This article adapts Auclert’s (2019) redistribution-channel framework to Brazil using a Bayesian New Keynesian DSGE model with Ricardian and non-Ricardian households, financial frictions, endogenous banking spreads, macroeconomic uncertainty, and public transfers, estimated on quarterly data for 2010Q1-2025Q3. A contractionary monetary shock reduces output, inflation, and aggregate consumption, with non-Ricardian consumption falling much more sharply than Ricardian consumption. Consequently, the consumption gap widens on impact, while the income gap briefly turns negative before converging back toward zero. An Auclert-inspired diagnostic decomposition shows that earnings heterogeneity is the main source of the widening consumption gap, partly offset by intertemporal substitution; Fisher redistribution, transfers, and unhedged interest-rate exposure play smaller roles. Variance and historical decompositions reveal that transfer shocks dominate consumption-gap fluctuations, whereas monetary-policy shocks dominate the income gap, with uncertainty and sovereign-risk shocks becoming important during episodes of macro-financial stress. Monetary policy in Brazil is therefore not distributionally neutral, and consumption- and income-based asymmetries reflect distinct transmission mechanisms.

This paper studies how extreme droughts generate inflation inequality across income groups and regions. We develop a regional heterogeneous-household New Keynesian model in which drought shocks affect agricultural productivity, raise marginal costs in the food sector, and transmit to consumer prices through household-specific consumption baskets and regional supply linkages. The model is calibrated to Brazilian states using household expenditure data, regional supply-use tables, satellite-based precipitation measures, and agronomic estimates of crop sensitivity to water conditions. The results show that extreme droughts increase the inflation faced by lower-income households more than that faced by higher-income households. Under average crop water sensitivity, local drought shocks raise twelve-month inflation inequality by about 0.10 to 0.30 percentage points in most states. When shocks occur during critical stages of the crop cycle, the effects rise to about 0.20 to 0.50 percentage points. The impact is strongest in states where lower-income households devote larger expenditure shares to food, consumption is more dependent on locally produced goods, and regional market integration is weaker. Counterfactual exercises show that reducing home bias in food consumption mitigates, but does not eliminate, the distributive effects of drought shocks. Robustness exercises using severe, rather than extreme, drought episodes preserve the qualitative results. The simulated effects on inflation inequality are also positively associated with food insecurity, especially its severe form. The findings indicate that climate shocks can act as a structural source of inflation inequality, with regional integration and consumption diversification operating as partial risk-sharing mechanisms.

We document new facts on how Brazil’s demographic transition reshapes labor-market composition over the life cycle. Using household survey microdata and administrative demographic statistics, we characterize age profiles of formal employment, informality, and self-employment and quantify how shifts in the age distribution affect aggregate workforce composition. We develop a multi-period overlapping-generations model with incomplete markets, three labor-market sectors (formal, informal, self-employment), and an endogenous retirement decision. Calibrated to Brazilian moments in 2023, the model implies that aging lowers formal employment, raises self-employment and retirement, and affects macro aggregates such as labor, output, and savings. Policy experiments increasing the minimum retirement age and/or required contribution years mitigate these reallocations by raising formal participation among older cohorts.

This paper explores to what extent inflation persistence in Brazil is affected by fiscal disturbances perceived as unfunded, that is, not backed by future primary surpluses in an environment with fiscal-monetary interaction. The paper proceeds in two steps. First, it documents reduced-form evidence from rolling half-life measure of inflation persistence and compares Brazil with a set of peer economies. The results indicate that Brazilian inflation is more persistent in normal times, but that the post-2020 surge in persistence unwinds faster in Brazil than in most peers. Second, it estimates a DSGE model with partial fiscal backing for Brazil. The results show substantial but incomplete fiscal backing in the country: about 36 percent of aggregate shocks load on the unfunded, fiscal-led component. These findings suggest that Brazil remains predominantly monetary-led, a regime in which fiscal debt is stabilized by expected primary surpluses and real rates, while fiscal valuation effects play a meaningful role in inflation dynamics and persistence.

We examine the effects of rising steady-state debt-to-GDP ratios on the output--inflation trade-off faced by central banks. To this end, we calibrate a heterogeneous agent New Keynesian (HANK) model for Brazil, a country characterized by historically high and persistently projected levels of public indebtedness. We show that natural interest rates rise in the presence of fiscal imbalances. Beyond this long-run effect, high levels of government debt also worsen macroeconomic trade-offs. For debt ratios below unity, the deterioration is relatively mild; however, at very high levels of indebtedness, the trade-off implies substantially deeper recessions are required to achieve a given reduction in inflation.

Relying on a micro-level dataset of US firms' inflation expectations, we document that firms in industries with larger Domar weights and more flexible prices have more accurate inflation forecasts. To explain these patterns, we develop a rational inattention model of price-setting firms within a production network. In equilibrium, firms' inflation forecast accuracy depends on their attention to marginal costs, the comovement between their marginal cost and aggregate inflation, and the change in the marginal cost. When calibrated to US input-output data, the model replicates the cross-sectoral relationship between forecast accuracy, Domar weights, and price flexibility. Quantitatively, we find that production networks endogenously increase nominal rigidity. This channel accounts for about 30 percent of the network's attenuation of the monetary transmission on inflation, deepening the standard flattening of the Phillips curve from input-output linkages.

This paper develops a quantitative framework to analyze how informal labor markets shape employment dynamics over the business cycle in developing economies. I construct a search-and-matching model with heterogeneous workers, endogenous vacancy creation in formal and informal sectors, and aggregate productivity shocks. The model’s block-recursive structure provides a tractable way to incorporate both worker heterogeneity and aggregate uncertainty. Calibrated to the Brazilian labor market, the model reproduces key qualitative patterns: informality and unemployment are countercyclical, low-skilled workers are more exposed to informal employment, and informal jobs are easier to access but less stable than formal jobs. The quantitative exercise highlights an equilibrium trade-off between search access and job stability. Informal submarkets expand the set of opportunities available to unemployed workers, especially when formal vacancy creation is weak. At the same time, they expose workers to higher separation risk. In a counterfactual economy that shuts down informal vacancy creation, workers lose this adjustment margin, but employment shifts toward more stable formal matches. Under the current calibration, the stability margin dominates: aggregate unemployment falls modestly and low-skilled welfare rises. These results emphasize that the macroeconomic role of informality cannot be summarized solely by its effect on employment levels. Equally important is how informality affects job-finding prospects, match survival, and the distribution of gains across workers over the cycle.

I develop a labor market framework in which monopsony power, firm informality, and self-employment coexist, a salient feature in developing economies’ labor markets. I build a quantitative model of heterogeneous firms with endogenous entry and formalization decisions. Firms choose whether to operate formally or informally, while workers choose among formal wage work, informal wage work, and self-employment. From the firm’s perspective, labor supply elasticity is endogenous to competition with other firms, substitution across local markets, and the availability of self-employment as an outside option. I calibrate the model using Brazilian administrative and survey data and quantify the equilibrium effect on welfare stemming from firm informality (+1%), self-employment (-6%) and market power (-5%). The channels interact: in the absence of market power, firm informality becomes welfare reducing (-0.7%) and self-employment’s negative impact is strongly dampened (-0.8%). A payroll-tax cut financed by higher corporate profit taxation improves welfare by simultaneously raising equilibrium wages, reallocating labor towards formal firms, and reducing self-employment.

This paper investigates the impact of trade liberalization on structural transformation, leveraging quasi-experimental tariff reductions from Brazil’s 1990s reforms. Building on recent cross-regional macroeconomic literature, we extend the traditional shift-share framework to estimate both local and spatial spillover effects on sectoral employment shares, providing insights into aggregate outcomes. Regions more exposed to tariff reductions experienced a reallocation of employment away from manufacturing and non-tradables and toward primary activities. Consistent with a comparative-advantage channel, the shift toward the primary sector is stronger in regions with higher agricultural potential. Spillover effects — mediated through migration or gravity — partially offset local deindustrialization in manufacturing but amplified declines in non-tradables. These effects persisted over two decades and remained robust to alternative structural transformation hypotheses, concurrent economic shocks, and controls for expenditure exposure. Overall, our findings support theories highlighting comparative advantage as a key driver of trade-induced structural transformation.

