"Flooded House or Underwater Mortgage? The Macrofinancial Implications of Climate Change and Adaptation”

I study how climate change affects housing markets, mortgage credit, and private adaptation in a general-equilibrium framework with overlapping generations. Households are exposed to physical climate risks that damage housing and degrade land, which is inelastically supplied. While exposure to future climate risk lowers expected resale values, realized climate damages reduce habitable land, driving house prices up over time through scarcity. In frictionless markets, forward-looking prices support efficient private adaptation. However, constrained households underinvest in resilience, implying that pricing alone may be insufficient. Unequal adaptation amplifies wealth inequality and accelerates land degradation. Over generations, the adaptation gap widens endogenously as tightening credit constraints further limit investment in resilience. A counterfactual shift from constrained homeownership to landlord-based ownership shows that separating adaptation investment from borrowing constraints can restore efficiency.

Redistribution and unemployment insurance

This paper analyzes the interactions between redistribution and unemployment insurance policies and their implications for the optimal design of tax-benefit systems. In a setting where individuals with different earnings abilities are exposed to unemployment on the labor market, I derive a Pareto-efficiency condition linking taxes and transfers when employed to benefits when unemployed. This Pareto-efficiency condition extends the standard Baily–Chetty formula for optimal unemployment insurance and shows that redistribution through unemployment benefits is efficient: (i) replacement rates should be monotonically decreasing with earnings, from 1 at the bottom of the earnings distribution to almost 0 at the top, (ii) the more redistributive the tax-transfer is, the more redistributive unemployment benefits should be, and vice versa. Empirical applications to the US and France show that these interactions between redistribution and unemployment insurance have sizable quantitative implications for optimal policy.

 “Causal Inference for Asset Pricing” w/ Valentin Haddad, Zhiguo He, Péter Kondor & Erik Loualiche.

This paper provides a guide for using causal inference with asset prices and quantities. Our framework revolves around an elementary assumption about portfolio demand: homogeneous substitution conditional on observables. Under this assumption, standard cross-sectional instrumental variables or difference-in-differences regressions identify the relative demand elasticity between assets with the same observables, the difference between own-price and cross-price elasticity. However, we uncover a missing coefficient problem: cross-sectional estimators mechanically absorb substitution patterns across assets. Recovering substitution is essential to answer many natural counterfactual questions, and requires analyzing the response of portfolios to exogenous time-series variation. The same principles apply to the estimation of multipliers measuring the price impact of supply or demand shocks. Our assumption maps to familiar restrictions on covariance matrices in classical asset pricing models, encompasses demand models such as logit, and accommodates rich substitution patterns even outside of these models. We discuss how to design experiments satisfying this condition and offer diagnostics to validate it.

"Magical Implemenation"

A principal must decide which of two parties deserves a prize. Each party privately observes the state that determines which of them is deserving. The principal provides each party with a text describing the conditions for deserving the prize and asks them to report the state of nature. The parties are not strategic. Each party activates a cheating procedure that relates to the state and the text provided. The principal magically implements his goal if he can devise a pair of texts that allow him to recognize a cheater by applying what we call the “one-way cheating principle”.

“What’s in a u?”

We revisit the long-lasting debate about the meaning of the utility function used in the standard Expected Utility (EU) model. Despite the common view that EU forces risk aversion and diminishing marginal utility of wealth to be pegged to one another, here we show that this is not the case. Marginal utility for money is an input into risk attitude, but it is not its sole determinant. The attitude towards ‘pure risk’ is also a contributing factor, and it is independent from the former. We discuss several theoretical implications of this result, for the following topics: (i) non-neutral risk attitudes for profit maximizing firms; (ii) risk aversion over time lotteries in the presence of discounting, and convex time budget decisions; (iii) the equity premium puzzle. We also discuss matters of identification: (i) for firms; (ii) via proxies; (iii) via standard MLE methods under parametric restrictions; (iv) in intertemporal-choice problems; and (v) cross-context elicitation in multi-dimensional settings, and its relationship with the methods and results from the psychology literature.

Bank specialization with Production Networks

This paper studies the benefits and costs of lending specialization along supply chains, where banks serve as common lenders. Using firm-to-firm transaction and credit registry data from Belgium, we show that common lending is persistent and widespread. We develop and estimate a structural model of credit demand and supply in imperfectly competitive markets, where firms are connected through the production network. Our estimation results reveal that firms prefer to borrow from the same bank as their suppliers or customers, which gives common lenders market power and enables them to charge higher markups. At the same time, the network effects of common lending give banks an incentive to offer lower interest rates to maintain their role as common lenders. Exploiting the closure of a large manufacturing plant as an exogenous shock, we show that common lending also creates costs: the shock propagates through the production network, reducing credit demand along the supply chain and among common lenders, with banks more exposed to the affected network experiencing significantly larger declines in lending.

