NBER Working Papers Summary

In instrumental-variable (IV) studies, researchers often evaluate multiple candidate instruments and selectively report the specification with the most favorable first- or second-stage statistics. We show that this form of instrument selection (“instrument-hacking”) induces median bias in IV estimators toward the OLS estimand, undermining the rationale for using IV. When all candidate instruments have the same true strength, median bias increases monotonically with the number of available instruments. More generally, when candidate instruments differ in true strength, monotonicity need not hold because increasing the number of instruments can lead researchers to select genuinely stronger instruments. Nonetheless, in simulations calibrated to the empirical distribution of instrument strengths in the IV literature, median bias increases monotonically with the number of available instruments. Furthermore, it is substantial in magnitude, even when only a few instruments are available.

We design a method for measuring the risk preferences of agents in the deep past. The method combines a structural model of crop choice as a portfolio allocation with machine-learning prediction of expected crop returns, using historic agronomic and climate data. We estimate county-level risk preferences for the United States and farmer-level preferences in Kansas from 1889 to 1929. More risk averse farmers leveraged less, were less likely to purchase novel WWI Liberty Bonds, and were more likely to participate in local risk-sharing institutions. We show that higher risk aversion predicts slower tractor adoption and farm mechanization during the 1920s.

We examine whether occupational licensing improves service quality and safety using trip-level Uber data that include driver ratings and telematics-based measures of driving behavior. Exploiting quasi-random assignment from proximity-based dispatch, we compare trips served by licensed and unlicensed drivers in two settings: a cross-border comparison between New York City and New Jersey, and a deregulation event in Houston. Across settings and specifications, including instrumental variable estimates, we find no consistent evidence that licensing improves consumer outcomes. In Houston, post-deregulation entrants are indistinguishable from previously licensed drivers on ratings and driving behavior, despite differing markedly in experience and age.

I decompose stock returns into a duration-matched Treasury component, identified from monetary policy surprises, and a payoff component. Risk and returns rise much less with duration for stocks than for their matched Treasuries. Stock volatility is dampened by rate insurance: rates fall in bad times, so the bond inside a stock provides insurance against the stock’s payoff risk. Expected stock returns are dampened because the insurance works in reverse: rates rise in good times, so stocks’ payoff gains hedge the losses borne by investors holding net duration, notably government bonds when Ricardian equivalence fails. This framework helps reconcile positive bond premia with negative stock-bond covariance, sheds light on equity anomalies and the collapse of the value premium, implies that fiscal and monetary policy shape bond and equity premia, and motivates a two-factor model that jointly prices stocks and bonds. Rate insurance can even turn the price of long-run risk negative, explaining why long bonds beat long stocks.

Firms shape public policy not only from the outside through lobbying and campaign contributions, but also from the inside when business owners hold public office. We study this channel using a novel dataset that links state legislators’ personal financial disclosures to bill sponsorship records across 26 U.S. states from 2009 to 2023. The disclosures allow us to observe business ownership during legislative service and to distinguish entrepreneurs, defined as legislators who both own and actively manage a firm, from passive shareholders and employees. Applying a large language model to bill text, we classify legislation as pro-business and identify a subset of pro-entry bills that reduce barriers facing new firms. Entrepreneurs are a substantial presence in state legislatures, accounting for over 40 percent of legislators, and their representation varies primarily across states rather than within states over time. Although entrepreneurs do not sponsor more bills overall, they initiate a greater share of bills as first or sole primary sponsor. They also do not appear to be generic advocates for business. Relative to legislators with other business ties, entrepreneurs are no more likely to sponsor pro-business bills or bills endorsed by state Chambers of Commerce. Instead, they selectively advance pro-entry legislation, especially bills related to deregulation and innovation rather than antitrust or access to capital. These findings document an important channel through which entrepreneurs shape the policy environment for entrepreneurship from within political institutions.

How slow are bank transfers, and how do transfer delays affect deposit demand? Using transaction-level data from millions of depositors, we measure transfer delays by matching debits and credits across accounts held by the same depositor. Shorter delays correlate with more transfers and lower balances. Exploiting county-level exposure to Zelle’s staggered rollout, we find that faster payments reduce delays and deposit growth. Calibrating a deposit-management model, we find that transfer delays raise deposit demand, and the magnitude of this effect varies with interest rates and consumption volatility. Payment frictions therefore shape transactional deposit demand and monetary transmission.