Market power among firms is a common feature in many economies, including in developing countries where a significant share of firms operate informally. This paper examines the interaction between market power and informality, and how this interaction shapes the impact of government policies. We develop a general equilibrium model with firm heterogeneity and oligopolistic output markets, in which firms endogenously decide whether to operate formally or informally. Informal firms evade taxes but incur enforcement costs. We calibrate the model using comprehensive administrative data from Brazil and perform several counterfactuals.

This paper develops a unified data-driven framework for assessing changes in the welfare of a representative consumer over time. The framework combines structural break detection with welfare analysis by interpreting each estimated pair of deterministic consumption growth rate and shock volatility as representing a distinct economy. We apply the framework to U.S. consumption data over the period 1834--2019 by estimating regime-switching ARMA models and find evidence of a single common structural break in the late 1940s. The postwar economy is characterized by both higher consumption growth and lower volatility, leading to a substantial improvement in representative-agent welfare. To quantify the contribution of each factor, we perform a decomposition based on counterfactual exercises analogous to the analyses of the welfare cost of business cycles and the welfare cost of reducing growth developed by \cite{Lucas1987}. Our results indicate that most of the welfare gains are attributable to higher growth, although the contribution of lower volatility becomes increasingly important as the coefficient of relative risk aversion rises. The findings are robust to adjustments that reduce measured prewar volatility to account for potential historical measurement error.

This paper examines the relevance of monetary policy shocks for fluctuations in macroeconomic variables in Brazil. We construct an external instrument from a monetary policy reaction function, estimate a Structural Vector Moving Average model with external instruments (SVMA-IV), and compute Forecast Variance Ratios. Our findings indicate that monetary policy shocks contribute substantially to the forecast variance of the Selic rate and moderately to inflation, while they appear to be only weakly relevant for forecasting output and the exchange rate. These results generally hold when the instrument is obtained using a high-frequency approach, but they differ with yield-curve-based instruments. This underscores the importance of considering different dimensions of monetary policy when assessing its macroeconomic effects.

We construct an index to measure monetary policy uncertainty in Brazil (MPU-Br) from the textual analysis of over 1.2 million newspaper articles published between 2007 and 2025, using natural language processing techniques. The validity of the index is assessed through three complementary strategies: a comparison with a proxy based on the dispersion of market expectations for the Selic rate, which shows a correlation of 0.55 through the end of 2022; a human audit of more than 6,000 news articles, whose manual classification displays a correlation of 0.76 with the MPU-Br; and a historical validation, in which the main peaks of the series coincide with relevant episodes of uncertainty in the conduct of Brazilian monetary policy. We then estimate the effects of a shock to the MPU-Br using a Bayesian Structural VAR, and the results indicate significant responses, with a decline in inflation, in the one-year interest rate and in the Ibovespa, as well as an increase in overall Brazilian economic uncertainty. Finally, we investigate one of the main transmission mechanisms of monetary policy uncertainty to the real economy: the financial frictions channel. We find evidence that increases in the MPU-Br are transmitted to credit conditions persistently, raising bank spreads and reducing the volume of credit granted to the productive sector, operating through both the price and the quantity side of credit.

Materials policy in the United States and elsewhere treats recyclable materials as a largely undifferentiated category: no material-specific virgin taxes, recycling subsidies and consumption taxes on recyclables are nearly absent. Yet materials are highly heterogeneous in the share of extracted material that is ever usable, in how readily recycled inputs substitute for virgin ones, and in the landfill damage each imposes at disposal. We develop a dynamic general equilibrium model of the material cycle in which a final good is produced from capital, labor, and a menu of materials, each combining virgin extraction and recycling in an endogenously chosen mix, and characterize the material-specific recycling subsidies and virgin material taxes, together with a uniform consumption tax, that implement the efficient allocation as a competitive equilibrium. We calibrate the model separately for five material categories -- ferrous metals, non-ferrous metals, glass, paper, and plastics -- accounting for the majority of U.S. virgin material extraction and municipal solid waste. The resulting heterogeneity is large: the optimal recycling subsidy ranges from 0.89 percent (Non-Ferrous) to 33.66 percent (Ferrous), while the optimal virgin material tax ranges from 4.07 percent (Paper) to 329.78 percent (Plastic); the optimal consumption tax, by contrast, is small and uniform at 0.59 percent. These findings indicate that a single, material-blind instrument cannot approximate the efficient policy for any of the three margins we study, underscoring the need for materials policy to be designed at the level of the material rather than the waste stream as a whole.

We study the long-run evolution of real interest rates in small open economies with imperfect financial integration. We develop a quantitative overlapping-generations model with imperfect capital mobility that provides a tractable decomposition of domestic real interest rates into global and domestic drivers. We calibrate the model to South Korea and show that it successfully reproduces the observed evolution of the real interest rate, current account, and investment since 1980. A structural decomposition indicates that both global and domestic factors contributed substantially to the decline in Korea's real interest rate, even in an environment with high financial openness. Forward-looking projections suggest that population aging will continue to exert downward pressure in the coming decades. A recalibration to Sweden demonstrates that the framework also performs well in an economy with markedly different demographic dynamics.

Migrants in developing countries face substantial uncertainty over earnings at potential destinations, yet the causal role of this risk in shaping bilateral flows has remained elusive. Using 2000 and 2010 Brazilian Demographic Census data on microregion-to-microregion flows, we estimate the causal effects of destination mean and variance of log earnings on bilateral migration. Identification exploits two orthogonal shift-share instruments---the 2003 legalization of genetically engineered soy and the post-2000 China import shock---that move the two destination moments in distinct ways. Using a control-function PPML estimator, we find that destination earnings risk deters migration at roughly twice the rate that higher mean earnings attract it. Embedding these estimates in a quantitative spatial model with CRRA preferences and Diamond-Mortensen-Pissarides frictional labor markets, we recover structural parameters and use the calibrated model to decompose the mean and variance channels. Destination earnings risk is a first-order deterrent to internal migration, which suggests that policies compressing destination earnings dispersion---such as unemployment insurance---can act as indirect tools for encouraging productivity-enhancing migration.

Expectations are central to macroeconomics and finance. Standard models assume forecasters operate in isolation, learning only from their own information and from public data, but never from other forecaster. I study a channel where they do learn from each other: when the media reports the inflation forecast of a prominent — salient — institution, others can partially infer its idiosyncratic view, so that one institution’s private noise can move the whole distribution of expectations. I extend a standard model of expectation formation with what I call salient signal and bring it to a panel of Brazilian institution-level inflation forecasts (2010–2021), using a large language model to identify the days a forecaster is publicized to others and exploiting the differing volatility of market-implied inflation on those days to identify the effect. Forecasters do not react on the day a peer’s view becomes news; their expectations then move gradually toward it, reaching a one-for-one response within about three weeks and overshooting thereafter. One institution’s idiosyncratic information thus leaves a lasting footprint on aggregate professional expectations.

In this paper, we analyze sector-differentiated subsidies in production and investment networks. Our work expands Liu’s (2019) model of industrial policy in production networks to include the economy's investment goods network. We show how Liu’s main theoretical results must be modified when two distinct networks connect the sectors of the economy. We then present two applications of this theory to Brazil. Our first finding is that incorporating the investment network leads to changes in the sector priority ranking indicated by the theory. For our second application, we analyze the impact of sector-differentiated federal tax exemptions in Brazil in 2019, which were estimated at R$ 265 billion, or 3.6% of Brazil's GDP. We calculate that the impact of these subsidies on aggregate income was nearly zero. However, in a counterfactual scenario where subsidies are redistributed among sectors based on the theory-derived ranking, income increases by up to 1.7% within the investment-augmented production network.

