每周一14：00自动抓取最新NBER工作论文，中文内容由Kimi翻译。欢迎关注微信公众号：芾椿YA
(Updated on: 2025-11-24)
Forecasting Social Science: Evidence from 100 Projects
Authors: Stefano DellaVigna, Eva Vivalt
Abstract: Forecasts about research findings affect critical scientific decisions, such as what treatments an R&D lab invests in, or which papers a researcher decides to write. But what do we know about the accuracy of these forecasts? We analyze a unique data set of all 100 projects posted on the Social Science Prediction Platform from 2020 to 2024, which received 53,298 forecasts in total, including 66 projects for which we also have results. We show that forecasters, on average, over-estimate treatment effects; however, the average forecast is quite predictive of the actual treatment effect. We also examine differences in accuracy across forecasters. Academics have a slightly higher accuracy than non-academics, but expertise in a field does not increase accuracy. A panel of motivated repeat forecasters has higher accuracy, but this does not extend more broadly to all repeat forecasters. Confidence in the accuracy of one’s forecasts is perversely associated with lower accuracy. We also document substantial cross-study correlation in accuracy among forecasters and identify a group of “superforecasters”. Finally, we relate our findings to results in the literature as well as to expert forecasts.
摘要: 关于研究发现的预测影响着关键的科学决策，比如研发实验室投资于哪些治疗方法，或者研究者决定撰写哪些论文。但我们对这些预测的准确性了解多少呢？我们分析了一个独特的数据集，包含了2020年至2024年在社会科学预测平台上发布的所有100个项目，这些项目总共收到了53,298个预测，其中66个项目我们也有结果。我们发现，平均而言，预测者高估了治疗效果；然而，平均预测对实际治疗效果的预测相当准确。我们还检查了预测者之间准确性的差异。学者的准确性略高于非学者，但在某个领域的专业知识并不能提高准确性。一个由积极重复预测者组成的小组具有更高的准确性，但这并不广泛适用于所有重复预测者。对预测准确性的信心与准确性呈反比关系。我们还记录了预测者之间在准确性上的大量跨研究相关性，并识别出一群“超级预测者”。最后，我们将我们的发现与文献中的结果以及专家预测联系起来。
The Lending Technology of Direct Lenders in Private Credit
Authors: Young Soo Jang, Dasol Kim, Amir Sufi
Abstract: We compare the lending technology of direct lenders, banks, and finance companies using a unique data set on secured borrowing by the universe of U.S.-based private middle market firms. The borrowers of direct lenders are distinct relative to those of traditional lenders; they are younger, more likely to be in intangible capital industries, and more likely to be located in the biggest cities in the United States. These differences reflect the focus of direct lenders on private equity-owned firms; direct lenders have negligible impact in industries and cities with low private equity presence. The lending technology of direct lenders is distinct from banks: they have almost no branch network, they are geographically distant from borrowers, they write collateral claims more focused on the continuation value of firms after a default, and they have a higher degree of specialization in certain industries. Direct lenders and private equity sponsors match on industry specialization more strongly than geographic proximity. The findings suggest that direct lenders are not a general substitute for traditional lenders in middle market business lending, but they are instead specialized lenders focused on a particular segment of the U.S. economy with a distinct lending technology.
摘要: 我们使用一个独特的数据集，对美国本土私营中型企业的担保借款进行了比较，比较了直接贷款机构、银行和金融公司的借贷技术。直接贷款机构的借款人与传统贷款机构的借款人有所不同；他们更年轻，更有可能处于无形资本行业，并且更有可能位于美国最大的城市。这些差异反映了直接贷款机构对私募股权拥有的公司的专注；在私募股权存在较低的行业和城市中，直接贷款机构的影响微乎其微。直接贷款机构的借贷技术与银行不同：他们几乎没有分支机构网络，他们与借款人地理位置相距甚远，他们撰写的抵押索赔更加注重公司在违约后的持续价值，并且他们在某些行业的专业化程度更高。直接贷款机构和私募股权赞助商在行业专业化方面的匹配程度比地理位置更强烈。这些发现表明，直接贷款机构并不是传统贷款机构在中型企业贷款中的一般替代品，而是一种专注于美国经济特定细分市场的专业化贷款机构，拥有独特的借贷技术。
The Rise of Market Power and the Macroeconomic Implications: Comment
Authors: C. Lanier Benkard, Nathan H. Miller, Ali Yurukoglu
Abstract: De Loecker et al. (2020) (DEU) estimate that markups increased significantly in the United States from 1955 to 2016. We find this result is sensitive to unreported sample restrictions that drop 27% of the available observations. Applying the methodology as described in the article to the full sample, markup increases are more muted until late in the sample period, and are almost entirely driven by Finance and Insurance firms. If these firms are removed, markup increases are modest. We conclude that the DEU methodology and data, as they are described in the article, do not support the conclusion that broad-based increases in market power have occurred in recent decades. This paper was motivated by an earlier NBER working paper . The authors of that paper have posted a response to this paper . The authors of this paper have prepared a rejoinder to the response.