"Public Insurance Design and Coverage Gaps under Electoral Competition"

We study public insurance when its design results from political equilibrium, focusing on health insurance as our central motivation.

Electoral candidates compete over policies that allow for any coverage level of public insurance, moving beyond the often-debated binary dichotomy between universal coverage and pure private insurance. Voters differ in income and privately-observed health risk. In a one-dimensional policy version of the model where candidates propose proportional coverage of health costs financed by a linear income tax, the equilibrium level of coverage is determined by the median voter's preferred policy.

The political feasibility of universal coverage is characterized by the median value of an index representing, for each voter, the ratio of the voter’s expected health losses to her income. Universal coverage emerges as a political equilibrium whenever the median of voter’s index exceeds a population average version of the index. This index remains a sufficient statistic even when voters also differ in their risk preferences. 

Numerical simulations calibrated on workers in a large US firm predict strong support for near-universal coverage, with a median bliss point of 81% coverage in the baseline model.

We make two main theoretical innovations. First, we incorporate private insurance markets into our model in order to study the extent to which the availability of private health insurance weakens support for public coverage. We find that the equilibrium level of public coverage is similar to that without private insurance, declining modestly to a median of 78%.

Second, we generalize the policy space by introducing a voting model in which electoral candidates compete over mechanisms, which may be described as income-contingent menus of insurance contracts. We show that the equilibrium policy coincides with that of a utilitarian social planner weighing voters according to their sensitivity to policy. The analysis predicts that some income groups receive full coverage irrespective of their health risk, while others receive a partial coverage baseline, akin to a limited insurance mandate, together with an option to acquire more insurance. The analysis provides a political-economy rationale for public insurance programs like Medicaid.

“Disclosing Proxies”

This paper studies strategic disclosure when the interpretation of evidence depends on an uncertain informational model. An agent privately observes evidence about an underlying state and chooses whether to disclose it to an observer. The observer interprets both disclosure and non-disclosure through a model that specifies a prior and an evidence distribution, but the true model may itself be unknown. Model uncertainty introduces a new channel: disclosure conveys information not only about the state but also about which model is true. We show that this ``proxy'' effect can either increase or decrease equilibrium disclosure, depending on the correlation between the models' priors and their evidence structures, and that model uncertainty can raise welfare even for both possible models. We interpret these results as describing environments where a designer---such as a policymaker imposing color blindness or a digital platform encouraging data provision---chooses whether to reveal or conceal the underlying model. In both cases, concealing the model alters agents' incentives to disclose evidence, and thus endogenously reshapes the information available in equilibrium.

“AI Sycophancy and Decisions”

We examine whether sycophantic AI advice distorts decisions. Our experiment involves 1,500 participants in 30 decision environments that span core domains in economics and the social sciences. We find that interacting with an AI that is broadly representative of consumer-facing models depolarizes choices on average, moving participants away from their initial leanings. This result appears despite the LLM being measurably sycophantic: it disproportionately supports users’ initial leanings and uses agreeable, flattering language. Depolarization occurs across moral and non-moral, objective and subjective, strategic and non-strategic, and complex and simple tasks, and despite the vast majority of researchers in an expert survey expecting polarization. Next, we show that sycophancy is behaviorally relevant—a treatment increasing AI sycophancy reduces depolarization—but that it is outweighed by the apparent informativeness of LLM advice. Finally, we ask whether market forces are likely to push toward greater polarizing effects outside our experiment or in the future. On the supply side, we show that our baseline LLM’s level of sycophancy is typical of leading models and that these models are not becoming substantially more sycophantic over time. On the demand side, we show that participants do not prefer greater sycophancy, do not select into AI advice in tasks where it is more polarizing for them, and exhibit greater depolarizing effects when they are more frequent AI users outside of the experiment.

"Robust Pricing for Cloud Computing"

We study the robust sequential screening problem of a monopolist seller of multiple cloud computing services facing a buyer who has private information about his demand distribution for these services. At the time of contracting, the buyer knows the distribution of his demand for various services and the seller simply knows the mean of the buyer’s total demand. We show that a simple “committed spend mechanism” is robustly optimal: it provides the seller with the highest profit guarantee against all demand distributions that have the known total mean demand. This mechanism requires the buyer to commit to a minimum total usage and a corresponding base payment; the buyer can choose the individual quantities of each service and is free to consume additional units (over the committed total usage) at a fixed marginal price. This result provides theoretical support for prevalent cloud computing pricing practices while highlighting the robustness of simple pricing schemes in environments with complex uncertainty.