We examine how contractions in local labor demand during the Great Recession affected children's academic achievement. We combine county-level test scores for grades 3–8 from the Stanford Education Data Archive with a shift-share design that interacts counties' 2005 industry composition with national industry employment growth, isolating demand-driven changes in local employment. Following recent advances in the shift-share literature, we validate the design with balance, pre-trend, and Rotemberg-weight diagnostics and report exposure-robust standard errors throughout. A one-standard-deviation adverse shock lowers mathematics achievement by about 0.03 student-level standard deviations and widens the White–Black and economic-disadvantage achievement gaps in both subjects. Within a common geography, losses concentrate among economically disadvantaged students. English language arts estimates point in the same direction but are harder to separate from the Great Recession's housing bust, and we interpret them as the combined effect of the labor-demand contraction and the associated decline in house prices. The achievement response is concentrated in the recession window. The post-2014 period, identified mainly by the oil-price cycle, yields a precise null, so our estimates measure the response to severe contractions rather than a general business-cycle parameter. School funding did not respond contemporaneously to these shocks, while family income, child poverty, and house prices all did.

How do firms set wages? How should governments set income taxes? If labor supply is inelastic to wages, firms can pay workers less than their marginal products, and governments can increase taxes without eroding the base. However, the structure of labor supply elasticities in the economy is complex. Recent empirics document variation across workers, firms, and margins (which firm to work at versus how many hours to work). To account for this rich structure of labor supply elasticities we extend the neoclassical model to include a discrete choice over which firm to work at, production complementarities and strategic interaction between heterogeneous, granular firms. In terms of wage setting, we find that novel effects of worker heterogeneity account for 78 percent of the variable component of labor supply elasticities and markdowns, and 89 percent of markdown differences between large and small firms. In terms of policy, higher progressivity makes labor supply less elastic, eroding the tax base by widening markdowns and worsening sorting. These channels (i) produce large declines in earnings following increases in marginal tax rates, consistent with empirical studies, and (ii) reduce optimal tax progressivity by one-third and associated welfare gains by two-thirds.

Public AI benchmarks steer research and allocate investments. They are therefore market designs. Public examples can reveal the process behind a private final test, while a finite public score cannot cover a broad task space inherent to general intelligence. I show how both gaps become profitable when scores move capital and how targeted effort erodes the signal used by later investors. The market design lesson is to separate development from certification: publish practice tasks, but choose the investment-consequential generator after the submitted system's evaluation policy is fixed.

The possibility of fiscal dominance in the representative-agent New Keynesian model (RANK) hinges on the assumption that income is perpetually demand-determined: fiscal deficits can drive output and inflation within that model only insofar as they trigger infinitely lasting, self-sustained shifts in aggregate spending and income. Moving to heterogeneous-agent New Keynesian models (HANK) opens the door to a different pathway: classical non-Ricardian effects, due to finite horizons or liquidity constraints. A refinement motivated by the model's intended focus on short-run phenomena—requiring a return to flexible-price outcomes in finite time—arrests the infinite feedback loop between spending and income, leaving only the classical non-Ricardian mechanism, and makes sure that the study of monetary-fiscal interactions is not centered on hard-to-test assumptions regarding beliefs at infinity.

Survival analysis methods are often used to predict the time until the onset of an event in settings when the true time-to-event (TTE) may be censored during training. Such approaches typically assume uncensored data are representative of censored data and that the probability of censoring conditioned on the covariates remains constant over time, i.e., there is no censoring distribution shift. However, both assumptions can fail in practice when censoring results from interventions targeted to individuals with particular comorbidities or genetic markers (such as prophylactic surgery when predicting time to cancer onset, or scheduled cesarean delivery and induction when predicting time to spontaneous labor) and changes in clinical policies alter which individuals are targeted for these interventions over time. To address this, we propose a new approach, cluster-weighted inference of time-to event (CWITE), that remains accurate when these assumptions do not hold. Unlike existing approaches that ignore times-to-censoring (TTC) or treat them only as a lower bound of the TTE, CWITE leverages the insight that a subset of censored individuals are likely censored close to their true TTEs, and uses a novel mechanism to learn from such individuals. On the task of predicting time to spontaneous labor using real-world data, CWITE improves TTE accuracy for individuals similar to censored training data (mean absolute error: 6.50 days, 95% CI: [5.55, 7.40] vs. 7.82 days, [6.82, 8.82]) while maintaining comparable performance for those similar to uncensored training data (6.50 days, [5.54, 7.61] vs. 6.63 days, [5.67,7.69]). Our results demonstrate that incorporating more specific supervision from censored training data can significantly improve TTE predictions in settings with limited overlap and censoring distribution shift, challenges common in real-world clinical data. Code to implement CWITE and reproduce all experiments in the paper is available at https://github.com/MLD3/CWITE.

