research
Here are some of my papers and working projects
2026
- When Public Demand Becomes Bad News: Government Procurement and Stock Market ReactionsWorking Paper, 2026
This paper examines how Chinese equity markets react to government procurement contracts. We find that procurement announcements are associated with significantly negative abnormal returns: the Fama-French three-factor CAR is-0.258% over [-2, +2] trading days and-0.729% over [-5, +5]. The negative response is concentrated among high-Tobin’s-Q firms, while low-Q firms earn positive abnormal returns. Non-state-owned firms react more negatively than state-owned firms. Negative returns appear both before and after announcements and around non-overlapping signing windows, consistent with gradual repricing rather than a single day-0 shock. High-Q firms’ cash holdings fall and long-term debt rises after announcements, consistent with contract execution absorbing cash and requiring financing. Overall, these findings suggest that an apparently favorable government contract may be bad news for suppliers whose valuations embed high growth expectations.
@article{GOVPROCUREMENT, title = {When Public Demand Becomes Bad News: Government Procurement and Stock Market Reactions}, author = {He, Sherry Xinyao and Hsieh, Jarvinia Chinhui and Luo, Shouyi and Wu, Jingwei}, journal = {Working Paper}, doi = {10.2139/ssrn.7573758}, year = {2026} } - Operational Hedging Under Lockdown: Cross-Border Acquisitions and Working-Capital Frictions in ChinaWorking Paper, 2026
This paper studies whether localized operating disruptions change firms’ boundary choices. China’s zero-COVID policy created substantial geographic variation in restrictions on production, mobility, and logistics. Combining province-level policy stringency from the Oxford COVID-19 Government Response Tracker (OxCGRT) with CSMAR MA and financial-statement data for Chinese listed firms, we employ a staggered difference-indifferences design that compares firms in high-stringency provinces with those in less restricted provinces. More stringent restrictions increase cross-border MA activity by approximately 0.56 to 0.66 percentage points, a sizable effect relative to the low unconditional acquisition rate. Event-study estimates show no differential pre-trends and a positive acquisition response following treatment. The effect is strongest among manufacturing acquirers and is not accompanied by broad increases in capital expenditure, RD, or cash holdings. Instead, treated firms exhibit lower incremental operating payables, larger reductions in operating receivables, and higher interest expense. Taken together, the evidence is consistent with firms using cross-border acquisitions as an operational hedge when local lockdowns tighten trade credit and increase the costs of geographically concentrated operations.
@article{LOCKDOWN, title = {Operational Hedging Under Lockdown: Cross-Border Acquisitions and Working-Capital Frictions in China}, author = {He, Sherry Xinyao and Hu, Wenbin}, journal = {Working Paper}, doi = {10.2139/ssrn.7527438}, year = {2026} } - RewardLens: Same Static Accuracy, Different Intervention Behavior in Multimodal JudgesWorking Paper, 2026
Can similar static preference scores conceal different responses to changing visual evidence? RewardLens compares multimodal judges matched by independent static accuracy on a separate controlled CLEVR audit. Questions and candidate contents stay fixed; relevant edits reverse the programmed gold preference, whereas irrelevant edits preserve it. Relevant Adaptation (RA) and Irrelevant Invariance (II) measure edited-image correctness conditional on a correct base decision. Across eight judges, a same-domain Count comparison provides a concrete counterexample: Qwen and Skywork score 100.0% and 99.5% on independent CLEVR static items, but their RA is 34/200 versus 125/199. On a separate 20-triplet order subset, both are base-correct in both orientations; the contrast persists after exchanging A/B contents (3/20 versus 13/20). The complete one-point same-domain census is heterogeneous: among 42 matched pairs, 16 have RA gaps of at least 10 percentage points and 16 have zero gap, with a median of 2 points. A post-freeze two-person audit of 320 triplets finds stronger human/gold agreement for Count and Attribute than Presence and Spatial. A scene-level recheck reproduces all 960 sampled gold labels, but human disagreement is strongly associated with candidate order in the latter factors; its cause remains unresolved. RewardLens thus exposes specific behavioral differences missed by measured static scores, not universal separation, population equivalence, or a position-independent account of visual grounding.
@article{REWARDLENS, title = {RewardLens: Same Static Accuracy, Different Intervention Behavior in Multimodal Judges}, author = {He, Sherry Xinyao and Daoson, Benjamin}, journal = {Working Paper}, year = {2026} } - Text-Based Interpretation of Innovation: The Causes and ConsequencesWorking Project, 2026
@article{NOVA, title = {Text-Based Interpretation of Innovation: The Causes and Consequences}, author = {He, Sherry Xinyao and Zhang, Bohui and Zhao, Xiaofeng}, journal = {Working Project}, year = {2026} } - Money or PowerWorking Paper, 2026
@article{MP, title = {Money or Power}, author = {He, Sherry Xinyao and Wang, Qinyun and Wang, Xuebin and You, Yang}, journal = {Working Paper}, year = {2026} }
2025
- Uncovering Missing R&D: Machine Learning Insights into Innovation and Q-TheoryWorking Paper, 2025
A substantial fraction of firms do not report R&D expenditures, posing significant challenges for accurately measuring innovation capital. To address this issue, we employ various machine learning techniques to estimate missing R&D investments. Using a novel out-of-sample validation—firms that report R&D for the first time—we observe that neural networks provide the most accurate estimates of missing R&D. These ML-estimates reliably predict both the quantity and value of future patents for non-R&D-reporting firms. Moreover, we find that the Q-theory holds for uncovered missing R&D investments, thereby resolving a significant portion of the empirical gap in the investment-Q relationship caused by missing data. Our findings carry important implications: over time, the magnitude and economic impact of estimated missing R&D may rival or even surpass those of reported R&D itself.
@article{MLRD, title = {Uncovering Missing R&D: Machine Learning Insights into Innovation and Q-Theory}, author = {He, Sherry Xinyao and Kang, Jun-Koo and Zhang, Hong}, journal = {Working Paper}, year = {2025} }
2024
- Ripple Through Garbage: The Impact of China’s Solid Waste Import Ban on U.S. FirmsWorking Paper, 2024
15th Financial Markets and Corporate Governance Conference
The global supply chain encompasses not only good production but also the byproduct of waste. Do firms exploit this property to benefit domestic production? Utilizing China’s Solid Waste Import Ban (CSWIB) as a regulatory shock, we observe that the Producer Price Index (PPI) for U.S. waste-generating industries rose by 2.94% due to increased production costs post-CSWIB. Furthermore, firms within waste-generating sectors also exhibit a 6.49% drop in Tobin’s Q. Our findings suggest that global waste management significantly affects inflation and firm valuation in the U.S., which carry important normative implications for global green transitions.
@article{CSWIB, title = {Ripple Through Garbage: The Impact of China's Solid Waste Import Ban on U.S. Firms}, author = {Chen, Tao and He, Sherry Xinyao and Sun, Stephen Teng and Zhang, Hong}, journal = {Working Paper}, year = {2024}, doi = {10.2139/ssrn.5494686}, }