“Partial disability insurance”

To curb the rising costs of disability insurance (DI) programs, some countries have implemented partial DI programs, which require recipients to work part time to receive benefits. However, the welfare implications of such programs remain unclear. The analysis in this paper is twofold. First, we introduce a conceptual framework that identifies the central trade-offs between partial-DI programs, full-DI programs, and ordinary employment. Second, we study a reform in Denmark that expanded the Danish partial-DI program. We find that the reform substantially altered selection into the program by shifting individuals who would otherwise have entered the full-DI program into the partial-DI program. At the same time, we find no evidence of negative selection from ordinary employment. However, existing partial-DI participants work fewer hours when starting new spells, suggesting that imperfect screening entails efficiency costs.

"The Role of Moral Values" w/ Fiona Paine, David Thesmar

This paper tests how people’s moral values influence their views of debt contracts. We ask participants to make decisions about debt contracts in different hypothetical situations (vignettes). We separately measure their moral values using the Moral Foundations Questionnaire (Graham et al., 2009). We have three main sets of findings. First, differences in moral values strongly explain the cross-section of participants’ debt decisions.

Participants with more conservative values show more support for credit score-based loan pricing, stricter forms of collateral, and tougher bankruptcy resolution. Second, when we randomly change the economic costs and benefits of debt within our vignettes, we find that participants change their answers in the direction predicted by economic theory.

Third, participants’ mental models of the functioning of the credit market strongly correlate with their moral values. Participants with conservative values are more likely to believe that strict enforcement and risk-based loan pricing provide incentives and are economically efficient. More liberal participants believe that insurance against unlucky shocks are important. Consistent with moral values being distinct from Bayesian beliefs, financial literacy does not attenuate the effect of moral values on beliefs about what is economically efficient.

"Robust Delegation"

We study delegation rules by principals uninformed of the underlying state (external
uncertainty) and the preferences of better-informed agents (internal uncertainty).
Evaluating delegation sets with a max–min criterion, we show that in multidimensional environments optimal delegation sets are simple: for broad classes of preference uncertainty, optimal delegation sets are convex. Thus, interval delegation is always (robustly) optimal when the action space is unidimensional. Internal uncertainty can justify greater discretion, allowing actions that are never optimal for the principal in any state; and a version of the ally principle holds: alignment along enough dimensions implies unconstrained delegation along all dimensions.

"Rollover Lotteries as Markets" 

We develop a new model of rollover lottery ticket sales. Treating the monetary loss on tickets as an implicit price, lottery design yields a precise inverse supply curve. Growing jackpots shift the inverse supply down, and help identify the falling demand curve arising from thrill heterogeneity. We nonparametrically estimate the demand for Powerball.

This implicit market model allows risk aversion or risk-loving preferences, but we show that even slight deviations from risk neutrality deviate from the data tremendously. This is a high stakes empirical deduction (based on 160 million gamblers) that risk aversion vanishes at very large stakes.

This model offers a unified theory of all aspects of the rollover lottery. For instance, ticket sales grow in the jackpot in a log-log convex fashion, which we verify in the data. It also implies how sales rise with inflation surges, but less so at higher jackpots. We predict a neutrality of ticket price increases, and to inflation surges. It explains why jackpot odds of US state lotteries are linear in the population, and why many countries cap their jackpots.

“Pricing News and No News With Heterogeneous Beliefs”

We study a general-equilibrium economy where a continuum of agents trade stocks and derivatives under heterogeneous beliefs along two dimensions: news intensity and content. When intensity disagreement dominates, implied volatility appears persistent—quiet periods shift wealth toward calm-world believers, compressing risk-neutral tail probabilities and raising prices. When content disagreement dominates, volatility appears mean-reverting—news shift wealth between optimists and pessimists. The information structure of news process matters for the persistence of intensity disagreement: in a Poisson limit, intensity disagreement survives but is eliminated in a Brownian limit. The framework endogenizes implied volatility smirk and U-shaped cross-section of derivative positions across subjective return beliefs.