Urbanization in developing countries has generated sharp spatial inequalities within cities, with important consequences for earnings and intergenerational mobility. While the literature emphasizes slums as an equilibrium response to high urban housing costs, this paper shows that distant poor neighborhoods (formal residential areas located far from city centers) constitute a distinct and quantitatively important urban margin. Using Brazilian data from 1980 to 2010, we document that distant poor neighborhoods (DPNs) host a large and growing share of the urban population and that, despite better housing conditions than slums, residents face earnings penalties and children experience lower upward educational mobility due to weak access to jobs and schools, respectively. To interpret these patterns, we develop a dynamic general equilibrium model with residential sorting across rural areas, slums, DPNs, and cities, in which housing costs, commuting frictions, labor market access, and schooling jointly determine earnings and mobility. Quantitative policy experiments show that eliminating slums without addressing spatial frictions reallocates households toward DPNs, dampening aggregate earnings and mobility, whereas policies that integrate DPNs into the urban core (through transportation improvements or school integration) generate sizable mobility gains. The results highlight DPNs as a central but overlooked margin in urban policy and show that cities’ capacity to function as engines of opportunity depends critically on spatial access to jobs and education.

We study how sovereign risk weights shape the transmission of monetarypolicy to bank credit in an emerging-market small open economy. In a New KeynesianDSGE model calibrated to Brazil, a representative bank holds long-duration nominalgovernment bonds and floating-rate government debt indexed to the central bank'spolicy rate (the Selic rate), while facing a regulatory capital floor. A monetarytightening lowers the price of long bonds, generating mark-to-market losses thaterode bank capital. When sovereign debt carries a low risk weight, these losses reducethe numerator of the capital ratio without commensurately reducing risk-weightedassets; once the floor binds, the bank contracts private credit.The central result is an activation frontier in the space of sovereign exposureand the effective capital buffer. Below the frontier, the same tightening has mutedbalance-sheet effects; above it, credit contracts sharply, with a contraction an orderof magnitude larger when sovereign debt is zero-weighted than when it is fullyweighted. Mapping supervisory balance-sheet data into the model, direct securitiesholdings place the Brazilian banking system near the activation threshold, while abroader repo-inclusive exposure measure places it inside the constrained region. Twofurther results sharpen the mechanism. A larger share of policy-rate-indexed debtattenuates the channel by reducing duration exposure. For the same cumulativedisinflation, a front-loaded tightening produces larger duration losses and a strongercredit contraction than a gradual one, especially when the bank portfolio is tiltedtoward long-duration debt. The mechanism does not rely on sovereign-risk mispricing,distorted beliefs, default shocks, or disaster shocks. The key policy margin is thesovereign risk weight itself rather than the fine design of a smooth macroprudentialrule.

Modeling business cycles in Brazil is particularly challenging due to recurring structural shifts, external shocks, and successive stabilization plans. We model GDP growth from 1980 to 2025 using a Markov-Switching specification that combines independent variance filtering with an absorbing block transition matrix. This flexible framework provides a statistically better fit than standard Markov-Switching models previously applied to Brazilian data, while isolating temporary high-volatility episodes and preventing outlier distortion in business cycle classification. Without imposing predetermined break dates, the model endogenously identifies a structural break in 1999, separating two distinct statistical environments. The break coincides with the adoption of a floating exchange rate and inflation targeting regimes, which have remained the cornerstones of Brazilian macroeconomic policy to this day. Before the break, business cycles were characterized by higher volatility, weaker expansions, and shorter, more frequent recessions. Afterward, volatility declined, expansions strengthened markedly, and recessions became more persistent. To the best of our knowledge, this is the first application of this framework to Brazilian GDP. The model closely reproduces the official CODACE chronology and provides a flexible framework for analyzing economies characterized by structural breaks and periods of high volatility.

I construct a measure of global temperature shocks and examine their distributional effects on labor income growth by workers’ formality status, using individual-level Brazilian data. To disentangle the roles of the extensive and intensive margins, I also analyze both the distributional effects on labor income growth among workers who remain employed and transitions between and within different employment states---formal employment, informal employment, and unemployment. I document five main facts: (i) for both formal and informal workers, the effects on labor income growth are negative and U-shaped across the labor income distribution, with high-labor income workers being the least affected; (ii) informal workers are, on average, more adversely affected than formal workers; (iii) the estimates imply higher inequality in both sectors, with a stronger increase among informal workers; (iv) the effects operate mainly through the intensive margin, reflecting changes in labor income growth among those who remain employed; and (v) global temperature shocks leave transition probabilities among initially formal workers largely unchanged, but increase the probability of persistence in the informal sector, mainly at the expense of a lower probability of transitioning from informal to formal employment.

This paper studies to what extent international trade accounts for recent trends in business dynamism and its implications for labor misallocation. Using Brazilian administrative data, I document that an aggregate decline in business dynamism conceals substantial heterogeneity across sectors, closely mirroring exposure to the China Shock. I implement a shift–share strategy to confirm that trade exposure causally affects firm entry, exit, and local output per worker. To study aggregate implications of these patterns, I develop and calibrate a multi-sector model of oligopsonistic labor markets with firm dynamics and heterogeneous demand shocks. Aggregate effects depend both on shock magnitude in each market and on the preexisting profile of market power, with shock incidence determining the qualitative response. Trade shocks explain between a fifth and a quarter of the difference in firm aging across sectors; negative demand shifts in large sectors increase labor market concentration and misallocation, more than offsetting efficiency gains from positively affected sectors.

Using longitudinal data from Add Health, this research connects intensive video game use in early adulthood with negative outcomes later in life, such as lower earnings, worse mental health, and the absence of a partner and children. To interpret these facts, a framework is developed where gaming provides immediate gratification but is addictive, causing individuals to shift away from activities that support long-term economic and social achievement. The implications of this framework are expressed through lemmas. The model is calibrated to align with key trends observed in the Add Health data and demonstrates a strong fit. Several policy simulations aimed at discouraging video gaming during early life are also carried out.

This paper implements the Multivariate Core Trend (MCT) framework to analyze thestructural layers of inflation persistence and volatility in Brazil, using a high-dimensionaldataset of seasonally adjusted monthly Broad National Consumer Price Index (IPCA)data from August 1999 to May 2026. The state-space filtering model with stochasticvolatility and outlier adjustments processes the specific inflation rates and time-varying ex-penditure weights of over 350 subitems across the nine IPCA groups. The main empiricalfindings are: (i) the aggregate MCT-IPCA successfully isolates underlying trend inflationby smoothing out short-lived disturbances; (ii) there is an asymmetry between the volatil-ity components, with transitory noise being more than three times greater than permanentvariance; (iii) the long-run trend is heavily dominated by sector-specific innovations ratherthan common macroeconomic drivers; (iv) the structural layers of long-run persistence andshort-run transitory volatility behave as near-perfect inverted mirror images, where priceswith intense high-frequency volatility (e.g., Transportation and Housing) lack permanentmemory, while others (e.g., Household Goods and Health and Personal Care) exhibitminimal noise but a persistent trend; and (v) an out-of-sample multi-horizon forecastingexercise via a machine learning algorithm confirms the superior predictive accuracy andstructurally lower systematic tracking bias of the aggregate MCT-IPCA signal. There-fore, while aggregate demand management via Selic interest rate modifications effectivelydisciplines cyclical components coupled with the macro trend, over-tightening policy tooffset supply-driven idiosyncratic shocks might be applied with caution by the monetaryauthority.

This paper estimates the dynamic macroeconomic effects of sovereign default risk shocks. Identifying these shocks, however, is difficult because default risk is endogenous and typically rises when fundamentals deteriorate. We address this challenge by exploiting the 2002 Brazilian presidential election, an episode in which electoral prospects abruptly moved investors’ perceived probability of default. Building on a detailed narrative account of the episode, we identify default risk shocks in a structural VAR through narrative sign restrictions. We find that these shocks are stagflationary: they raise sovereign spreads, depreciate the exchange rate, increase inflation, and contract output, while also tightening monetary and credit conditions and generating persistent declines in investment and foreign capital inflows. Quantitatively, default risk shocks account for about half of the fluctuations in the exchange rate and for substantial shares of the variation in output and inflation. Our results show that default risk can be costly even when default never materializes.