Support for populist and authoritarian regimes is rising worldwide, despite evidence that they tend to underperform economically. We examine the role of (mis)perceptions of regime performance as drivers of political attitudes, leveraging two survey experiments with 11,377 respondents during Argentina’s 2023 presidential elections. Optimistic beliefs about the performance of populist and non-democratic regimes were widespread, and displayed a strong correlation with support for these regimes. When exposed to randomly assigned informational treatments challenging optimistic views about these political regimes, individuals significantly adjusted their beliefs, and reduced their support for candidates they associate with populist and authoritarian leanings. Exploring the impact of different information sources, we find that academic sources and newspapers were more influential than social media. Although individuals adjusted their beliefs and attitudes in response to information on regime performance, contradicting their prior beliefs reduced their demand for additional information, consistent with an important role for motivated reasoning.

The diffusion of technological innovation depends on incentives, regulations, and firms’ strategic behaviors. We study these intersections within cardiac procedure markets following Medicare’s expansion of non-hospital facility options for treatment, enabled by clinical advancements. State-level regulations restrain federal pro-competition policy. Where market entry occurs, business stealing is concentrated among the lowest cost treatment settings, rather than high-cost hospitals––increasing Medicare spending by approximately $2.5 million. Medicare policy also generates externalities for untargeted procedures and other payers, except when hospitals and physicians are vertically integrated. Federal rulemaking interacts with and is mitigated by complex market dynamics––including in potentially unanticipated ways.

We develop stress tests based on an original International Monetary System (IMS) model with regime switchings. The model is calibrated for nine reference currencies from the beginning of the Classical Gold Standard to the present. Regime switchings in currency dominance are related to combinations of conditions on a multidimensional environment variable that includes five classes of shocks: technology; development; monetary, financial and fiscal institutions; democracy and conflicts; and the regulatory environment. We provide an original database of events for these five classes of shocks, which is used for calibration. The calibration highlights the important role of the democracy and conflicts component in regime switchings. The calibrated model is then used to perform stress tests on the current prospects of currency dominance for a broad set of scenarios. A salient result from the scenarios we tested is that the dominance of the US dollar is at most marginally affected. No other currency emerges as a major player, suggesting strong inertia in the system’s current centripetal dynamics.

Buprenorphine prescribing for opioid use disorder remains far below the scale of the overdose crisis. We estimate the causal effect of county-level synthetic opioid overdose mortality shocks on the volume of buprenorphine treatment delivered, exploiting the staggered geographic diffusion of synthetic opioids from 2013 to 2023. Linking IQVIA Longitudinal Prescription Claims to National Vital Statistics System mortality records, we implement a difference-in-differences estimator with propensity score matching under three complementary exposure measures. Treatment volume increases significantly under all three definitions, with no evidence of differential pre-trends; point estimates for buprenorphine prescribing rates range from 5 to 17 percent of pre-treatment means, emerge within one to two quarters, and grow monotonically thereafter. Our design does not separately identify changes in patient demand and provider behavior, but the increase in unique prescribers is consistent with a provider-side response. The response is concentrated among advanced practice providers, Medicaid patients, and counties with pre-existing buprenorphine infrastructure; counties with little baseline prescribing show no response. Local mortality information acts as a determinant of treatment volume, with effects comparable in magnitude to major legislative interventions aimed at expanding treatment, but appear only where prescribing capacity already exists.

Childcare services can offer an opportunity for women to develop their careers and increase household income. At the same time, it can affect the development of children, depending on the quality of care. We first develop a collective model of the household as a conceptual framework to show how improving the availability of childcare can affect mothers’ labor market opportunities and earnings and how it can affect child development. We then review the literature on the effects of childcare on the employment and earnings of women and on child development. Childcare can increase mothers’ employment. However, limited work opportunities or strict norms against maternal work can keep mothers out of the labor force. When other family members provide care, childcare can free these caregivers to work or attend school. When studies measure the impact on child development, the results are mixed and depend on both context and quality of provision.