“The Metric Is the Message? How Inequality Metrics Shape Perceptions and Preferences for Redistribution”

Inequality can be measured in many ways. Economists traditionally rely on technical indices such as the Gini coefficient, valued for satisfying formal properties. More recently, scholars have emphasized measures like the income share of the Top 10%, which more directly evoke political conflict. We experimentally test whether the way inequality is communicated shapes public perceptions and support for redistribution. In an online experiment (N = 1,246) among laypeople, we show that “political” measures reduce acceptance of inequality relative to Gini coefficient, which itself has a desensitizing effect on fairness and emotional ratings. The behavioral effects of switching from the Gini to the Top 10% share is quantitatively similar to switching from the (moderately unequal) United States to (extremely unequal) South Africa. Academic experts and policy makers are less susceptible, but not immune to the metric effect, and predict a strong behavioral effects on the general public. Finally, we show that left-wing media are more likely to report Top-share metrics and less likely to report Gini coefficient than right-wing media, a pattern that amplifies political cleavages. Overall, our findings show that inequality statistics do not merely describe economic reality, but exert a strong influence on (normative) assessments of the economy.

The Ghost in the Machine: Generating Beliefs with Large Language Models

I introduce a methodology to generate economic expectations by applying large language models to historical news. Leveraging this methodology, I make three key contributions. (1) I show generated expectations closely match existing survey measures and capture many of the same deviations from full-information rational expectations. (2) I use my method to generate 120 years of economic expectations from which I construct a measure of economic sentiment capturing systematic errors in generated expectations. (3) I then employ this measure to investigate behavioral theories of bubbles. Using a sample of industry-level run-ups over the past 100 years, I find that an industry's exposure to economic sentiment is associated with a higher probability of a crash and lower future returns. Additionally, I find a higher degree of feedback between returns and sentiment during run-ups that crash, consistent with return extrapolation as a key mechanism behind bubbles.

“Learning to Harm: The Intergenerational Transmission of Violence Against Women”

This paper provides the first large-scale estimates on the intergenerational transmission of violence against women (VAW) and evidence on how this cycle can be broken. We show that sons of fathers suspected of VAW are more than twice as likely to become suspected perpetrators of VAW, and daughters exposed to violent fathers are almost twice as likely to partner with VAW men. We then examine whether removing fathers who commit VAW from the household weakens this transmission. To address identification concerns, we use three complementary identification strategies: a judge instrumental variables approach that exploits quasi-random variation in father removal, a family-fixed effects model, and a comparison across stepfathers and non-residing biological fathers. These three designs reveal that removing VAW fathers significantly decreases sons’ later perpetration of VAW, while removing non-violent fathers increases sons’ later perpetration of VAW. The protective effects are particularly pronounced when removal occurs at younger ages. While the gains from removing VAW fathers partly reflect reduced exposure, we show that father removal also leads to declines in mothers’ substance-related mental health problems, consistent with improved maternal capacity. Together, these results indicate that VAW is a learned behavior developed during childhood and that reducing exposure to VAW fathers can meaningfully weaken its intergenerational persistence.

"Him Too? Analyzing the Effects of Epstein Connections"

The Epstein files, released by the DOJ between September 2025 and January 2026, provide an unprecedented record of a stigmatized actor's ties to corporate leadership. We examine whether proximity to such an actor expands firms' access to elite networks, transmits tolerance for misconduct, or both. Searching over 1.3 million DOJ documents for 92,698 CEOs and directors of U.S. public firms and using LLM-based classification, we identify 825 directors with frequent documented contact with Epstein. 

We establish three results. First, S&P 500 firms linked to Epstein in post-release news coverage experienced cumulative abnormal returns of -3.7% over three trading days. Second, Epstein-mediated ties substantially densify the corporate board network. Third, firms with more Epstein-connected directors experience significantly more governance, employee, and total misconduct incidents, as well as more regulatory actions and litigation. 

Mechanism tests indicate that gains in a firm's own network centrality do not predict theseoutcomes, whereas indirect exposure through neighboring Epstein-connected boards consistently does, identifying norm contagion rather than expanded network position as the channel through which proximity to a bad actor reaches the boardroom. 

“When Skills Are Specific: Labor Market Entry and Occupational Mismatch for Apprenticeship Graduates”

Using administrative data on the largest group of labor market
entrants in Germany—graduates from an apprenticeship in a
specific occupation—this paper studies the effect of entry
conditions on occupational mismatch. I find that higher entry
unemployment leads to persistent earnings losses that are driven by
full-time wage effects. At the same time, occupational matching, as
measured by work in the training occupation or the task similarity
between training and occupation, persistently falls. Occupational
mismatch accounts for around 20% of the estimated wage effects. A
conceptual framework sheds light on underlying mechanisms and
shows that lower levels of matching can arise from adverse general
or occupation-specific shocks. The framework entails empirical
predictions that I test using novel data on occupation-specific
unemployment. The findings highlight that the high-quality
provision of specific skills can protect apprenticeship graduates from
unemployment due to adverse entry conditions, but may expose
them to occupational mismatch that can entail long-term
productivity effects.