We study the redistributive consequences of trade policy in a quantitative general-equilibrium model that combines the task-based offshoring framework of Grossman and Rossi-Hansberg (2008) with a Bewley-Huggett-Aiyagari incomplete-markets economy. In this environment, an increase in offshoring costs affects the wages of low-skilled workers through three channels: a labor-supply effect, which raises them, and productivity and capital-supply effects, which lower them—the latter being absent when capital is exogenous, as in the original Grossman and Rossi-Hansberg model. We calibrate the model to the United States and evaluate an offshoring-cost shock inspired by the 2025 'Liberation Day' tariffs. Higher trade barriers do reduce income inequality, but only slightly, and the adjustment occurs almost entirely through a lower skill premium rather than through capital income. These gains come with a substantial efficiency cost: aggregate output falls 2.81% and welfare 1.62%. At the disaggregated level, all groups lose income, wealth, and welfare, with the poorest households losing proportionally more wealth than the richest. Protectionism can therefore be rationalized as a redistributive policy, but one whose equity gains are small relative to its efficiency costs.

Contrary to the prediction of the traditional uncovered interest parity mechanism, a 50-basis-point contractionary shock to Brazilian monetary policy depreciates the domestic currency and, simultaneously, raises the sovereign risk premium. Following the shock, the BRL/USD exchange rate depreciates 3.64% on impact, while the EMBI+ and the five-year sovereign CDS rise significantly—a joint response consistent with risk-premium transmission, in which the perception of sovereign risk dominates the interest rate differential. The yield curve reinforces this evidence: the response of rates increases systematically with maturity up to five years, exhibiting greater amplification than that documented for developed economies. In the medium run, the exchange rate, the sovereign risk measures, and the yield curve display a highly synchronized reversal dynamic, providing additional evidence of a common transmission mechanism. These results contrast with recent evidence obtained through identification via heteroskedasticity at daily frequency, which rejects the fiscal dominance hypothesis for Brazil, suggesting that the importance of the risk-premium channel depends on the data frequency and the identification strategy. The results are obtained from a Structural Dynamic Factor Model (SDFM) identified by an external instrument, combining monetary surprises around Copom meetings, orthogonalization with respect to predetermined public information, and a sign filter that separates monetary policy shocks from central bank information shocks, using a panel of 106 Brazilian macroeconomic and financial series between 2013 and 2025.

We decompose 205 FOMC announcements (2000–2024) into pure Monetary Policy (MP) and Central Bank Information (CBI) shocks, following Jarocinski and Karadi (2020), and trace their transmission to equity markets in Germany, Brazil, Hong Kong, and India using Bayesian local projections with endogenous structural-break detection and cross-fitting for honest inference. MP shocks generate a trilemma-consistent hierarchy of negative responses governed by monetary-policy autonomy: Hong Kong's currency board absorbs the full blow, while India's managed float largely insulates against it. CBI shocks overturn this ranking entirely: Brazil, a commodity exporter whose earnings complement U.S. growth, responds most strongly and positively, while India's competing IT-services and financial sectors respond negatively to the same nominally favorable news. We label this an 'informational trilemma,' governed by productive-structure complementarity rather than exchange-rate architecture. Despite dramatic shifts in the Fed's conduct—from conventional targeting through the zero lower bound, quantitative easing, and the sharpest tightening cycle in four decades—both the MP and CBI transmission architectures remain structurally stable over a quarter-century. For emerging-market policymakers, the implication is that appropriate responses to Fed announcements hinge on shock decomposition, not direction, and that durable insulation requires structural reform of currency and capital-market exposures rather than cyclical policy coordination.

Word embeddings are vectors of real numbers associated with words, designed to capture semantic and syntactic similarity between the words in a corpus of text. We estimate the word embeddings of the European Central Bank's monetary policy statements by using a simple natural language processing model (Word2Vec), based only on the information and model parameters available as of each press conference. We show that a measure based on such embeddings contributes to improve core inflation forecasts multiple quarters ahead. Other common textual analysis techniques, such as sentiment or topic analysis, do not obtain the same results. At the same time, the performance is competitive with that of much more sophisticated large language models (LLMs), although the model we use avoids look-ahead bias. The information contained in the embeddings remains valuable for out-of-sample forecasting even after controlling for the central bank inflation forecasts, an important input for the monetary policy statements.

Using comprehensive data from the Brazilian equity market (2017-2022), we document that retail investors act as unintentional liquidity providers on days of high adverse-selection risk — accidental market makers. On such days, professional market makers withdraw while retail investors expand their passive limit-order volume, absorbing institutional demand precisely when it is most costly to supply. Retail investors fail to internalize the winner's curse embedded in passive order execution: day-zero returns on passive retail orders reach -59 basis points on buying days and -58 basis points on selling days, with no subsequent reversal, implying an annual wealth transfer of approximately USD 0.2 billion. A shift-share instrument based on regional flows into Brazil's government bond platform yields a 2SLS point estimate of 1.26 standard deviations improvement in Amihud liquidity per one-percent increase in retail volume, about seven times the OLS estimate of 0.18, consistent with retail investors participating disproportionately when market liquidity is already deteriorated.

We test whether lowering borrower switching costs intensifies bank competition. We exploit a 2014 reform by Brazil’s central bank that created a national framework for loan portability, allowing borrowers to refinance existing loans at rival banks. Using an event-study difference-in-differences design across municipalities, we find that effective annualized borrowing rates on payroll loans decline by about 19 basis points on average relative to the pre-reform mean, with larger reductions in ex-ante less concentrated markets. Credit quantities respond in the predicted direction: payroll-loan credit per capita rises by approximately 6\% overall and by up to 10\% in more competitive markets. Heterogeneity aligns with portability frictions, with larger effects for borrower groups and markets where outside options are more viable. Pre-treatment coefficients are small and statistically indistinguishable from zero, and the results are robust to alternative exposure and timing definitions. The joint price--quantity pattern indicates that reducing switching costs increases market contestability, compresses bank markups, and expands credit in the most exposed segments.

Fund portfolios reveal managers' private information and have long been known to predict future stock returns. The literature, however, relies mostly on quarterly disclosures and cannot measure how much information is lost when holdings are observed less frequently. For the first time, we identify this information cost using a unique monthly census of 3,000 Brazilian equity funds between 2005 and 2025. Holding fixed the forecasting procedure, the evaluation horizon, and the sample, we compare identical signals built from monthly and from simulated quarterly holdings. Moving from monthly to quarterly observations reduces the information coefficient by 14% and the portfolio alpha by 26%, and the loss stems mainly from staleness of the holdings rather than from less frequent updating of managers' skill. The signal is built from the active deviation, defined as the difference between a stock's weight in the fund's portfolio and its weight in the consensus portfolio, aggregated according to managers' historical accuracy. The predictive content is concentrated in smaller-capitalization stocks, where value-weighted abnormal returns remain economically significant, consistent with the limits of arbitrage. The results show that the informational value of portfolio disclosures depreciates rapidly and provide the first direct estimate of the cost of lower disclosure frequency. These findings contribute to regulatory debates on portfolio transparency and demonstrate that disclosure frequency is an economically important dimension of financial market information.

This paper examines how firms’ financial leverage shapes the transmission of bank credit supply shocks to employment. Using matched credit registry (SCR), employer-employee (RAIS), and firm-level accounting (Alexandria) data for Brazil between 2013 and 2022, we estimate exogenous, time-varying bank-level credit supply shocks following an industry-location-size-time fixed-effects approach and study their effects on firm employment growth. Over the full sample period, the average employment response to credit shocks is economically small and statistically insignificant. However, this masks substantial heterogeneity across macroeconomic regimes: during the 2013-2016 period of public credit expansion, positive credit shocks significantly increased employment growth, with the effect more than twice as large for highly leveraged firms; during the 2016-2017 recession, marked by a simultaneous contraction of public and private credit, adverse shocks reduced employment growth, again disproportionately among highly leveraged firms. No significant effects are found during the 2018-2020 recovery or the 2020-2021 pandemic period. We further show that these amplification effects are concentrated among small and medium-sized firms during the credit expansion and among large firms during the recession. Our findings indicate that financial leverage amplifies the employment effects of credit supply shocks, but only during episodes of pronounced aggregate credit fluctuations, highlighting the state-dependent nature of the credit channel of employment

This paper examines liquidity and price formation in Brazil’s FX futures market and whether the four daily windows used to calculate the official PTAX exchange-rate benchmark temporarily affect these dimensions. Using more than 8 billion raw order-level observations for B3 (Brasil, Bolsa, Balcão) U.S.-dollar (USD) mini futures from 2016 to 2019, we reconstruct the first level of the limit order book and analyze bid--ask spreads, order-flow-imbalance (OFI)-based price impact, efficient-price volatility, and informational leadership between the best bid and ask. Across the sample, spreads remain close to the minimum tick and display limited persistence, while price impact declines after 2017. PTAX windows do not systematically worsen spreads or price impact, but efficient-price volatility rises during the first three windows, with smaller effects later in the day. Information Share, Component Share, and a wavelet-based measure show frequent shifts in informational leadership, with no persistent dominance by either side. The results indicate that the market generally absorbs benchmark-related trading pressure without broad or lasting deterioration in liquidity, although PTAX concentrates information-driven price adjustment in specific intraday intervals.