摘要: De Loecker等人（2020年）（DEU）估计，从1955年到2016年，美国的加价显著增加。我们发现这个结果对未报告的样本限制很敏感，这些限制排除了27%的可用观测值。将文章中描述的方法应用于完整样本，加价增加在样本期的后期更为温和，几乎完全由金融和保险公司推动。如果排除这些公司，加价增加是适度的。我们得出结论，DEU方法和数据，如文章中所述，不支持近几十年市场力量普遍增加的结论。本文的动机是早期的NBER工作论文。那篇论文的作者已经对本文发表了回应。本文的作者已经准备了对回应的反驳。
An Anatomy of the Great Reallocation in US Supply Chain Trade
Authors: Laura Alfaro, Davin Chor
Abstract: This paper documents stylized facts about the “Great Reallocation” in US supply chain trade following the 2018–2019 tariff shocks and the April 2025 Liberation Day announcements. We find that: (i) The US has decoupled from China but not from the world overall. (ii) US imports diversified mainly among its top-20 partners, rather than expanding to new source countries. (iii) Local linear projections confirm ongoing declines in China’s import shares, with compensating increases from Vietnam, Mexico, and Taiwan. (iv) Most of this shift occurred along the product-level intensive margin, though extensive margin adjustments became more pronounced for Vietnam and India from 2021-2024. (v) After a period of “wait and see”, the decline in import shares from China spread to contract-intensive and relationship-sticky goods by 2021-2024. (vi) Early 2025 data suggest that trade reallocation has already accelerated after Liberation Day, in favor of trade partners facing lower additional tariffs and with geographically proximate supply networks. Together, these findings show that the US-China tariff shocks have unwound the US’ sourcing from China back to where it stood at the time of China’s WTO accession.
摘要: 本文记录了2018-2019年关税冲击和2025年4月解放日公告之后，美国供应链贸易中“大重组”的风格化事实。我们发现：（i）美国已经与中国脱钩，但并未与世界整体脱钩。（ii）美国的进口多元化主要在其前20大伙伴之间，而不是扩展到新的来源国。（iii）局部线性投影证实了中国进口份额的持续下降，而越南、墨西哥和台湾的份额相应增加。（iv）这种转变主要是在产品层面的密集边际发生的，尽管从2021-2024年，越南和印度的广泛边际调整变得更加明显。（v）经过一段“观望”期后，从中国的进口份额下降蔓延到合同密集型和关系粘性商品，时间跨度为2021-2024年。（vi）2025年初的数据表明，解放日之后贸易重组已经加速，有利于面临较低额外关税和地理位置接近的供应网络的贸易伙伴。综合来看，这些发现表明，美中关税冲击已经将美国从中国的采购回退到中国加入WTO时的水平。
The Social Rate of Return on Road Infrastructure Investments
Authors: Anusha Chari, Peter Blair Henry, Pablo Picardo
Abstract: A billion people live more than two kilometers from an all-season road, the vast majority of whom reside in emerging-market and developing economies (EMDEs), where the absence of paved roads hinders growth and development. Consistent with this shortage, we estimate that the median social rate of return to installing an additional kilometer of two-lane highway in EMDEs is 55 percent—roughly eight times the social rate of return on private capital in the US. The size and precision of the estimates vary widely across countries. But the eightfold excess social return on roads is, nevertheless, five times larger than the excess financial return on stocks in developing countries that once catalyzed the creation of emerging market equity as an asset class. The absolute and relative magnitudes of these excess returns thus suggest the possibility of substantial unrealized gains from reallocating developed-country savings toward public capital formation in EMDEs. We document these facts using a production function approach to estimate each country’s marginal product of public capital (MPX). We then pair these estimates with country-specific, hedonic road-construction prices (Px) from the Roads Cost Knowledge System to compute social rates of return (Rx). Because macro estimates face well-known identification and measurement challenges, we interpret our findings alongside complementary evidence on financial and micro-level returns.
摘要: 十亿人居住在距离全天候道路两公里以上的地方，其中绝大多数居住在新兴市场和发展中经济体（EMDEs）中，那里缺乏铺装道路阻碍了增长和发展。与这种短缺相一致，我们估计在EMDEs中，每增加一公里双车道高速公路的社会平均回报率为55%，大约是美国私人资本社会回报率的八倍。这些估计的大小和精确度在不同国家之间差异很大。但是，道路的社会回报率超过八倍，仍然是发展中国股票超额金融回报率的五倍，后者曾经催化了新兴市场股票作为一个资产类别的创造。这些超额回报的绝对和相对规模因此表明，将发达国家的储蓄重新分配到EMDEs的公共资本形成中，可能会有大量的未实现收益。我们使用生产函数方法来估计每个国家的公共资本边际产品（MPX），然后我们将这些估计值与来自道路成本知识系统（Roads Cost Knowledge System）的国家特定、享乐主义道路建设价格（Px）配对，以计算社会回报率（Rx）。由于宏观估计面临众所周知的识别和测量挑战，我们将我们的发现与金融和微观层面回报的补充证据一起解释。
Malleable Minds: The Effects of STEM- vs. Humanities-Focused Curricula
Authors: Robert Ainsworth, Rajeev H. Dehejia, Andrei Munteanu, Cristian Pop-Eleches, Miguel Urquiola
Abstract: We examine the impacts of assignment to STEM vs. humanities-focused curricula in Romania’s high school system. We apply a regression discontinuity design to administrative and survey data to estimate effects on educational pathways, desired careers, and non-cognitive outcomes. An overarching theme of our findings is the malleability of students to what they study. Assignment to STEM increases STEM college enrollment and technology or engineering career intentions by 25 pp. Exploring mechanisms, we find that STEM assignment changes students’ self-perceived academic abilities and their preferences over academic subjects and job tasks. STEM assignment is risky for low-achieving students, reducing their chances of passing a high school exit exam and enrolling in college. A final finding is that STEM makes boys more conservative, while shifting some of girls’ views to the left. Our results identify a strategy for promoting STEM higher education and careers, but also highlight potential tradeoffs.
