Research

 

Working Papers

1. Fintech Access and Consumption Smoothing (PDF)

(with Long Chen, Xavier Giroud, and Neng Wang; presented at NASM2024.)

What are the real effects of an innovative fintech credit service used by 500 million consumers?

We identify and quantify the effect of access to Fintech consumer credit service on consumption using granular data from a nearly one billion consumer population on China’s leading digital payment platform. We find that, despite sharp declines in consumption following the pandemic shock, consumers who used Fintech credit maintained a 28.7% higher monthly consumption in the six months afterward, indicating a consumption-smoothing effect. The effect persists at 22.0% on average over the two-year period. We explain the effect with two channels: the income channel, where the consumption-smoothing effect is stronger among low-income consumers and those from less developed regions, and the financial literacy channel, with a more significant effect for consumers with lower financial literacy. Fintech credit users do not show high delinquency rates but invest less in their wealth management account. Additionally, we find that the Fintech credit usage helps explain city-level variations in retail sales.

 

2. Regulating Information and Competition: Evidence from Fintech SME Loans (PDF)

(with André Sztutman and Robert M. Townsend)

How do market power and information availability affect the efficiency and welfare in credit markets?

Leveraging data from a major lending provider for SMEs in China and an interest rate discount policy, we analyze the presence of selection in lending markets and how it interacts with market power. Our findings reveal a decrease in the average probability of loan repayment following interest rate reductions, indicating advantageous selection. Alternative explanations, such as moral hazard, observable heterogeneity, and dynamic portfolio optimization, cannot fully account for that pattern. Building on our estimated demand elasticities and selection effects, we assess the welfare implications of information-sharing and pro-competition policies, finding that among a restricted simple set of policies, ensuring competition and mandating information to be shared across financial providers results in the largest welfare gains. 

 

3. Monopoly and Competition of Foundation Models (PDF)

 

Does competition in foundation models such as LLMs lead to greater informativeness and market efficiency? (Very likely, no.)

I consider the generation and provision of information products, such as generative artificial intelligence models, in the information markets. Sellers of information must make an investment to deliver quality experiments. The level of investment determines the informativeness of the best experiment a seller can provide. Heterogeneous buyers face a decision problem with the uncertainty of the true state and can purchase experiments to augment their private information. Sellers design a menu of experiments and prices for the market. I characterize the optimal menu given any investment level and derive the optimal investment. When two sellers compete with investment, we study an equilibrium in which two sellers split the market. Each seller specializes in generating a more informative signal about one of the states. Under a general assumption of cost structures, the monopoly seller always provides more informative experiments and to more buyers than in the case of duopoly competition.

 

4. Artificial Intelligence and Platform Credit Risk

(with Long Chen, Jon Frost, and Yi Huang, draft available upon request)

How does big tech manage credit risk? Do the algorithms show biases? How does big tech lending rely on its ecosystem?

Big tech companies are becoming increasingly active in finance around the world, including in credit markets. Their platform-based business model, information advantage, and the use of real-time big data and machine learning can benefit them in real-time credit risk assessment, contract enforcement and dynamic risk management. We show that the big tech lender actively manages credit risk with real-time adjustments in credit admission, line of credit, and interest rates. The big tech lender effectively maintains a low delinquency rate, even during shock. We explain the low risk with a mechanism of ecosystem dependency. Since borrowers’ business operations rely on the digital platform ecosystem, the implied punishment of platform exclusion is a powerful incentive for borrowers to repay the loans. We further investigate whether the adjustment algorithm exhibits algorithmic discrimination against specific groups.

 

 

Selected Research in Progress

 

5. Personalized Recommendation and Consumer Experiences

(with Alessandro Acquisti,)

After opting out of personalized recommendations on e-commerce platforms, do consumers buy less?

We study the impact of consumers opting out of personalized recommendations in a leading e-commerce platform. Young, educated males living in big cities are most likely to opt out, primarily due to dissatisfaction with recommendations. Opting out reduces browsing diversity and overall platform engagement.  However, consumers pay more attention to the items they click on. We find that opt-out does not affect total spending or satisfaction, as return rates remain stable. Consumers reallocate their purchases from recommendations to search.

6. Optimal Risk Sharing of SMEs as Households

(with Robert M. Townsend)

 

7. Estimating the Value of Data 

(with Daron Acemoglu, Nikhil Agarwal, and Tobias Salz)

 

8. Excessive Entry and Social Inefficiency in Information Market

How does making information free to sellers affect market efficiency and consumer welfare?