American colleges and universities are highly stratified by pre-college academic achievement, family background, and institutional resources. We study the meritocratic consensus in American higher education: colleges that high-testing students (who are generally also from high-income families) attend spend dramatically more on instruction than do those that enroll lower-testing students. Stratification by test scores has been largely stable since the 1960s, but the stratification of instructional resources has risen sharply since 1970 at both private and public institutions. Non-academic admissions criteria like athletics, legacy, and affirmative action are second-order in determining the allocation of students to universities. Potential economic justifications for the positive association of instructional expenditures with student prior achievement—q-complementarity between achievement and resources, convex social returns to high human capital, and incentives to invest in learning prior to college—have little empirical support. Resource stratification across universities has not increased in the past decade, largely due to increased public funding of universities that enroll lower-testing students through financial aid programs like California's CalGrant, but stratification within institutions is now rising swiftly.

Fertility has fallen in nearly all high-income countries. People also report wanting fewer children: between 2012 and 2022 the ideal number of children fell from 2.42 to 2.31 and the share of respondents intending a child within three years fell from 28 to 21 percent over a similar period. Why? Using two cross-national surveys that ask respondents about their attitudes and their fertility preferences, both fielded twice roughly a decade apart, we decompose each decline into the contribution of a broad and competing set of attitudes. Three stand out: children are increasingly seen as interfering with the freedom of parents; views on whether mothers of young children should work have become markedly more progressive and account for a substantially larger share of the decline among the tertiary-educated; and fewer people believe that women or men need children to lead a fulfilled life. This last attitude changed most over the decade and is the largest contributor to the fall in intended fertility, above all among less-educated women. The changing division of household work and increasing work-family conflict have minor contributions to changing fertility preferences.

We develop a general-equilibrium model of the global economy that integrates heterogeneous firms competing in product markets with countries that allocate capital around the world. Combining a hedonic demand system on the product side with a mean-variance portfolio system on the asset side, we obtain almost closed-form solutions for the equilibrium of the model. We use firm-level data on balance sheets, geographic breakdowns of revenue and employment, and business descriptions along with country-level data on bilateral equity holdings and trade costs to quantify the model to a cross section of roughly 23,000 listed firms in 48 countries. We use the model to evaluate the reallocation and welfare effects of globalization. Both financial and trade liberalization concentrate activity among the largest firms and raise welfare, with gains being larger in emerging and mid-sized open economies respectively. Product- and capital-market frictions amplify each other, meaning that liberalizing one market reduces the gains from liberalizing the other.

We introduce the notion of marginal cumulative multiplier — the effect of either government spending or taxation on output holding the other fiscal instrument constant — and apply it to a well known panel of consolidation episodes in 16 countries in the period 1978-2020. In our benchmark specification we estimate a marginal spending multiplier at two years of 1.5 and a marginal tax multiplier close to 0. We also estimate similar multipliers by applying the policy counterfactual method of McKay and Wolf (2023). In an extensive robustness analysis we never find spending multipliers below 1.3 or tax multipliers higher than -1. These findings are seemingly in contrast to those of much of the existing literature on fiscal multipliers, which typically finds higher tax than spending multipliers. We show that this contradiction disappears once the fiscal variables used in the literature are scaled by the proper factor.

The green energy transition will be powered by the mining and processing of lithium, nickel, and cobalt, which are critical for the production of advanced batteries. These minerals are concentrated geographically but traded globally, allowing key mining countries to exercise market power through policy intervention. Advanced batteries use multiple minerals in combination, and this joint use creates interdependence across mineral markets. We study the geopolitical implications of these forces and their consequences for green technology adoption worldwide. We quantify supply chain vulnerability, international policy spillovers, and the potential for mineral cartels.

The 2016 Panama Papers leak tightened regulatory enforcement around money laundering and offshore banking. We investigate whether the diversion of foreign aid in developing countries led to a shift to cryptocurrency as an alternative laundering platform. We develop a disbursement-timed forensic measure of cryptocurrency activity, combining on-chain Bitcoin transactions and wallet creation, off-chain exchange records, and IP-linked web traffic, and apply it to World Bank aid disbursements covering $238 billion across the 93 recipient countries in our estimation sample during 2018-2024. Exploiting the administrative timing of aid tranche arrivals, we find sharp, short-lived surges of crypto activity at the disbursement month, driven mainly by anonymous and newly created wallets on both tax-haven and mainstream exchanges. Blockchain forensics reveal patterns consistent with the placement, layering, and integration sequence of conventional money laundering. We estimate an implied leakage of 2 to 6 cents per aid dollar, which amounts to roughly 1.7 to 4.4 billion dollars of aid diversion across the tranche arrivals we study. Capture carries no funding penalty: the four sectors where we detect it, Transport, Water and Sanitation, Social Protection, and Governance, still absorb half of subsequent World Bank funding. Cryptocurrency facilitates aid diversion, but its transparent ledgers also leave forensic traces that may help detect and recover diverted funds.