摘要: 我们研究了在罗马尼亚高中系统中，被分配到STEM（科学、技术、工程和数学）与人文学科课程的影响。我们应用了一种回归不连续设计，结合行政和调查数据来估计其对教育路径、期望职业和非认知结果的影响。我们发现的一个总体主题是学生对他们所学内容的可塑性。被分配到STEM课程可以增加25个百分点的STEM大学入学率和技术或工程职业意向。在探索机制时，我们发现STEM课程的分配改变了学生对自身学术能力的看法，以及他们对学术科目和工作任务的偏好。对于成绩较低的学生来说，STEM课程的分配是有风险的，降低了他们通过高中毕业考试和进入大学的机会。最后一个发现是，STEM课程使男孩变得更加保守，而改变了一些女孩的观点向左倾斜。我们的结果识别了一种促进STEM高等教育和职业的策略，但也突出了潜在的权衡。
Beliefs and Portfolios: Causal Evidence
Authors: Johannes Beutel, Michael Weber
Abstract: We causally test alternative theories of expectation formation. Using a randomized information experiment we show overreaction is a key feature of individuals’ return expectations, and individuals’ response to the price-earnings ratio is opposite to the academic consensus. Our evidence is inconsistent with standard models of expectation formation but subjective mental models that deviate from objective benchmarks can jointly explain the updating behavior in the experiment, the link between individuals’ prior perceptions and expectations, and the heterogeneity of updating. Conditional on their beliefs, individuals’ sensitivity of equity shares in a hypothetical portfolio choice experiment is consistent with the standard Merton model.
摘要: 我们对预期形成的不同理论进行了因果检验。通过一个随机化的信息实验，我们展示了过度反应是个体回报预期的一个关键特征，个体对市盈率的反应与学术共识相反。我们的证据与标准的预期形成模型不一致，但偏离客观基准的主观心理模型可以共同解释实验中的更新行为、个体先前感知与预期之间的联系，以及更新的异质性。在他们的信念条件下，个体在假设的投资组合选择实验中对股票的敏感性与标准的默顿模型一致。
The Logic of State Surveillance
Authors: Gemma Dipoppa, Annalisa Pezone
Abstract: All states adopt systems to surveil political activists. How do they decide whom to watch and why? We study the logic of state surveillance using the first complete individual-level database of those monitored by a state — 152,000 Italians born between 1816 and 1932, encompassing both democratic and authoritarian regimes. We focus on education: exploiting a discontinuous expansion in primary schooling in municipalities above a population and age threshold, we show that cohorts exposed to more years of school experienced an uptick in surveillance. The effect is largest for working classes, who were monitored for longer periods, subjected to harsher measures, and disproportionately targeted when affiliated with communist ideologies. Yet treated cohorts did not become more politically active, indicating that surveillance expanded not in reaction to increased mobilization, but as a preventive strategy rooted in fears of working-class empowerment. These findings reveal how states view educated yet excluded groups as politically threatening and prioritize their surveillance, potentially generating inequalities in groups’ ability to influence political change.
摘要: 所有国家都采用系统来监视政治活动家。他们如何决定监视谁以及为什么监视？我们利用第一个完整的个体层面数据库来研究国家监视的逻辑——这个数据库包含了152,000名意大利人，他们出生于1816年至1932年之间，涵盖了民主和威权政权。我们关注教育：利用在人口和年龄门槛以上的市镇中，初等教育的不连续扩展，我们展示了接受更多年学校教育的群体经历了监视的增加。对于工人阶级来说，这种影响最大，他们被监视的时间更长，受到的措施更严厉，并且在与共产主义意识形态有关联时，他们被不成比例地针对。然而，受到处理的群体并没有变得更加政治活跃，这表明监视的扩大并不是对增加动员的反应，而是作为一种预防策略，根源于对工人阶级赋权的恐惧。这些发现揭示了国家如何将受过教育但被排除在外的群体视为政治威胁，并优先监视他们，可能在群体影响政治变革的能力上产生不平等。
GCAP Public Security-Level Data on U.S. Fund Holdings
Authors: Bruno Cavani, Matteo Maggiori, Jesse Schreger
Abstract: We construct representative security-fund-level longitudinal data for the United States using regulatory filings of portfolio holdings from Form N-PORT. We validate our dataset by comparing coverage and composition to official statistics from the Federal Reserve’s Financial Accounts of the United States (formerly known as the Flow of Funds) and Treasury International Capital (TIC) System, and to micro-level commercial datasets. We showcase an application by replicating and updating Maggiori, Neiman and Schreger (2020) findings on home currency bias using N-PORT instead of commercial fund holdings data. Our results confirm that N-PORT data offer a comprehensive, reliable, and public source for research in macroeconomics and finance. We make all the security-level data on holdings available in a public repository of the GCAP Lab and provide code for updating the data.