More Information, Less Learning from Prices

We revisit the question of whether investors learn from prices by directly conditioning on the arrival of information, including private information. A central challenge in this literature is that information arrivals are typically unobserved, making it difficult to separate learning from prices from reactions to information. We address this challenge using securities lending data to construct a real-time measure that captures informed trading, allowing us to identify the timing of information arrivals at the stock-day level. We find that following information arrivals, price informativeness declines, return volatility increases, and the correlation between trading volume and volatility strengthens. These patterns are inconsistent with noisy rational expectations models and instead support differences-of-opinion models in which investors overweight their own signals and underweight the information in prices. Our results persist outside periods of public news, highlighting the role of private information in financial markets.

“Hysteria: Medical Authority and the Confinement of Women in Nineteenth-Century England” 

Why did diagnoses of hysteria rise in nineteenth-century England, and why did they fall so heavily on women? I study how this diagnostic category spread through asylums and reshaped the recorded causes of women’s confinement. Hysteria is no longer a recognized disease, but nineteenth-century medical writing tied it to a broad set of nervous, emotional, reproductive, and behavioral explanations for women’s mental illness. Drawing on newly assembled data from asylum reports, medical records, and newspapers, I show that the growing use of the diagnosis coincided with a change in the recorded causes of women’s asylum admissions. The rise was steepest where local medical practice could most readily turn these ideas into certification and confinement. The findings suggest that the spread of hysteria reflected the diffusion of a gendered medical category through local certification and admission, as much as any underlying difference in health.

“Gender Norms, Stereotypical Beliefs, and Competitiveness”

Using an online experiment with 5,762 U.S. participants, we study beliefs and gender norms about a wide range of job-relevant behaviors and personality characteristics associated with seeking competition. Evaluators perceive competitive women as less pro-social, more career-oriented, less feminine, and more masculine than they are or state to be. Yet, there is no gender gap in belief accuracy. Although belief accuracy does not differ by gender, competitive women may still perceive social penalties—not for competing itself, which is socially accepted, but because behaviors associated with competition are perceived to violate gender norms for women. These penalties would largely disappear if beliefs were accurate.

“Tax Misperceptions and Labor Supply: A Two-Wave Survey Experiment in the Netherlands”

In countries with a complex tax system, it can be difficult for people to correctly assess their effective marginal tax rate. Misperceptions of the marginal tax rates can, in turn, lead to inefficient labor supply decisions. We conduct a two-wave survey experiment among a representative sample of the Dutch population (N ~ 2,500) to assess what misperceptions people have and whether information provision can influence their misperception and, in turn, their labor supply. In one information treatment, we provide information on the modal effective marginal tax rates in three different income brackets. In another treatment, we encourage respondents to visit a website that allows them to calculate their effective marginal tax rate. In the third treatment, we combine the two. Our data show that people with relatively low gross income tend to overestimate their effective marginal tax rate, while higher incomes tend to underestimate it. We find that providing information increases the accuracy of respondents’ estimate of their effective marginal tax rate two months later. There are also important consequences for labor supply. Among those who initially overestimate their effective marginal tax rate, work hours increase by about 1,5 hours per week. The implied compensated wage elasticity of labor supply is about +0.4.

“Dual Career Ladders: Individual Contributors in Modern Corporate Hierarchies”

Firms often maintain dual career ladders for managers and individual contributors (ICs). Using personnel records from 43 medium-sized and large firms, we reconstruct complete reporting chains and organizational layers to document how these tracks are structured and compensated. We show that ICs are prevalent throughout the hierarchy but earn substantially less than managers in comparable positions, with the penalty increasing toward the top. To interpret these patterns, we develop and estimate a model of dual careers in which firms allocate workers across management and IC tracks subject to coordination and problem-solving frictions. The estimates reveal distinct firm types with respect to their hierarchy shape, degree of managerial specialization, and workforce composition that generate varying IC penalties and respond differently to organizational shocks. Motivated by the large pay gains associated with promotions from IC to manager roles, we then analyze the demographic composition of the IC workforce. While pay within each track does not significantly vary across demographic groups, women and minority workers are disproportionately represented on IC tracks, especially at higher ranks. Thus, assignment to IC and managerial tracks emerges as a key mechanism through which within-firm career inequality contributes to broader gender and racial pay gaps.