Investments and Innovation with Non-Rival Inputs: Evidence from Chinese Artificial Intelligence Startups (with Kang Zhou), Review of Economics and Statistics, forthcoming.
Stepping Down from Life-long Posts: Layoffs, Fertility, and Educational Attainment in Urban China (with Junsen Zhang and Kang Zhou), Journal of Labor Economics, forthcoming.
Information Disclosure and Bidding Structure: Evidence from the London Bus Market (with Mike Waterson), Quarterly Review of Economics and Finance, 2026.
The Cultural Origins of Family Firms (with Song Yuan), Journal of Comparative Economics, 2025.
Dying to Survive: Unintended Consequences of Environmental Regulations on Industrial Accidents (with Xiaowei Chen, Qiyue Shen, Xuebo Wang, and Zhilong Zhang)
Abstract: We investigate the impact of environmental regulations on industrial accidents by exploiting the 2014 revision of China’s Environmental Protection Law (EPL), heralded as “the strictest environmental law in China’s history.” These tightened regulations may have significantly financially constrained pollution-intensive firms, forcing them to reduce precautionary spending on production safety in order to survive, thereby potentially increasing the risk of industrial accidents. With a difference-in-differences strategy, we compare changes in industrial accidents between pollution-intensive and non-pollution-intensive industries at the prefecture level before and after the implementation of the revised EPL. Our estimates indicate that stricter environmental regulations led to an approximately 60.4% increase in the number of industrial accidents in heavily polluting industries. Furthermore, we demonstrate that the reduction in firms’ safety investments, caused by tighter financial constraints following the stricter regulations, is a plausible underlying mechanism. Our study highlights an additional hidden cost of environmental regulations that has been overlooked in the literature, suggesting that traditional estimates of the costs of environmental regulations may have significantly underestimated their broader societal impact.
Surname-based Connections, Informal Institutions, and Local Innovation (with Po-Hsuan Hsu, Bernard Yeung, and Qinhong Yu)
Abstract: Surname-based ties facilitate local social networks and collaborations, particularly where formal institutions are weak. We hypothesize that local innovation, measured by local firms' patent activity, increases with surname-based connections among business owners, as innovation requires teamwork and the combination of complementary resources that such ties help coordinate. Using a county-year panel from China, we find that counties' businesses with stronger surname-based connections generate significantly stronger patent performance. An instrumental-variable strategy based on historical peasant rebellions in imperial China, which increased the importance of surname-based collaborations, supports a causal interpretation. More importantly, patenting rises disproportionately among firms controlled by business owners sharing more common surnames, allowing us to attribute local increases in innovation directly to surname-based ties. Consistent with surname ties substituting for formal institutions, the connection-innovation relationship weakens with better access to finance, improved public services, and stronger supply-chain cooperation, and declines over time as China's formal institutions strengthen. Additional tests show that surname-based connections are also associated with greater local entrepreneurship and economic development, highlighting how social networks shape the landscape of economic growth.
The Digital Second Shift: Gender Gap in Parenting App Usage in China (with Huan Cai and Lu Dong)
Abstract: This paper examines gender disparities in parenting in the digital domain, using a novel dataset that records the gender composition of users across more than 6,000 app-level observations in China. Two patterns stand out. First, parenting apps are strongly feminized: women account for nearly two-thirds of users, compared to fewer than half for the typical non-parenting app. Second, the female share is highest in cities where women enjoy greater income and educational attainment, and lowest in areas marked by more entrenched gender inequality. The women most engaged in digital caregiving are therefore those best positioned to transcend traditional roles. Mechanism analysis suggests that this is not driven by broader digital fluency among affluent women, but rather reflects their intentional choice for intensive parenting practices.
Heritage of Hostility: How Anti-Missionary Violence and Industrial Capacity Shaped China’s Quid Pro Quo for Foreign Technology (with Renliang Liu)
Abstract: Why do favorable policies sometimes fail to spur foreign technology adoption? This paper examines how anti-foreign sentiment, proxied by the incidence of anti-missionary violence, and industrial capacity jointly shaped the gains from China's 1983 Quid Pro Quo policy (QPQ, trading market access for technology). Our difference-in-differences analysis shows that the QPQ policy substantially increased foreign technology adoption in cities with more developed early industrial capacity. A subsequent triple-differences specification shows that anti-missionary violence erased around three-quarters of these gains, with the most pronounced effects on critical equipment, licensing agreements, and in cities where officials led the conflicts. Mechanism analysis suggests that anti-missionary violence contributed to this erosion by deteriorating bilateral municipal ties and deterring the entry of foreign-invested firms. Firm-level matching indicates a 15.8% productivity premium for technology adopters, implying that anti-foreign sentiment reduced average city-level productivity by 0.128% over 1983-1995.
AI in the Court (with Natarajan Balasubramanian, Chenyang Pan, and Wenjian Xu)
Abstract: This paper examines how the adoption of artificial intelligence (AI) technologies influences judicial performance in China. We use large language models to identify AI procurement in comprehensive court purchasing records and construct measures of court performance from more than 146 million judicial decisions. Employing a staggered difference-indifferences framework, we find that AI adoption reduces appeal rates by roughly 15% and reversal rates by 21%, while substantially speeding up case processing. These performance gains are especially pronounced for more complex cases and in less developed regions, suggesting that AI helps alleviate capacity and information constraints in resourcelimited judicial settings. We further show that AI adoption generates significantly larger improvements compared with other digital technologies, highlighting its unique contribution to judicial decision-making. Taken together, the results temper concerns about AI use in the public sector and demonstrate its potential to strengthen institutional performance.