摘要: 我们使用美国监管文件中的投资组合持有情况，从N-PORT表格构建了代表性的美国证券基金层面的纵向数据。我们通过与联邦储备的美国金融账户（以前称为资金流动）和财政部国际资本（TIC）系统的官方统计数据，以及微观层面的商业数据集进行比较，来验证我们的数据集。我们通过使用N-PORT而不是商业基金持有数据，复制并更新Maggiori, Neiman和Schreger（2020）关于本币偏好的发现，展示了一个应用案例。我们的结果证实，N-PORT数据为宏观经济学和金融研究提供了一个全面、可靠且公开的数据来源。我们在GCAP实验室的公共存储库中提供了所有证券级别的持有数据，并提供了更新数据的代码。
Laptops in the Long Run: Evidence from the One Laptop per Child Program in Rural Peru
Authors: Santiago Cueto, Diether W. Beuermann, Julian Cristia, Ofer Malamud, Francisco Pardo
Abstract: This paper examines a large-scale randomized evaluation of the One Laptop Per Child (OLPC) program in 531 Peruvian rural primary schools. We use administrative data on academic performance and grade progression over 10 years to estimate the long-run effects of increased computer access on (i) school performance over time and (ii) students’ educational trajectories. Following schools over time, we find no significant effects on academic performance but some evidence of negative effects on grade progression. Following students over time, we find no significant effects on primary and secondary completion, academic performance in secondary school, or university enrollment. Survey data indicate that computer access significantly improved students’ computer skills but not their cognitive skills; treated teachers received some training but did not improve their digital skills and showed limited use of technology in classrooms, suggesting the need for additional pedagogical support.
摘要: 本文检验了在秘鲁531所农村小学进行的“每个孩子一台笔记本电脑”（OLPC）计划的大规模随机评估。我们使用10年来的学术表现和年级晋升的行政数据来估计增加计算机使用对（i）学校随时间的学术表现和（ii）学生教育轨迹的长期影响。随着时间的推移对学校进行跟踪，我们发现对学术表现没有显著影响，但有一些证据表明对年级晋升有负面影响。随着时间的推移对学生进行跟踪，我们发现对小学和中学的完成情况、中学的学术表现或大学入学没有显著影响。调查数据表明，计算机使用显著提高了学生的计算机技能，但并没有提高他们的认知技能；接受治疗的教师接受了一些培训，但没有提高他们的数字技能，并且在课堂上对技术的利用有限，这表明需要额外的教学支持。
Tracking the Short-Run Price Impact of U.S. Tariffs
Authors: Alberto Cavallo, Paola Llamas, Franco M. Vazquez
Abstract: We use high-frequency retail microdata to measure the short-run impact of the 2025 U.S. tariffs on consumer prices. By matching daily prices from major U.S. retailers to product-level tariff rates and countries of origin, we construct price indices that isolate the direct effects of tariff changes across goods and trading partners. Prices began rising immediately after the broader tariff measures announced in early March and continued to increase gradually over subsequent months, with imported goods rising roughly twice as much as domestic ones. Our estimated retail tariff pass-through is 20 percent, with a cumulative contribution of about 0.7 percentage points to the all-items Consumer Price Index by September 2025. Our results show that tariff costs were gradually but steadily transmitted to U.S. consumers, with additional spillovers to domestic goods.
摘要: 我们利用高频零售微观数据来衡量2025年美国关税对消费者价格的短期影响。通过将美国主要零售商的每日价格与产品级别的关税税率和原产国相匹配，我们构建了价格指数，这些指数隔离了关税变化对商品和贸易伙伴的直接影响。自3月初宣布更广泛的关税措施后，价格立即开始上升，并在随后的几个月中持续逐渐上升，进口商品的价格上涨幅度大约是国产商品的两倍。我们估计的零售关税传导率为20%，到2025年9月对所有商品消费者价格指数的累积贡献约为0.7个百分点。我们的结果表明，关税成本逐渐但稳定地转嫁给了美国消费者，并对国内商品产生了额外的溢出效应。
Internal vs. External Shocks in Weakening Democracies: Evidence on Migration and Foreign Investment
Authors: Assaf Razin
Abstract: This paper investigates the consequences of regime change for both migration and foreign direct investment (FDI) by employing quasi-natural experiments that exploit external and internal shocks to democratic institutions. It compares evidence from Europe, which was afflicted by the “Syrian Shock”—an external institutional stress testing administrative and fiscal capacity—and Israel, which experienced the “Corruption Shock”—an internal credibility crisis that eroded judicial independence and policy predictability. These two shocks provide a natural experiment to examine how weakening democratic institutions influence both capital mobility and people mobility.