We estimate a horizon-specific risk premia for seven US macroeconomic and financial factors. The Bayesian hierarchical model extracts latent priced shocks from returns and traces their propagation into macroeconomic variables. Using returns from currencies, bond futures, and government bonds, we find that four out of the seven factors have statistically significant premia: CPI inflation, the Fed Funds Rate, the yield curve slope, and the US-Europe yield differential. In each case, premia are small at short horizons and grow with the investment horizon. The other three factors, GDP, the real yield, and the VIX are not significantly priced. This pattern contrasts sharply with the literature’s equity results, where real-activity variables bear the dominant long-horizon premia. In our bond-and-currency cross section, the priced disturbance is concentrated in the nominal-monetary block rather than in real activity. We derive a closed-form decomposition showing that the model-implied mimicking portfolio preserves the same asset-space direction across all horizons, with only its leverage changing. The Bayesian implementation delivers coherent inference over the entire term structure and over portfolio weights in a setting where non-parametric alternatives would be underpowered.

This article studies how news and implied volatility (IV) influence volatility predictionsfor the Brazilian stock exchange Ibovespa index ETF and five widely traded Brazilian stocks.We proposed a model which integrates IV and LDA-derived news topics and employed ran-dom forest to tackle nonlinearities and high dimensionality. Our results indicate that a modelrelying only on external variables (IV and news topics) significantly improves forecast accu-racy for horizons beyond one day compared to autoregressive models, even when accountingfor asymmetries and discontinuities. Our analysis of variable importance shows that newstopics are crucial for long-term forecasts, while IV significantly influences short-term pre-dictions. This paper contributes to the literature by underlining the importance of textual dataanalysis and IV in volatility forecasting

Do bettors misperceive expected returns, and does bringing their beliefs closer to the regulatory benchmark change betting and financial allocation decisions? Using a survey experiment in Brazil, we document that bettors are motivated by financial gain and entertainment and expect to receive approximately R$1.75 for every R$1 wagered, far above the R$0.85 regulatory benchmark. A randomized information treatment lowers bettors' reported beliefs about expected returns by 28% to 37% but reduces the share of windfall income allocated to betting by only 2.1 percentage points. The treatment has a larger effect on the probability of allocating any money to betting than on the amount allocated among those who continue to bet. A parsimonious model rationalizes these findings: when betting provides substantial direct utility, more accurate beliefs have limited effects on betting expenditure and saving. Still, reducing overconfidence limits expected wealth losses, because overconfident bettors consume more today in anticipation of betting winnings that do not materialize on average. Information disclosure may therefore make beliefs more accurate without substantially reducing betting expenditure.

Can a new payment technology boost other payment methods and improve banking outcomes? Using individual-level data from Brazil, we study Pix, a real-time payment system, and its impact on financial behavior. Exploiting exogenous flood shocks, we show Pix plays a key role in informal insurance and expands the use of financial products. Pix complements, rather than replaces, traditional methods like card payments, and increases bank deposits and loans. Non-bank financial institutions also benefit without siphoning funds from banks. Our findings suggest that digital payment technologies can promote financial inclusion and strengthen the broader financial system.

Asset managers intermediate retail savings through fund networks in which feeders raise capital for masters that trade stocks. Using consolidated Brazilian fund data, I define a manager's exposure to a stock as its network's direct and indirect participation in outstanding shares and estimate a Bayesian dynamic demand model manager by manager. A granular network instrument identifies each manager's exposure elasticity, the equilibrium response of its exposures to rivals' demand shocks; positive elasticity defines replicant behavior. Consistent with a game in which mismatching rivals is costly, replicant managers are a minority in most stocks. Between 2016 and 2021, a portfolio long accommodated stocks and short replicated stocks earns 7.6\% per year at 15.47\% volatility, with modest market exposure and losses concentrated in the COVID crash. Imitation among intermediaries is a priced risk that originates in quantities.

This paper studies how court congestion affects bankruptcy resolution. We compile case-level data from the State of São Paulo in Brazil, where court workload varies significantly and cases are randomly assigned to courts. We find that firms assigned to more congested courts are less likely to be liquidated and more likely to remain active after filing. We also show that these courts operate more slowly, leading to longer total time in court, particularly in liquidation procedures. To explore the underlying mechanisms, we first document that court congestion increases the duration of early procedural stages in reorganization, such as plan submission and voting. These delays can affect the bargaining environment between debtors and creditors. Consistent with this idea, we then show that more congested courts are associated with deeper debt renegotiation in reorganization plans. We interpret these findings as evidence that court congestion alters bargaining conditions during bankruptcy by reducing expected liquidation values, weakening creditors’ outside option, and shifting bargaining power toward debtors

We propose a dynamic urban economy model in which risk is priced, and use it to organize spatial heterogeneity in residential real estate returns. We take the model to NCREIF total-return data on institutional multifamily properties, which report the income and capital components of return directly at the property level. Risk exposure to the national apartment index is priced across U.S. metropolitan areas, and within them excess returns rise with distance from the primary employment center, at a baseline 3.26 basis points per mile per quarter; the within-city gradient survives neighborhood demographic characteristics. We interpret the gradient through a combination of priced risk exposure and an operating leverage channel: operating costs excluding property taxes decline less than one-for-one with rents across locations, at an elasticity of 0.80 (s.e. 0.04) in expense accounts, so locations with lower rents have thinner operating margins that amplify exposure to aggregate shocks and command higher expected returns. A fixed-cost benchmark rationalizes the profile of income returns across distance quartiles and recovers a negative risk-adjusted growth rate for local cash flows, the condition under which the model predicts cap rates that rise with distance. A time-varying extension, in which operating leverage interacts with the consumption-wealth ratio to predict excess returns, is consistent with the data.

We study open banking as a macro-financial policy that reallocates access to privately produced, nonrival data. In an overlapping-generations model, sharing creates cross-source synergies but weakens incentives to produce proprietary information; credit, concentration, and welfare therefore depend on openness and network design. Conditional on a 5 percent external-data share, calibration to Brazilian bank and Open Finance data implies an effective loading of 0.125. The baseline raises credit-to-output by 3.5 percentage points. Across Brazil-specific external anchors, loan rates, spreads, and risk fall, while transition welfare remains negative despite modest steady-state gains.

I use the VIX futures term-structure to build novel currency risk factors and show that they are priced in the cross-section of currency returns and carry incremental predictive information for future returns. Currencies with greater exposure to the level and curvature factors earn a lower risk premium, acting as insurance, while those more exposed to the slope factor earn a higher premium. A portfolio that is long high-exposure currencies and short low-exposure currencies captures this risk-return dispersion, and its excess return can be understood as compensation for bearing a globally priced volatility shock. Decomposing the factor returns into news about expected future volatility and news about the variance-risk premium, I find that the risk premia are driven almost entirely by variance-risk-premium news, while expected-volatility news carries no price of risk. A Bayesian factor-selection procedure confirms that the term-structure factors enter the stochastic discount factor with high posterior probability. I rationalize these results with a two-agent preferred-habitat model in which experts have horizon-specific volatility hedging demands and arbitrageurs face funding and margin costs.

