ABOUT ME
My name is Fanwen Zhu. I am a Ph.D. job‑market candidate in Economics at UCLA working at the intersection of entrepreneurship, trade, and economics of education. My research combines empirical causal inference with structural modeling, drawing on evidence on both public innovation dataset and a series of field experiments in education.
Across projects I ask three connected questions: who becomes an entrepreneur when large geo‑economic shocks take place; how founder types and subsequent endogenous organizational behaviors shape the formation of dynamic capabilities and influence firms’ later development; and how K-12 schools can optimally roll out GenAI tools through a well-designed transformation path.
Research
Working papers
1. Sanctions and Startups: Trade Shocks and Inventor Entrepreneurship in the U.S.-China Trade War (Job Market Paper)
Transformative entrepreneurs, in contrast to subsistence ones, are a key source of creative destruction and long-run economic growth. However, we still know little empirically about how macroeconomic shocks shape inventors’ entrepreneurial choices as potential transformative founders and their early-stage innovation strategies. In my JMP, I exploit cross-industry exposure to the 2018 U.S.–China trade war, together with a newly constructed, million-scale inventor–firm matched dataset, to study the entrepreneurial consequences of tariffs (a market-size shock) and the Entity List (a negative supply shock). The empirical analysis shows that tariff changes have limited effects, whereas the Entity List induces pronounced two-sided selection in which types of inventors continue to found startups, and leads to substantial reductions in post-entry innovation outcomes. The paper also highlights a new channel through which trade shocks affect innovation at the individual level, rather than only at the firm level.
2. Born to Invent, Bound to Succeed? Inventor Founders and Firm Success
Do inventor-founded enterprises behave differently from their counterparts? I investigate the large-scale inventor and firm registry database from 2000 in China and develop a Schumpeterian innovation model that explains the difference in their innovation trajectories and provide policy suggestions.
Work in progress
3. Adopting LLMs at Work: A Dynamic Model with Field Experiment
What is the optimal way for organizations to roll out worker–GenAI collaboration? Through field experiments, we reveal a sigmoid adoption path with an early dip and a persistent group of non-adopters. Guided by survey evidence, we build a learning model with learning costs and psychological anchors that generate these patterns and offer intervention strategies for smoother organizational rollout.
3. Adopting LLMs at Work: A Dynamic Model with Field Experiment
What is the optimal way for organizations to roll out worker–GenAI collaboration? Through field experiments, we reveal a sigmoid adoption path with an early dip and a persistent group of non-adopters. Guided by survey evidence, we build a learning model with learning costs and psychological anchors that generate these patterns and offer intervention strategies for smoother organizational rollout.
3. Adopting LLMs at Work: A Dynamic Model with Field Experiment
What is the optimal way for organizations to roll out worker–GenAI collaboration? Through field experiments, we reveal a sigmoid adoption path with an early dip and a persistent group of non-adopters. Guided by survey evidence, we build a learning model with learning costs and psychological anchors that generate these patterns and offer intervention strategies for smoother organizational rollout.
We conduct a survey covering 4,927 teachers in 339 Chinese K–12 schools on GenAI usage. Regular use clusters around 30% and 50% across schools and we found that adoption is weakly related to use in nearby schools but strongly associated with same-school peer use. We propose a model explaining this pattern through shared, nonrival knowledge which creates reinforcing low- and high-use equilibria. Counterfactual analysis suggests that recurring support and shared-knowledge infrastructure can expand active-user cohorts, strengthen knowledge spillovers, and turn dispersed experimentation into an organizational resource that unlocks broader productivity gains across the organization.
1. From Substitution to Complementarity: The Spillover Structure of Generative AI Adoption in Collaborative Teams
Motivated by our field experiment, we construct a game model showing that peer efforts in GenAI usage become complementary when shared information continually expands the solution frontier, but substitution and free-riding emerge when marginal returns diminish rapidly. High-ability leaders can increase teammates’ returns to adoption, supporting targeted usage-cost subsidies. Effective AI diffusion therefore requires workflows that make one member’s output another’s input rather than a direct substitute.