Does impairment of political speech affect financial speech? We exploit the introduction of the National Security Law (NSL) in Hong Kong in June 2020 to answer this question. We find that after the NSL enactment, local analysts self-censor their reports, compared to foreign analysts covering the same firms. Specifically, when firm-specific bad news hits, local analysts shade up their forecasts, use vaguer language, and respond more slowly to earnings announcements. This pattern is especially true for central state-owned enterprises as negative opinions on their poor performance may be deemed unpatriotic. Markets are aware of this self-censorship and respond accordingly.

-Presentation: WFA 2026 (scheduled), ABFER 2025, ABFC 2025, University of Sydney 2025, UTS 2025, Curtin University 2025

-By co-author (selected): University of Chicago 2025, University of Washington, National Taiwan University 2025, National Chengchi University 2025, University of Tokyo 2025, Waseda University 2025, Hitotsubashi University 2025, Seoul National University 2025, KAIST 2025

Access to Credit and Medical Decisions: Can Finance Save Lives?

The poor often delay seeking medical treatment, even when it is free, to avoid losing income due to missed workdays. I use a regression discontinuity design to show that access to credit can remedy this issue. I exploit a setting in Korea in which individuals below a certain income threshold are eligible for cheap credit. My main finding is that such individuals have a 46% lower probability of dying or disability than otherwise similar individuals who are ineligible for cheap credit. This reduction is partly due to the fact that the eligible individuals see a doctor more promptly than the ineligible individuals after the onset of disease symptoms, as reflected in a lower likelihood of emergency room visits. Further, these individuals start treatment on average 62 days earlier following their initial diagnosis. Another major determinant of the decline in mortality and disability is that eligible individuals with chronic diseases are more likely to visit clinics for regular treatment. I explicitly show that such differences in the timeliness of treatment are due to individuals with access to credit being 44% more likely to take time off from work to seek medical care. Overall, access to credit can make a life-or-death difference to the poor even when medical treatment is affordable.

-Presentation: UNSW 2026, FIRN 2024

The Carbon-Capital Trade-off in the Age of AI: Evidence from Data Centers with Zonghe Guo, Yeejin Jang, and Sheng Xu 

This paper examines how data center investments-critical AI infrastructure-affect firms' environmental footprints and capital allocation, amid surging energy demands and emissions. Using staggered datacenter activations from 2005 to 2025 via event study and difference-indifferences design, we find datacenter operations sharply raise Scope 2 location-based emissions intensity with negligible Scope 1 effects, alongside surging mutual fund holdings driven primarily by non-ESG funds. Data center growth fueled by AI risks locking in fossil fuel power and diluting emissions targets as utilities scale back commitments amid surging demand. Overall, our results highlight a central trade-off in the AI-driven economy: digital infrastructure attracts capital but increases environmental costs, with important implications for sustainable finance.

Access to Credit and Medical Decisions: Can Finance Save Lives?