We study how belief shocks affect asset prices when uncertainty alters investors’ information processing. We introduce a Subjective Inattention framework in which limited attention and subjective beliefs jointly shape return expectations. Using Brazilian financial data and heteroskedasticity-identified SVARs, we show that belief shocks generate muted average price responses but strong and persistent effects during crisis periods. Rational and single-bias benchmarks do not reproduce this nonlinear transmission. U.S. evidence supports the external validity of the mechanism and shows that market composition determines which information-processing-capacity segment drives amplification. Subjective inattention therefore helps explain crisis-period price sensitivity and tail risks.

Exploiting the exogenous implementation of a nationwide government-backed debt renegotiation program in Brazil, we show that beneficiaries experienced a short-term reduction of approximately 15 to 18 log points in default rates. However, these effects proved temporary, as default levels among treated individuals gradually returned to their pre-program baseline within roughly 18 months of the program’s launch. Moreover, the benefits of the program were unevenly distributed across the population. The reduction in default rates was concentrated among individuals with formal employment, and thus with stable income streams, whereas the most vulnerable segment of the population (those engaged in informal employment) experienced no statistically significant improvement. We further show that the smallest banks, which also exhibited the highest levels of loan loss provisions, benefited disproportionately from the decline in default rates. Overall, our findings suggest that government-backed debt renegotiation programs primarily benefit individuals with more stable cash flows and smaller, riskier financial institutions. We also discuss potential unintended consequences of such policies, including moral hazard incentives for smaller banks to expand riskier lending and for households to accumulate additional debt in anticipation of future renegotiation programs.

Recursive nonlinear impulse responses require an estimated innovation law wheneverthe impact shock is normalized by innovation ranks and future innovations are integratedout. The closest semiparametric recursive construction in the literature estimates therelevant innovation quantile functions smoothly and discusses a direct empirical-residualimplementation without developing its complete first-order inference theory. We tackle thisgap in a finite-dimensional nonlinear structural autoregression with unrestricted continuousmarginal innovation distributions and a fixed normal-rank shock. Our estimator replaceseach innovation quantile function with the empirical quantile of generated structuralresiduals and iterates the same structural transition. For any fixed collection of responses,we establish a joint ?T asymptotic linear representation with four components: directtransition estimation, the effect of transition estimation on residual order statistics, ordinaryinnovation-quantile estimation, and the shifted impact quantile. After projection throughthe recursion, the quantile terms admit a residual-rank-and-spacing representation, yieldingfeasible inference without innovation-density estimation or quantile smoothing. We thencharacterize the propagated bias from smoothing, establish validity of a full recursiveresidual bootstrap, and derive the additional covariance contribution from a finite numberof simulated paths, providing bandwidth-free inference for the empirical-residual version ofthe same normal-rank response used in the smooth recursive construction.

Pooled calibration of conformal prediction intervals, as in the EnbPI algorithm of Xu and Xie (2023), carries a persistent bias under seasonal heteroskedasticity: the per-season coverage gap contains a term that does not vanish with sample size, so volatile seasons are systematically undercovered. Two remedies remove it and form the endpoints of a one-parameter family: stratifying the calibration buffer by season, and standardising residuals by a season-specific scale before pooling. The second is more efficient but valid only when seasons differ in scale alone; when they differ in shape as well, it leaves the coverage spread where pooling left it, while stratification cuts it by two thirds.Stratification is usually thought too costly at the per-season sample sizes a monthly series provides. We show most of that cost belongs to the interval construction, not the conditioning: the empirical quantile of a short buffer is biased inwards, and the minimum-width line search among many candidate quantile pairs alone costs four to six points of coverage. Building the interval from conformal order statistics with a fixed asymmetry repairs both defects, and stratified coverage at twenty residuals per season rises from 0.79 to its nominal 0.90 across four seasonal designs, the residual cost being width rather than coverage. On Brazilian food-price inflation the same change takes coverage from 79.2% to exactly 90.0% with no dispersion across the twelve months; on export growth it raises the worst month from 70% to 90%. A permutation test rejects the pure-scale restriction on both series.

This paper provides a semiparametric foundation for local-projections difference-in-differences with covariate adjustment. We treat every horizon-specific clean-control stack as an average-treatment-effect-on-the-treated problem and develop regression-adjustment, inverse-probability-weighting, inverse-probability-tilting, and doubly robust estimators within a common architecture. The resulting DRLPDID framework covers both staggered absorbing adoption and non-absorbing binary treatments. For absorbing adoption, we establish that the horizon-specific local estimand equals a treated-entry-weighted aggregation of cohort–time average treatment effects. This result makes the target population and its horizon-dependent composition explicit. We also characterize aggregation over prespecified post-treatment horizons. For switching treatments, we formulate sustained switch-in estimands indexed by realized treatment histories. Identification requires clean pre-event histories, no anticipation, restrictions on carryover, and stabilization of priortreatment effects. The framework then permits comparisons with stable-zero units and, under stabilization, stable-one units. We derive cluster-level influence functions for the full dynamic path and for prespecified linear summaries, yielding cluster-robust standard errors and simultaneous multiplier-bootstrap confidence bands. Simulations reproduce the predicted double-robustness pattern and document the behavior of the proposed estimators under nuisance-model misspecification. Empirical applications revisit no-fault-divorce reforms under absorbing adoption and democracy and economic growth under treatment switching. The proposed estimators are implemented in an open-source Python Package.

This paper proposes a Markov-switching vector autoregressive (MS-VAR) framework for analysing time-varying Granger causality in the presence of conditional heteroskedasticity. Existing approaches in the tradition of Psaradakis (2005) embed both mean dynamics and covariance structure within the same regime indicators, creating a filter misattribution problem; volatility clusters and genuine causality regime changes are observationally similar from the perspective of the Hamilton filter, so the former can be spuriously attributed to the latter. The paper resolves this by separating the two channels, allowing regime switching only in the conditional mean while modelling second-moment dynamics through a state-invariant GARCH-DCC structure. Monte Carlo evidence with T=300 confirms that the fixed-covariance Markov-switching estimator produces substantially worse regime classification across all four predictive states, with mean classification error roughly twice aslarge as the proposed estimator; differences are highly significant in paired comparisons conducted on the common-convergence sample. An empirical application to the trade-weighted U.S. dollar index and the Trade Policy Uncertainty (TPU) index of Baker(2016) illustrates the practical importance of the approach: ARCH-LM tests strongly reject homoskedasticity in theVAR residuals, and the MS-VAR with GARCH--DCC uncovers four distinct predictive regimes whose timing maps cleanly onto the major episodes of U.S. trade policy history, from the Plaza Accord era and the China shock of the early 2000s to the two Trump trade-war episodes of 2017-2020 and 2025-2026, with TPU-driven dollar dynamics dominating in both trade-war periods and a persistent No Causality regime prevailing during intervals of broadly stable trade rules.

This paper proposes model-averaging methods for estimation and inference of quantile treatment effects (QTE and QTT) under treatment selection based on observables. To address propensity score misspecification, we develop an estimator averaging QTE/QTT across models. To assign optimal model weights, we employ a model-averaging estimator that minimizes the asymptotic mean squared error. We establish the asymptotic distributions of the estimators, enabling the construction of valid confidence intervals, under two different scenarios. First, when the covariate set for each sub-model satisfies the unconfoundedness assumption. Second, when some (but not all) covariate sets violate the unconfoundedness assumption. Moreover, we construct a test for unconfoundedness for a given covariate set, assuming that the union of all covariates satisfies the assumption. We also introduce the “unconfoundedness signature plot” as a data-driven tool for assessing covariate relevance. Simulations show strong finite-sample performance of the proposed methods. We also illustrate the approach by estimating the effect of inherited control on firm performance.

Empirical research in economics relies on assumptions such as exogeneity and correct model specification to justify causal interpretation and identification. In practice, these assumptions are typically assessed through restrictions on conditional means, due to their analytical and computational convenience. However, such approaches are inherently limited: violations may arise through higher-order moments of the distribution without affecting conditional means, rendering mean-based tests asymptotically powerless. This paper develops a nonparametric framework for testing equality of conditional distributions in selection models. The proposed hypothesis is fully distributional and encompasses implications of exogeneity and correct specification as special cases. The method is based on the Maximum Mean Discrepancy (MMD). We show that existing MMD baed tests are invalid under irregular selection mechanisms, leading to size distortions. To address this, we introduce a regime-specific smoothness condition that ensures valid asymptotic inference. For global inference, we combine local test statistics using a Poisson-Binomial distribution, which accommodates heterogeneous behavior across the covariate space. The proposed test is consistent, and achieves the optimal local detection rate. Simulation results confirm good size and power properties.