摘要: 本文通过利用外部和内部对民主制度的冲击所进行的准自然实验，研究政权更迭对移民和外国直接投资（FDI）的影响。它比较了来自欧洲和以色列的证据。欧洲遭受了“叙利亚冲击”——一个外部制度压力测试，测试了行政和财政能力；以色列则经历了“腐败冲击”——一个内部信任危机，侵蚀了司法独立和政策可预测性。这两个冲击提供了一个自然实验，以检验民主制度的弱化如何影响资本流动和人员流动。
Playing Catch-up with Health Savings
Authors: Jacob Berman, Adam Bloomfield, Sita Slavov
Abstract: We use comprehensive tax data to study how saving behavior responds to the Health Savings Account (HSA) “catch-up” contribution provision, which raises HSA contribution limits for individuals aged 55 and older. Using a regression discontinuity design, we find a sharp increase in contributions among those previously near the limit and smaller increases among unconstrained savers. Induced contributions are not immediately withdrawn and do not appear to crowd out retirement savings. Responses are strongest among payroll contributors and long-term savers. However, married couples do not appear to coordinate their HSA behavior to take advantage of the complex spousal rules governing catch-up contributions. Our findings highlight how tax incentives shape HSA saving and suggest that tax-advantaged account design meaningfully affects household financial behavior.
摘要: 我们利用全面的税收数据研究储蓄行为如何响应健康储蓄账户（HSA）的“追加”供款规定，该规定提高了55岁及以上个人的HSA供款限额。通过使用回归不连续设计，我们发现那些之前接近限额的人的供款急剧增加，而不受限制的储蓄者的增加较小。诱导的供款并没有立即被提取，并且似乎并没有挤出退休储蓄。反应最强的是工资单供款者和长期储蓄者。然而，已婚夫妇似乎并没有协调他们的HSA行为以利用复杂的配偶追加供款规则。我们的发现突出了税收激励如何塑造HSA储蓄，并表明税收优惠账户设计显著影响家庭财务行为。
Authors: Aakash Bhalothia, Gavin Engelstad, Gaurav Khanna, Harrison Mitchell
Abstract: We estimate the economic value of cities worldwide, using detailed job histories for 513 million workers in 220,000 cities across 191 countries. These estimates allow us to identify why some cities are more productive than others and to quantify the earnings gains from migration throughout the development process. Our data contain job spells—with start and end dates, establishment names, locations, job titles, and effective salaries—enabling an event-study movers design with individual and time fixed effects. Moving to higher-value cities leads to immediate increases in job seniority, shifts into better-paid industries and occupations, and large overall earnings gains. The global scope of the data lets us compare internal and international moves and assess how the productivity advantages of cities differ by country income level. Across borders, 93% of wage changes reflect city effects, while within countries this share ranges from 45–73%. High-income countries exhibit stronger ability-based sorting, reducing the proportion attributable to place. City effects rise with industrial diversity and population, consistent with agglomeration economies, and more productive cities allocate workers to higher-productivity firms. The wide dispersion of city effects within countries highlights substantial potential gains from migration, particularly in low-income, less-urbanized economies. Reallocating workers to match the US distribution yields sizable wage gains in developing countries.
摘要: 我们通过分析全球191个国家220,000个城市中5.13亿工人的详细就业历史，估算了世界各地城市的经济价值。这些估算使我们能够识别为什么一些城市比其他城市更具生产力，并量化整个发展过程中迁移带来的收入增长。我们的数据包含工作记录——包括开始和结束日期、机构名称、地点、职位头衔和有效薪资——使我们能够采用个体和时间固定效应的事件研究迁移者设计。迁移到价值更高的城市会导致工作资历的立即增加，转向薪资更高的行业和职业，以及整体收入的大幅增长。数据的全球范围让我们能够比较国内和国际迁移，并评估城市生产力优势如何因国家收入水平而异。跨国迁移中，93%的工资变化反映了城市效应，而在国内这一比例从45%到73%不等。高收入国家表现出更强的能力基础排序，降低了归因于地点的比例。城市效应随着工业多样性和人口的增加而上升，与集聚经济一致，更具生产力的城市将工人分配到更高生产力的企业。城市效应在国内的广泛分散突出了迁移的巨大潜在收益，特别是在低收入、城市化程度较低的经济体中。将工人重新分配以匹配美国的分布，在发展中国家能够带来可观的工资增长。
Misaligned by Design: Incentive Failures in Machine Learning
Authors: David Autor, Andrew Caplin, Daniel J. Martin, Philip Marx
Abstract: The cost of error in many high-stakes settings is asymmetric: misdiagnosing pneumonia when absent is an inconvenience, but failing to detect it when present can be life-threatening. Accordingly, artificial intelligence (AI) models used to assist such decisions are frequently trained with asymmetric loss functions that incorporate human decision-makers’ trade-offs between false positives and false negatives. In two focal applications, we show that this standard alignment practice can backfire. In both cases, it would be better to train the machine learning model with a loss function that ignores the human’s objective and then adjust predictions ex post according to that objective. We rationalize this result using an economic model of incentive design with endogenous information acquisition. The key insight from our theoretical framework is that machine classifiers perform not one but two incentivized tasks: choosing how to classify and learning how to classify. We show that while the adjustments engineers use correctly incentivize choosing, they can simultaneously reduce the incentives to learn. Our formal treatment of the problem reveals that methods embraced for their intuitive appeal can in fact misalign human and machine objectives in predictable ways.
