Masahiko Aoki Best Paper Award (青木昌彦经济学论文奖)
Presentations: AEA, CUFE, CUHK Shenzhen, Jinan, Lingnan, Macau, NBER China Economy, NBER Summer Institute-Political Economy, NYU-Shanghai, PKU HSBC, Shanghai Tech, SITE 2025 (Stanford), University of Chicago-Northwestern China Workshop, UCSD, Virginia Darden, World Bank
Abstract: We decode China's industrial policies from 2000 to 2022 by employing Large Language Models (LLMs) to extract and analyze rich information from a comprehensive dataset of 3 million documents issued by central, provincial, and municipal governments. Through careful prompt engineering, multistage extraction and refinement, and rigorous verification, LLMs allow us to extract structured information on detailed policy dimensions, including context and scope, targeted industries, tools, implementation mechanisms, and intergovernmental relationships, etc. Combining these newly constructed industrial policy data with microlevel firm data, we document a list of facts about China's industrial policy that explore the following critical questions. Which industries are targeted and how does this align with local comparative advantage? What policy tools are deployed, and how does their use vary across different levels and regions of governments, as well as over the various phases of development of an industry? We also examine the impact of these policies on firm behavior, including entry, production, and productivity growth, and highlight the heterogeneous effects of different policy tools. In addition, we explore the political economy of industrial policy, focusing on top-down transmission mechanisms, policy diffusion, and persistence across regions. Finally, we document spatial inefficiencies and industry-wide overcapacity as potential downsides of industrial policies.
CSEF-RCFS Best Paper Award
Presentations: AFA, CSEF-RCFS Conference on Finance, Labor and Inequality, CICF, EFA, FMA, PHBS-CHUKSZ, UBC, UNSW
Abstract: We study the role of national culture in explaining within-firm pay inequality in closely-held firms owned by immigrants using a unique employee-employer matched dataset linked with firm ownership and immigrant records in Canada over the 2001 – 2017 period. We find that culture that immigrant owners carry from their home countries is an economically significant determinant of pay inequality within their firms. We show that Hofstede’s individualism is a key cultural dimension affecting within-firm pay inequality: firms owned by individuals from more individualistic countries have larger pay inequality. We show that the impact of culture on within-firm pay inequality is causal. In a difference-in-differences setting using firms that undergo ownership changes, we find a significant increase in within-firm pay inequality after the firm was taken over by immigrant owners from a country with higher within-firm pay inequality or from a more individualistic country. We find similar results among employee stayers; among employee stayers in firms within a labor-intensive industry where production technology is comparable; when the owner changes were caused by deaths of prior owners. Overall, our findings suggest that informal institutions such as national culture are important determinants of income inequality.
Presentations: Nankai, PHBS-CHUKSZ, Shanghai Tech Household Finance Conference
Abstract: This paper studies the implication of workplace networks on the housing market, using a unique employer branch–employee–residence–housing transaction linked panel for Massachusetts (2006–2023) constructed by linking LinkedIn work histories to property transactions and residential records covering both owners and renters. Home purchases comove strongly within branches: controlling for sorting and shared firm-level shocks with occupation-year and company-year fixed effects, an employee's home purchase probability increases by over 20% of the sample mean following one additional coworker's purchase in the last three years. Exploiting the post-2019 shift to remote work we confirm social interaction as the key mechanism: within firm-years, the coworker effect declines significantly for remote-suitable occupations post-pandemic, with no significant trend before the pandemic. The network's influence extends beyond the current firm, as purchases by out-of-city former coworkers or current coworkers' former colleagues also predict home purchases. Crucially, the network transmits information about economic fundamentals: employees buy more when their employers expand and less when employer risk increases, and the performance of a former coworker's current firm also predicts an employee's purchase, indicating cross‑firm transmission. Finally, the influence of coworker network and firm fundamental increases in hot and uncertain markets. Overall, by transmitting firm-level shocks to the housing decisions of current employees and their former coworkers, workplace networks act as an important micro-level channel that propagates local economic shocks into the housing market.
Presentations: ABFER, AIDE, AFA, Bocconi, Cambridge, CEBRA, NBER China Economy Group, Oxford, PKU HSBC School, Shanghai Virtual Finance Seminar, Texas AM
Abstract: By comparing uncollateralized business loans made by a big tech lending program with conventional bank loans, we find that big tech loans tend to be smaller and have higher interest rates and that borrowers of big tech loans tend to repay far before maturity and borrow more frequently. These patterns remain for borrowers with access to bank credit. Our findings highlight the big tech lender’s roles in serving borrowers’ short-term liquidity rather than their long-term financing needs. Through this model, big tech lending facilitates credit to borrowers underserved by banks without experiencing more-severe adverse selection or incurring greater risks than banks (even during the COVID-19 crisis).