Systemic discrimination arises when apparently neutral actions produce unequal treatment of equally qualified individuals. We formalize one channel: the discrimination that emerges when equally qualified individuals with different protected characteristics (e.g., gender or race) occupy, on average, different positions within social networks. Building on the peer-effects literature, we show that identifying this network-mediated discrimination reduces to identifying exogenous and endogenous peer effects, which runs into two typical challenges: the simultaneity of outcome formation (i.e., friends affect each other mutually) and the self-selection of friendships. We address the first, under exogenous network formation, with the approach of Bramoull´e et al. (2009) and the second with that of Jochmans (2023). We extend the latter on two fronts: controlling for estimated sociability fixed effects and allowing for heterogeneous peer effects, and derive the corresponding 2SLS estimators. We also make explicit a limitation shared by this literature and not resolved here: identification restricts homophily on unobserved productivity. Finally, two empirical applications estimate gender discrimination in school networks, using data from schools in Recife (Brazil) and from the Add Health program, decomposing total discrimination into a direct component and a systemic component; the latter split into a network-mediated channel and the remaining (attribute-mediated) systemic discrimination. In the Recife data the network channel is large enough to offset direct discrimination, suggesting that network structure can be a quantitatively relevant vector of gender discrimination.

We study sample maxima of Markov processes through random-measure lifts and product disintegrations. Conditional on the latent path, the observations become independent with edge-local conditional laws, yielding an exact quenched factorization for the distribution of the maximum. Under stationarity, ergodicity, and logarithmic integrability, we obtain the almost-sure exponential rate at fixed thresholds and compare it with the corresponding annealed probability. A stationary AR(1) example makes the construction explicit, where the lift preserves the law of the original dependent process, while conditioning on the latent trajectory expresses the no-exceedance probability as a product of univariate Gaussian distribution functions. For moving thresholds, we separate the conditional log-probability into an averaged term and a centered fluctuation term, providing sufficient conditions for quenched and annealed convergence.

This paper develops quasi-maximum likelihood (QML) estimation and inference for a threshold Spatial Durbin panel model with unit fixed effects, in which the spatial-lag coefficient, the covariate slopes, and their spatial-lag coefficients all switch jointly at an unknown threshold. The threshold also modifies the spatial multiplier through which shocks propagate, so the model permits regime-contingent direct, indirect, and total effects. Estimation proceeds by profile QML after removing the unit effects via the within transformation, with analytical corrections for the finite-sample bias in the regular parameter estimates and for the degrees-of-freedom distortion in the estimated disturbance variance. Under a large N, large T framework, we establish consistency and asymptotic normality for the regular parameters, derive the nonstandard convergence rate of the threshold estimator, construct likelihoodratio confidence sets for it, and propose a bootstrap supremum test for parameter homogeneity. Monte Carlo experiments show that the bias corrections materially improve variance estimation and inferential accuracy, and that the corrected QML estimator performs favorably relative to a homogeneous Spatial Durbin model, a non-spatial panel threshold model, and a correctly specified spatial two-stage least squares estimator, with the largest gains for the regime-specific spatial coefficients and the threshold location. An application to sixty economies observed over four decades identifies an initial-income threshold near eight thousand five hundred dollars per capita, above which estimated growth transmission through the spatial network is substantially stronger than below it, a form of heterogeneity that neither a homogeneous spatial model nor a non-spatial threshold specification can represent.

Synthetic controls make donor weights transparent, but not the pre-treatment periods that determine them. We develop residual and influence diagnostics by viewing the estimator as a simplex-constrained least-squares problem and exploiting the geometry of its active donor set. Conditional on active-set stability, we derive closed-form measures of period leverage, deletion and local influence, and active-set residual standardization, with robust versions for serial correlation and heteroskedasticity. The diagnostics distinguish diffuse from concentrated pre-treatment fit, identify periods that drive the post-treatment gap, and reveal when a low MSPE masks fragility. They are designed as pre-treatment audit tools, not specification-selection rules, and are paired with a reporting protocol to limit post-treatment cherry-picking. Simulations and an application show how they complement standard synthetic-control evidence.

Every random utility model rests on an assumed error distribution, usually chosen for tractability. We characterize when this choice matters. In a local misspecification framework in which every object is parametrized by the model's explanatory power (R²), we classify all infinitesimal perturbations of the error distribution by a single quantity: the tail-dominance index, the decay rate of the perturbation's tail mass relative to the assumed kernel's own losing-probability decay. The classification is a trichotomy with an explicit critical rate. Subcritical misspecifications, typically reshaping the center, the modality, or the skewness of the distribution, are neutralized in predicted choice probabilities as R² grows; supercritical ones, which add polynomial, subexponential, or one-sided heavy tails, are amplified without bound. Because the critical rate is a property of the assumed kernel, the probit is strictly more tail-fragile than the logit. Coefficient structure, in contrast, is recovered almost independently of the error distribution. Monte Carlo evidence, including families designed adversarially against the classification, confirms the theory's predictions and extends them to large deviations. High explanatory power is a license to ignore the shape of the noise, but not its tails.

In this paper I develop a breakdown frontier approach to assess the sensitivity of Local Average Treatment Effects (LATE) estimates to violations of monotonicity and independence of the instrument. I parametrize violations of independence using the concept of $c$-dependence from Masten & Poirier (2018) and allow for the share of defiers to be greater than zero but smaller than the share of compliers. I derive identified sets for the LATE and the Average Treatment Effect (ATE) in which the bounds are functions of these two sensitivity parameters. Using these bounds, I derive the breakdown frontier for the LATE, which is the weakest set of assumptions such that a conclusion regarding the LATE holds. I derive consistent sample analogue estimators for the breakdown frontiers and provide a valid bootstrap procedure for inference. Monte Carlo simulations show the desirable finite-sample properties of the estimators and an empirical application shows that the conclusions regarding the effect of family size on female labor force participation from Angrist & Evans (1998) are highly sensitive to violations of independence and monotonicity.

As predictive algorithms grow in popularity, using the same dataset to both train and test a new model has become routine across research, policy, and industry. Sample-splitting attains valid inference on model properties by using separate subsamples to estimate the model and to evaluate it. However, this approach has two drawbacks, since each task uses only part of the data, and different splits can lead to widely different estimates. Averaging across multiple splits, I develop an inference approach that uses more data for training, uses the entire sample for testing, and improves reproducibility. I address the statistical dependence from reusing observations across splits by proving a new central limit theorem for a large class of split-sample estimators under arguably mild and general conditions. Importantly, I make no restrictions on model complexity or convergence rates. I show that confidence intervals based on the normal approximation are valid for many applications, but may undercover in important cases of interest, such as comparing the performance between two models. I develop a new inference approach for such cases, explicitly accounting for the dependence across splits. Moreover, I provide a measure of reproducibility for p-values obtained from split-sample estimators. Finally, I apply my results to two important problems in development and public economics: predicting poverty and learning heterogeneous treatment effects in randomized experiments. I show that my inference approach with repeated cross-fitting achieves better power than existing alternatives, often enough to reveal statistical significance that would otherwise be missed.

This paper studies the consequences of heterogeneous treatment effects in panel-data settings where potential outcomes follow a linear factor structure. We show that, when the treatment-effect heterogeneity admits a linear factor structure, the interactive fixed effects (IFE) estimator fails to recover an interpretable average of treatment effects. This failure occurs because IFE absorbs the treatment effect heterogeneity, creating a bad-control problem. With time-invariant factors or unit-invariant loadings, identification further fails due to multicollinearity. In contrast, alternative estimation methods~--- such as synthetic-control and imputation-based approaches~--- that exclude treated units in post-treatment periods from the factor estimation, are robust to this problem.