摘要: 在许多高风险情境中，错误的代价是不对称的：在没有肺炎的情况下误诊肺炎是一种不便，但如果在存在肺炎时未能检测到它，则可能危及生命。因此，用于协助此类决策的人工智能（AI）模型通常使用包含人类决策者在假阳性和假阴性之间权衡的不对称损失函数进行训练。在我们关注的两个应用中，我们展示了这种标准的对齐实践可能会适得其反。在这两种情况下，最好训练机器学习模型使用一个忽略人类目标的损失函数，然后根据该目标调整预测结果。我们使用具有内生信息获取的经济激励设计模型来合理化这一结果。我们理论框架的关键洞见是，机器分类器执行的不仅仅是一个激励任务：选择如何分类和学习如何分类。我们展示了虽然工程师使用的调整正确地激励了选择，但它们可能同时减少了学习的激励。我们对问题的正式处理揭示了因其直观吸引力而被采用的方法实际上可能会以可预测的方式使人类和机器的目标不一致。
The Spatial Distribution of Income in Cities: New Global Evidence and Theory
Authors: Peter Deffebach, David Lagakos, Yuhei Miyauchi, Eiji Yamada
Abstract: We study how the spatial distribution of income and commuting patterns within cities vary across the development spectrum, drawing on new granular data from 50,000 neighborhoods in 121 cities across developed and developing countries. We document that in developing countries, poorer urban households are significantly more likely to live far from city centers, in hilly terrain, and near rivers. These patterns are absent or reversed in developed cities. Commuting shares decline more sharply with distance in less developed countries, indicating higher commuting costs that exacerbate spatial inequality in job access. Job-access measures are considerably worse for the urban poor than for the urban rich in developing countries, while the opposite is true in developed countries. We interpret these findings in a quantitative urban model and show that a parsimonious set of factors—nonhomothetic preferences over amenities, commuting costs, and the spatial concentration of jobs—helps explain most of the cross-country patterns we document.
摘要: 我们研究了收入空间分布和城市内通勤模式如何在不同发展水平的国家之间变化，利用来自发达国家和发展中国家121个城市的50,000个社区的新细致数据。我们记录了在发展中国家，较贫穷的城市家庭显著更有可能居住在远离市中心、多丘陵地形和靠近河流的地方。这些模式在发达国家城市中不存在或相反。在不太发达的国家，通勤份额随着距离的增加而更急剧地下降，表明通勤成本更高，加剧了工作机会的空间不平等。在发展中国家，城市贫困人口的工作机会衡量指标明显比城市富裕人口更差，而在发达国家则相反。我们在一个定量的城市模型中解释了这些发现，并表明一组简洁的因素——对便利设施的非同质偏好、通勤成本和工作的地理集中——有助于解释我们记录的大多数跨国模式。
Paternalistic Social Assistance: Evidence and Implications from Cash vs. In-Kind Transfers
Authors: Anna Chorniy, Amy Finkelstein, Matthew J. Notowidigdo
Abstract: We estimate and compare impacts of cash and in-kind transfers on the consumption of temptation goods in the same population, and explore normative implications. We use two decades of data from South Carolina on cash benefits from Supplemental Security Income (SSI) and in-kind benefits from the Supplemental Nutrition Assistance Program (SNAP) linked to detailed data on adults’ health care use. Our empirical strategy examines outcome changes in the several days following each transfer’s scheduled monthly payout. Emergency department visits for drug and alcohol use increase by 20-30 percent following SSI receipt, but do not respond to SNAP receipt. Fills of prescription drugs for new illnesses also increase following SSI receipt but do not respond to SNAP receipt. Motivated by these non-fungibility results, we develop a model of a paternalistic social planner choosing the mix of cash and SNAP for a fixed-budget transfer program when consumers have self-control problems and may engage in mental accounting. We show that the planner’s optimal SNAP share is strictly positive and weakly increasing as self-control worsens. Moreover, with heterogeneity in self-control and mental accounting, the planner may choose to use SNAP even when they have access to a uniform Pigouvian tax on the temptation good.
