Sixuan Li

We conduct a field experiment in China, offering startups and SMEs conditional social media subsidies to stimulate business growth. The intervention increases revenue, employment, and access to fintech credit: treated firms are more likely to open online stores and obtain online loans, while traditional bank credit remains unchanged. Our results uncover a "fintech accelerator" mechanism: digital investment boosts both sales and digital footprints, such as customer ratings, which are crucial for fintech lending eligibility. These findings highlight the role of targeted digital interventions in promoting financial inclusion and reshaping credit allocation for startups and small businesses.

This paper investigates how integrating large language models (LLMs), such as ChatGPT or DeepSeek, into robo-advisory platforms can improve investor decision-making, using financial literacy as the main setting. In a pilot program with a major brokerage firm, the study enhances a robo-advisor with a back-end LLM that provides personalized and conversational support to investors, helping them distinguish between beta and alpha in mutual funds and stocks. Compared with a traditional rule-based chatbot and standard non-premium human assistance, the LLM-augmented robo-advisor significantly increases investor engagement and understanding, narrows the behavioral gap between one-click automatic enrollment and self-constructed portfolios, and builds relational trust similar to high-touch wealth management services. The LLM support also encourages ETF adoption, especially in trust-sensitive markets, and generates positive spillovers by making investors more likely to pursue international diversification even when such options are not actively promoted. Overall, integrating LLMs into robo-advisors produces substantial benefits, with treated investors earning monthly returns 40 basis points higher and realizing an average monthly gain of 750 RMB.

This paper conducts a randomized controlled trial to examine how contract education and contract legal consulting affect entrepreneurs’ outcomes in venture capital and seed funding. Entrepreneurs in the treatment group received targeted training on VC contract terms and startup financing agreements. The results show that treated entrepreneurs are less likely to obtain initial angel or seed funding and are more likely to enter traditional employment rather than continue their startup ventures. However, conditional on receiving angel or VC investment, treated entrepreneurs are more likely to secure follow-on funding, achieve higher later-stage valuations, and face a lower risk of involuntary removal as co-founders. Overall, the findings suggest that greater contract knowledge helps entrepreneurs make more informed career choices and negotiate better terms, but may also reduce their entry into the VC funding pipeline.

This paper examines what determines firms’ decisions to exit Russia after the Ukraine invasion and whether investors value such social image management. It finds that firms withdrawing from Russia generally have higher ESG scores, especially social scores, but firms with higher ESG scores tend to announce their withdrawal later and choose softer forms of disengagement. Exit decisions are also more likely among firms headquartered in countries with greater security concerns, stronger public awareness, better institutional quality, common-law legal origins, or sanctions against Russia. However, these firms experience a 58-basis-point stock price decline after announcing business suspension, with no evidence of a stock market “reputation premium” relative to firms that remain. Overall, the evidence supports the reputation-concern channel rather than purely altruistic motives.

This paper examines whether large language models, such as ChatGPT, can use their reasoning capabilities to generate portfolio recommendations beyond conventional textual analysis. By fine-tuning the model with task-specific training data and adjusting ChatGPT’s parameters to allow more flexible outputs, the paper shows that ChatGPT can construct portfolios that outperform market benchmarks in out-of-sample tests. Using two types of multilingual textual data—Wall Street Journal articles from the U.S. and Chinese government policy announcements—the study finds that ChatGPT-generated portfolios can achieve monthly three-factor alphas of up to 3%, especially in response to policy-related news. In contrast, traditional textual analysis methods fail to generate portfolios with positive alpha.

Approximately 37% of Chinese listed firms possess medical expertise, as indicated by the presence of board members and senior executives with a medical degree or experience in the medical industry. Using the COVID-19 outbreak in China as a natural experiment, we find that the stock returns of firms with medical expertise, excluding those within the healthcare and pharmaceutical industries, are significantly higher than those of firms without such expertise. The positive impact is more pronounced when the CEO or Chairman has medical expertise and when the firm is not state-owned. Overall, this study underscores the importance of diversified executive human capital on firm performance by disentangling the effects of macroeconomic shocks.