What gave rise to the mosaic of cultural expressions across the globe? Why are some societies more culturally diverse than others? This study develops a unified theory of cultural evolution, characterizing the forces that have governed this process over the course of human existence. The research advances the hypothesis that population movements and socioeconomic transformations generated a mismatch between inherited cultural traits and prevailing ecological and societal conditions, setting in motion an evolutionary process governed by adaptation-based vertical transmission, horizontal diffusion, and fitness-based evolutionary selection. The theory generates novel testable predictions concerning the origins of the global cultural mosaic and the cross-societal variation in cultural diversity. First, greater ecological distance between ancestral environments generated greater cultural distance across societies. Second, prehistoric differences in ancestral diversity contributed to divergent adaptive capacity and greater cultural distance. Third, societies descended from more diverse prehistoric populations exhibit greater cultural diversity despite convergence induced by adaptation to a common environment and societal norms. Drawing on measures of cultural distance and diversity derived from folkloric and musical traditions among ethnic groups, visual representations of common concepts, and the dispersion of norms, values, and attitudes, the empirical analysis lends credence to the predictions of the theory.

In recent decades, economists studying international trade have gained access to an unprecedented volume and variety of data. These data have provided new insights and a more granular understanding of the mechanics of trade. Analyzing a large corpus of papers, we document that the nature of research has also shifted. Previously, the typical trade paper was either purely theoretical or an empirical paper testing theoretical predictions. More recently, scholarship has shifted toward a more integrated approach, particularly quantitative modeling. This development is surprising. We might have expected the data revolution (and contemporaneous credibility revolution) to increase the share of primarily empirical papers—although the markers of these revolutions are clearly evident in the rise of causal inference within empirical work. This article reviews these developments, assessing the strengths and limitations of different modes of inquiry, and plotting a path forward to harness the growing richness of data within the field.

Ordinal mechanisms use rankings but not preference intensities. Cardinal mechanisms also elicit school utilities, allowing them to use students’ preferences over assignment lotteries. We compare probabilistic serial with four cardinal mechanisms that maximize the same welfare objective under successively stronger restrictions: capacity-only, ε-envy-free, envy-free, and a welfare-maximizing cardinal-preference pseudomarket. Because the feasible sets are nested, maximum welfare weakly falls as the restrictions tighten from capacity alone to approximate no-envy, exact no-envy, and equal budgets with common prices. We establish positive and negative results on large-market truthtelling. All 44 theorem parts in the paper are machine-checked in the Lean 4 proof assistant. We develop methods for computing cardinal-preference pseudomarket equilibria for Seattle’s 898 students and 11 schools. Using set-identified cardinal preferences from Seattle high-school choice data, we estimate that, under exact capacity, capacity-only raises mean welfare over probabilistic serial by 0.053, equivalent to shifting 5.3 percentage points of assignment probability from the average student’s worst school to the top school. Exact envy-freeness retains 79% of this gain; best-found pseudomarket equilibria yield a mean gain of 0.002. The estimates show both the value of cardinal information and the welfare cost of pseudomarket fairness restrictions.

As of July 2026, 23 states had approved policies that disallow spending of SNAP benefits on sugary drinks. The effects of such restrictions on consumption are unclear, however, since most households can simply switch to buying sugary drinks with non-SNAP funds. Using a difference-in- differences design with nationwide grocery purchase panel data, we estimate that restrictions in the first 10 states reduced the average SNAP household’s retail purchases of excluded drinks by 12.4 percent (standard error = 1.0) over the first half of 2026. In states that excluded only some sugary drinks, SNAP households partially substituted to non-excluded drinks. Surveys we carried out before and after implementation show that the restrictions increased SNAP recipients’ perceptions of stigma. We combine our estimated consumption reductions with external parameters to model welfare effects given an over-consumption internality and a fiscal externality via public health care spending. In our model, excluding all sugary drinks from SNAP nationwide would provide benefits of about $1.1 billion per year, of which about 70 percent is from reduced health care costs.