摘要: 我们估计并比较了现金和实物转移支付对同一人群中诱惑品消费的影响，并探讨了其规范性含义。我们使用了来自南卡罗来纳州二十年的数据，这些数据涉及补充保障收入（SSI）的现金福利和补充营养援助计划（SNAP）的实物福利，并与成年人的医疗使用详细数据相链接。我们的实证策略检查了在每次计划的月度支付后的几天内结果的变化。急诊科对药物和酒精使用访问在收到SSI后增加了20-30%，但对SNAP的接收没有反应。新疾病的处方药填充也在收到SSI后增加，但对SNAP的接收没有反应。基于这些非货币化结果，我们开发了一个模型，模型中的家长式社会规划者在消费者存在自我控制问题并可能进行心理会计时，为固定预算转移计划选择现金和SNAP的组合。我们展示了规划者的最佳SNAP份额是严格正的，并且随着自我控制的恶化而弱增加。此外，随着自我控制和心理会计的异质性，规划者甚至可能选择使用SNAP，即使他们可以访问统一的Pigouvian税对诱惑品征税。
Linear Regressions with Combined Data
Authors: Xavier D’Haultfoeuille, Christophe Gaillac, Arnaud Maurel
Abstract: We study linear regressions in a context where the outcome of interest and some of the covariates are observed in two different datasets that cannot be matched. Traditional approaches obtain point identification by relying, often implicitly, on exclusion restrictions. We show that without such restrictions, coefficients of interest can still be partially identified, with the sharp bounds taking a simple form. We obtain tighter bounds when variables observed in both datasets, but not included in the regression of interest, are available, even if these variables are not subject to specific restrictions. We develop computationally simple and asymptotically normal estimators of the bounds. Finally, we apply our methodology to estimate racial disparities in patent approval rates and to evaluate the effect of patience and risk-taking on educational performance.
摘要: 我们研究了在线性回归的背景下，感兴趣的结果和一些协变量在两个不同的数据集中被观察到，而这些数据集无法匹配。传统的方法通过依赖排除限制来获得点识别，通常是隐含的。我们展示了即使没有这样的限制，感兴趣的系数仍然可以部分识别，并且尖锐的界限具有简单的形式。当在两个数据集中都观察到但不包括在感兴趣的回归中的变量可用时，我们可以获得更紧密的界限，即使这些变量不受特定限制。我们开发了计算简单且渐近正态的界限估计器。最后，我们将我们的方法应用于估计专利批准率中的种族差异，并评估耐心和冒险行为对教育表现的影响。
The Local-Area Incidence of Exporting
Authors: Christoph Boehm, Aaron B. Flaaen, Nitya Pandalai-Nayar, Jan Schlupp
Abstract: We develop a new algorithm to map confidential firm-level export transactions to their underlying establishments that can be implemented on U.S. microdata. Using this procedure, we construct a novel micro-dataset of U.S. exports at the plant level. Aggregation of these data permits more accurate measurement of exporting at the subnational level (e.g., county, MSA, etc.) than was previously possible. The data reveal exports to be much more geographically concentrated than both employment and manufacturing sales, implying that some regions are heavily reliant on foreign demand. To illustrate the consequences of such exposure, we study the effects of the trade collapse during the Great Recession on local labor markets. Counties experiencing greater declines in foreign demand performed worse in terms of employment, pay, and wages during the Great Recession. A similar analysis implemented with publicly available imputed export data—a common practice in the literature—fails to replicate these estimates.
摘要: 我们开发了一种新算法，将机密的公司层面出口交易映射到它们背后的具体工厂，该算法可以应用于美国的微观数据。通过这一程序，我们构建了一个新颖的美国工厂层面的出口微观数据集。这些数据的聚合使得在次国家级（例如，县、大都市区等）的出口测量比以往更为准确。数据显示，出口在地理上的集中度远高于就业和制造业销售，这意味着一些地区严重依赖外国需求。为了说明这种暴露的后果，我们研究了大萧条期间贸易崩溃对当地劳动力市场的影响。在大萧条期间，外国需求下降更大的县在就业、薪酬和工资方面表现得更差。如果使用公开可获得的估算出口数据进行类似的分析——这是文献中的常见做法——则无法复制这些估计值。
Target Allocation Funds, Strategic Complementarities, and Market Fragility
Authors: Chuck Fang, Itay Goldstein
Abstract: Target allocation funds (TAFs) make predictable rebalancing trades to maintain portfolio weights across asset classes. During the COVID-19 stock market crash, TAFs sold $59 billion of bond fund shares and simultaneously purchased a similar amount of equity fund shares. We show that TAF rebalancing triggers strategic complementarity: bond mutual funds facing larger rebalancing-induced sales by TAFs experienced greater outflows from other non-TAF investors. This effect was particularly pronounced for illiquid bond funds and amplified total bond fund outflows during COVID-19 by an additional $27 billion. Rebalancing by TAFs, together with the strategic amplification by other investors, transmits equity market shocks to bond returns, accounting for 17% of the rise in stock-bond correlation over the past decade.