In dynamic principal-agent games, relaxing the full-commitment assumption has strong practical implications: we can no longer use the Revelation Principle (RP), the main tool that makes the study of these problems tractable. Several economic applications depart from full-commitment, by assuming the principal has no intertemporal commitment power. This paper provides a toolkit for the study of principal-agent games when the principal has this form of limited commitment and there are finite types. We propose the quasi-direct mechanism as a canonical mechanism that generates any equilibrium payoff of any principal-agent game. A quasi-direct mechanism is one in which the message space is the space of types plus the silent message. We then provide a quasi RP for limited commitment: (i) any Perfect-Bayesian Equilibrium payoffs of any principal-agent game can be generated by quasi-direct mechanisms; and (ii) it is without loss to focus on equilibria in which incentive-compatibility holds.

This paper studies how communication across experts prior to aggregation by a decision-maker affects the efficiency of forecast combination. When experts exchange information before reporting their forecasts, their signals become correlated through the communication network, altering aggregation efficiency even when forecasts are unbiased. The analysis introduces a statistic that characterizes how network structure shapes aggregation efficiency and shows that degree heterogeneity plays a central role. Among connected networks, regular networks attain the minimal level of aggregation distortion, while star networks generate the largest distortions within sparse connected structures. Random network benchmarks show that aggregation efficiency approaches the regular-network benchmark when expected degree either vanishes or becomes large as network size increases, whereas networks with constant expected degree generate intermediate distortions. These results provide a theoretical foundation for understanding how communication across experts affects forecast combination and establish a connection between the forecast combination literature and models of social learning in networks.

Arbitrators for high-stakes conflicts, as well as judges and other officials, are often appointed through structured bargaining protocols. The theoretical literature models these protocols as extensive-form games with perfect information, evaluating them based on the merits of their subgame-perfect equilibria, such as efficiency. However, decision makers often fail to implement backward induction and exhibit other-regarding preferences. In a large experiment, we compare two prominent protocols and show that those concerns affect outcomes. Bargaining protocols whose equilibria are unfair (in a maximin sense) fare poorly compared to those favoring compromises. However, lengthy protocols face limitations because they elicit non-equilibrium behavior.

This paper develops a hybrid model that extends the Ramsey-Friedman taxation framework by incorporating heterogeneous agents, competitive markets, and a policy maker responsible for fiscal policy and public debt management. We establish an equilibrium existence theorem and show that quasi-linear preferences guarantee time-consistent policy. Within this framework, a virtual monetary unit emerges endogenously as a transaction-clearing service. We further show that, under suitable conditions on agents' preferences, the optimal taxes on capital and non-durable consumption goods are zero, thereby extending the classical optimal taxation results to this environment. The model provides a characterization of money in a frictionless general equilibrium economy formulated as a Stackelberg game and proposes a digital monetary system that preserves purchasing power across regions and productive sectors in a stable, non-inflationary manner, without distorting resource allocation.

Pollsters are often cited as main players during campaigns and as major trendsetters for electoral outcomes, but little study has been made to understand how they affected by market-forces. In this paper, we provide a partial equilibrium model to study how individual pollsters make rational decisions to maximize profit in response to market structures and how those decisions affect overall poll precision. We show that market equilibrium is not always efficient, especially in the presence of asymmetric cost parameters or correlation among polls. We analyze a policy intervention in which a minimum level of precision is required for polls to hit the market, and conclude that it can have significant effects on reallocating welfare, increasing efficiency, and changing market concentration. We then introduce the behavioral element of correlation neglect and show that there are parameters under which it induces pollsters to make more precise polls and ultimately increase welfare.

We study adverse selection in repeated trade with divisible contracts and interdependent values. A privately informed individual sequentially meets short-lived dealers, posts a quantity, and dealers compete on prices after observing the full trading history.We propose a simple belief formation rule and show that, under separation, it is equivalent to D1 in this dynamic setting. The equilibrium is unique, separating and history-dependent. Early on, dynamic incentives amplify the static distortion: anticipating that worse types might mimic them to secure better prices in the future, almost all types scale back quantities relative to the static benchmark. As public history accumulates, the marginal value of additional signaling falls, and allocations move monotonically toward the first-best. As the game horizon lengthens, terminal allocations converge to efficiency. Our work clarifies how quantities and timing interact as signals, showing that repeated trade endogenously weakens the signaling role of retention over time. We illustrate the results in three applications: insurance, financing and the job market.

I introduce stochastic non-tâtonnement processes and argue that they are a natural outcome of general equilibrium theory. To this end, I revisit classical demand theory to define a normalized Walrasian demand and a diffeomorphism that flattens indifference curves. These diffeomorphisms are applied to the three canonical manifolds in the consumption domain (i.e., the indifference and the offer hypersurfaces, and the trade hyperplane) to analyze their images in the normalized and flat domains, and relations to the set of Pareto optimal allocations in Arrow-Debreu economies are discussed. Then, I derive, for arbitrary non-tâtonnement processes, an Attraction Principle based on the dynamic of marginal substitution rates seen in the “floor” of the flat domain. This motivates the definition of stochastic non-tâtonnement processes and, specifically, Edgeworth’s ones. When utility functions are attractive and sharp, these Edgeworth’s processes are particularly well behaved and lead to the calculation of stochastic trade outcomes over the contract curve, which can be used to model price stickiness and sustained economic disequilibrium, and to prove a stochastic version of the First Welfare Theorem.

This paper investigates the premature termination of continuous-time dynamic contracts with CARA utility. In the dynamic contracting literature, particularly in Sannikov (2008), contract termination arises endogenously as a retirement decision driven by the wealth effect: as the continuation value increases, the cost of providing incentives becomes high enough to render continuation inefficient. Unlike this mechanism, we show that contractual termination may occur even in the absence of a wealth effect. The model introduces a lower barrier for the continuation value promised to the Agent, interpreted as a minimum viability threshold for the contractual relationship. Since this promise evolves stochastically, adverse shocks may lead the process to reach this boundary, generating a \textit{first-passage} event and the premature termination of the contract. This mechanism breaks the stationarity of contracts under CARA utility highlighted by Holmstrom and Milgrom (1987), introducing state-dependent incentives: effort increases when the promise of utility approaches the liquidation barrier and decreases when the contract is far from the stopping region. Near the boundary, the risk of termination increases and the provision of incentives becomes more costly for the Principal.

Buy now, pay later (BNPL) is a business strategy where a merchant joins with a provider to offer point-of-sale financing, typically at no interest. We present a formal microeconomic model that explains this practice. BNPL acts as an interest rate subsidy that separates consumers in two groups, allowing the merchant to effectively practice third-degree price discrimination. Consumers who need credit spend less money to buy due to subsidized interest, while consumers who don't are charged a higher price. We show the (optimal) contract between the merchant and the BNPL provider is a three-part tariff consisting of a fixed fee to split profits, an ad valorem commission to share risk, and an interest rate. If consumers are rational, BNPL benefits consumers who need credit and harms those who don't, with ambiguous overall welfare impact. However, if consumers are overoptimistic about the probability of repaying their debts, BNPL may harm both groups of consumers.

Which firms within an industry capture the regulator? We develop a three-stage model of regulatory capture in oligopoly---Cournot competition with heterogeneous marginal costs, common agency with truthful contribution schedules, and a fixed cost of political participation---in which regulation operates through a political technology: an incidence profile that determines how much each firm benefits from the regulatory instrument. The direction of capture is not intrinsic to the industry; it is selected by an observable feature of policy design. When the instrument is specific (per-unit, uniform incidence), market power and political power align: because the equilibrium price--cost margin equals the firm's own output, the efficient, high-markup firms have the most profit exposed to the regulated price, gain the most from participation, and lead the capture. A unique participation threshold partitions the industry into politically active market leaders and inactive marginal firms. When the instrument is ad valorem, incidence rises with marginal cost and the direction of capture inverts: the least efficient firms lead a defensive lobby for the protection on which their survival depends. The classical comparison of specific and ad valorem instruments, developed for tax incidence under imperfect competition, thus determines which firms lead regulatory capture. Welfare implications differ sharply by regime: under alignment, political power compounds the allocative distortion of market power and competition policy addresses both at once; under the defensive regime, the two policy margins separate.