Abstract: I examine racial bias in the most popular online home valuation algorithm and study the impact of the algorithmic information on racial price differentials in the U.S. housing market. I find that racial bias in the algorithm is much smaller than racial price differentials in the market. For example, while Black (Hispanic) households overpay (oversell) by 9.3% (1.9%) in prices relative to White households for similar homes, the algorithm only overvalues the same transactions by 1.1% (0.6%). The algorithm inadvertently learns racial bias from patterns in historical transaction prices. The algorithmic racial bias is small partly because the algorithm is designed to be insensitive to transitory pricing factors including buyer or seller race. Exploiting the staggered coverage of the algorithm across counties, I study the causal impact of the algorithmic information on a sample of neighboring ZIP Codes in bordering counties. I find that if the algorithmic valuation is available for all the homes in an area, it reduces the overpayment of Black buyers relative to White buyers by 4.8%, negating more than half of the corresponding racial price differentials. The findings suggest that a slightly biased algorithm may still reduce racial disparity in a market, and caution against regulation on algorithms solely based on the existence of algorithmic bias.
Abstract: We show that firms sometimes pay to license patents even after they expire. Because legal exclusivity ends at expiration, these payments reveal the existence of de facto barriers to entry that operate beyond formal intellectual property rights. Building on the literature on technology appropriability and technology diffusion, we use large‑language‑model (LLM) analysis of patent text to construct five measures of an invention: (i) its reliance on firm‑specific complementary assets, (ii) relatedness to the firm’s existing patent portfolio, (iii) disclosure opacity in omitting critical know-how, (iv) the extent to which an invention builds on frontier and complex science as opposed to applied know‑how, and (v) application scope. Patents scoring high on the first three dimensions are more likely to be licensed post‑expiration. Such patents also exhibit higher technology concentration as measured by the HHI of subsequent citations. Conversely, science‑anchored patents—those building more on frontier research than on applied know‑how—are less likely to be licensed post‑expiration and diffuse more widely (lower HHI). Application scope is unrelated to either outcome. The results are robust to controlling for a rich set of established patent characteristics. While our focus on post-expiration licensing already controls for legal barriers, we further show that the results are robust to the inclusion of licensor and licensee fixed effects capturing firm-level heterogeneity (e.g., a licensor’s marketing power or a licensee’s lack of expertise) or the inclusion of licensor-licensee fixed effects and measures capturing the relationship dependence between the licensor and licensee. We interpret the overall patterns as evidence that complementary assets, portfolio embeddedness, and opaque disclosures create durable technology‑specific “moats”, while science‑anchored inventions facilitate broad technology diffusion.
Abstract: Employees spend most of their time working and learning at firms. Firms play a critical role in developing employees’ human capital and matching skills to productive tasks. Yet, the role and mechanisms of firms in developing and deploying human capital, and the implications of human capital accumulation and (mis)matching remain underexplored. We compiled a unique dataset of time varying skills from resume data and matched them to job posting and employee reviews to study the mechanism through which firms influence employee human capital accumulation, matching skills to tasks, and their implications on firms, employees, and local labor market.
Abstract: We examine why environmental and social (E/S) performance vary across countries and firms, and evaluate the value implications. Using a sample of 33,021 firm-year observations representing 4,587 firms from 43 countries over the 2003–2015 period and applying hierarchical linear modeling, we find that individualism is positively associated with firm-level E/S performance. We show that two country-level channels— freedom of the press and protection of equal rights—and three firm-level channels—managerial discretion, board diversity, and corporate transparency—link individualism to E/S performance. We find a positive association between firm-level E/S performance and firm value, with three firm-level channels—cash flows, cash flow variability, and cost of equity—linking E/S performance to firm value. This positive association is stronger in more individualistic countries. Finally, we find that internationalization weakens the role of national culture; however, it accentuates the positive association between firm-level E/S performance and firm value.
Abstract: Job-market mobility prediction plays a crucial role in optimizing human capital usage for both employees and employers. Most conventional methods primarily focus on learning sequential career sequences while ignoring the sufficient information extraction of mutual entity correlations in the job market. In this work, we push forward to exploit the heterogeneous relational knowledge among the job market structures by proposing a model namely Attentive Heterogeneous Knowledge Learning and Synergy (AHKLS). Equipped with the subsequent module of time-aware perception, AHKLS achieves effective career trajectory encoding for job-market mobility prediction. To evaluate the AHKLS performance, we conduct extensive experiments on three real-world datasets with different sizes. The empirical analyses demonstrate not only the performance superiority of AHKLS over several competing methods, but also the module effectiveness and model compatibility with other methods in enhancing the mobility prediction tasks accordingly.