摘要: 目标配置基金（TAFs）进行可预测的再平衡交易，以维持各类资产的组合权重。在COVID-19股市崩盘期间，TAFs卖出了590亿美元的债券基金份额，并同时购买了类似金额的股票基金份额。我们展示出TAF的再平衡触发了策略互补性：面对TAFs更大再平衡诱导销售的债券共同基金，经历了来自其他非TAF投资者更大的资金流出。这种效应对于流动性较差的债券基金尤为明显，并在COVID-19期间额外增加了270亿美元的债券基金总流出。TAFs的再平衡行为，以及其他投资者的战略性放大效应，将股市冲击传递到债券回报，占过去十年股票-债券相关性上升的17%。
Shutting Down Japantown: The Effects of WWII Internment on Japanese American Enclaves
Authors: Martin H. Saavedra, Tate Twinam
Abstract: During World War II, the U.S. government incarcerated all West Coast Japanese Americans in internment camps. We ask how this forced displacement affected Japanese American enclaves. Using the recently digitized 1940 and 1950 full-count censuses, we measure changes in the racial composition of neighborhoods across 14 major cities. We find that internment reduced the Japanese American population of enumeration districts within the exclusion zone by 25–50% relative to their 1940 levels, and that these individuals were replaced by African American in-movers in a nearly 1-to-1 fashion. Outside the exclusion zone, new Japanese American enclaves formed, but did not approach the scale of their historic West Coast counterparts.
摘要: 在第二次世界大战期间，美国政府将所有西海岸的日裔美国人关押在拘留营中。我们询问这种被迫流离失所如何影响了日裔美国人的聚居区。利用最近数字化的1940年和1950年的完整人口普查数据，我们测量了14个主要城市中邻里种族构成的变化。我们发现，拘留使排除区内的日本裔美国人口比1940年的水平减少了25-50%，而这些个体几乎被非裔美国人以1对1的方式取代。在排除区之外，新的日裔美国人聚居区形成了，但没有达到他们历史上西海岸同类聚居区的规模。
Unraveling the Drivers of Energy-saving Technical Change
Authors: Diego R. Känzig, Charles T. Williamson
Abstract: We explore the increasing divergence between economic growth and energy consumption through energy-saving technical progress. Proposing a new measure of energy-saving technology, we study the underlying drivers in a semi-structural model of the U.S. economy. Our analysis shows that energy price shocks reduce consumption and stimulate energy-saving innovation, but also cause economic downturns and crowd out other innovations. Only energy-saving technology shocks can explain the negative co-movement between output and energy use. These sudden efficiency gains emerge as the primary driver of energy-saving technical change. Our findings highlight the importance of fostering energy-saving innovations in transitioning to a low-carbon economy.
摘要: 我们通过节能技术进步来探讨经济增长与能源消费之间日益扩大的分歧。提出了一种新的节能技术度量方法，我们在美国经济的半结构化模型中研究了背后的驱动因素。我们的分析表明，能源价格冲击会减少消费并刺激节能创新，但也会导致经济衰退并挤出其他创新。只有节能技术冲击才能解释产出与能源使用之间的负向联动。这些突然的效率提升成为节能技术变革的主要驱动力。我们的发现强调了在向低碳经济转型中促进节能创新的重要性。
Dilution vs. Risk Taking: Capital Gains Taxes and Entrepreneurship
Authors: Eduardo M. Azevedo, Florian Scheuer, Kent Smetters, Min Yang
Abstract: Recent proposals to tax unrealized capital gains or wealth have sparked a debate about their impact on entrepreneurship. We show that accrual-based taxation creates two opposing effects: successful founders face greater dilution from advance tax payments, whereas unsuccessful founders receive tax credits that effectively provide insurance. Using comprehensive new data on U.S. venture capital deals, we find that founder returns remain extremely skewed, with 84% receiving zero exit value while the top 2% capture 80% of total value. Moving from current realization-based to accrual-based taxation would reduce founder ownership at exit by 25% on average but would also increase the fraction receiving positive payoffs from 16% to 47% when tax credits are refunded. Embedding these distributions in a dynamic career choice model, we find that founders with no or moderate risk aversion prefer the current realization-based tax system, while more risk-averse founders prefer accrual-based taxation. We estimate that a 2% annual wealth tax has a similar impact on dilution as taxing unrealized capital gains but produces no risk-sharing benefits due to the absence of tax credits in case of down rounds.
摘要: 最近提议对未实现的资本收益或财富征税引发了关于它们对创业影响的辩论。我们表明，基于应计的税收制度会产生两种对立的效果：成功的创始人面临更大的提前缴税导致的稀释，而失败的创始人则获得税收抵免，实际上提供了保险。利用美国风险投资交易的全新综合数据，我们发现创始人的回报仍然极度倾斜，84%的人获得的退出价值为零，而前2%的人占据了总价值的80%。从当前基于实现的税收制度转变为基于应计的税收制度，平均会减少创始人退出时的所有权25%，但如果退还税收抵免，获得正回报的比例也会从16%增加到47%。将这些分布嵌入动态职业选择模型中，我们发现没有或适度风险厌恶的创始人更倾向于当前基于实现的税收制度，而更风险厌恶的创始人则更倾向于基于应计的税收制度。我们估计，每年2%的财富税对稀释的影响与对未实现资本收益征税相似，但由于在估值下调的情况下缺乏税收抵免，因此不会产生风险共担的好处。