with Zhiguo He and Jidong Zhou
Journal of Financial Economics
2021 European Finance Association Best Conference Paper Prize
Presentations: Hong Kong Shue Yan University*, Luohan Academy*, WUSTL Olin*, Norwegian School of Economics*, China Fintech Research Conference 2021, Annual International Industrial Organization Conference*, UChicago Booth Banking Workshop*, ES NA Summer Meeting 2021*, WFA 2021, Cambridge Centre for Alternative Finance Conference, China Financial Research Conference, CICF 2021, EFA 2021, Cambridge Corporate Finance Theory Symposium 2021, AFA 2022*, GSU Fintech Conference*, FTG Spring 2022 Meeting*, Workshop on Platforms and Data: Shaping the Future of Payments, NBER SI 2022 CF&RFI
with Kristian Blickle, Zhiguo He and Cecilia Parlatore
Forthcoming, Journal of Financial Economics
Presentations: Yale Junior Finance Conference 2022, Tsinghua PBC, Lone Star Finance Conference, WAPFIN (Stern)
with Shumiao Ouyang
Abstract: We model AI-based advising via large language models (LLMs) by introducing preference uncertainty---capturing soft information---alongside standard fundamental uncertainty, and validate model predictions through LLM-driven simulations. While human advisors can elicit soft information, their misaligned incentives lead to information loss (Crawford and Sobel, 1982). In contrast, LLMs are unbiased, but digitizing soft information is challenging: it is difficult to formalize and communication is constrained by the LLMs' limited memory. We model this communication as an optimal stopping problem with Brownian information flow and solve it in closed form. Simulations using multi-round LLM advising confirm our theoretical frictions and are benchmarked against standard portfolio questionnaires. Results show that deeper conversations mainly help investors clarify and learn their own objectives, improving decisions; however, impatience or early stopping reduces recommendation quality.
Previously circulated under the title "Specialized Lending when Big Data Hardens Soft Information"
with Zhiguo He and Cecilia Parlatore
Presentations: University of Washington (Foster)*, UBC Winter, Texas Finance Festival, BIS-CEPR-SCG-SFI Conference on Financial Intermediation (Gerzensee), FIRS, LSE & Bank of England Nonbank Conference, INSEAD Finance Symposium, Oxford FIT, NBER CF Fall 2024
Revise and Resubmit (2nd round), Review of Financial Studies
2022 Western Finance Association PBCSF Award for the Best Paper in FinTech
I study credit market outcomes with different competing lending technologies: A fintech lender that learns from data and is able to seize on-platform sales, and a banking sector that relies on physical collateral. Despite flexible information acquisition technology, the endogenous fintech learning is surprisingly coarse---only sets a single threshold to screen out low-quality borrowers. As the fintech lending technology improves, better enforcement harms, while better information technology benefits traditional banking sector profits. Big data technology enables the fintech to leverage data from its early-stage operations in unbanked markets to develop predictive models for expansion into wealthy markets.
Presentations: Workshop on Financial Intermediation and Regulation at Queens University 2022, WFA 2022, CICF 2022, Indiana Kelley, Toronto Junior Finance/Macro Conference, Berkeley Haas, FTG Spring 2023 (short presentation), FDIC, Korea University, FIRS 2023, Bank of Canada workshop on Payments and Securities Settlement, EFA 2023, INFORMS 2023
Abstract: I study information design on a financial network to maximize its stability, where banks' endogenous default outcomes are determined by a fixed point payment problem that accounts for both project qualities and interbank contagion. In addition to the cross-state risk sharing in previous work, the system-level design highlights the novel cross-bank risk sharing: a less discriminatory disclosure that reports the same signal on different banks reduces contagion, but may be costly due to banks' idiosyncratic shocks. The optimal disclosure is less discriminatory for high bank profitability or large counterparty exposure. The paper shows tractability in two cases: in the complete network, signals that are reported in the optimal policy must outperform a naive all-pass policy at one particular state; in general networks, when prior is sufficiently low, the optimal policy can be determined by ranking an efficiency index of alternatives.
Presentations: Duke Fuqua, UColorado Leeds, EIEF, Wisconsin School of Business at UW Madison, Oxford Financial Intermediation Theory Conference 2021, Boston Fed Conference on Stress Testing Research, Warren Center workshop on Networks in Finance, Princeton, SAET 2024
with Ling Ren
A rational miner may gain from withholding blocks and strategically timing the release to orphan the others' blocks, which questions the existence of honest majority and blockchain consensus. This paper reconciles the discrepancy between the absence of block-withholding attacks in practice and the profits as suggested by the literature. When the miner is impatient, selfish mining is a long-term deviation, and the short-term frictions incentivize honest mining. We present a continuous-time model that incorporates discounted cash flows, a new risk of being temporarily orphaned and endogenous difficulty adjustment. We show that, the gain from selfish mining relies on a quick difficulty adjustment, and the miner is most concerned with maintaining the new difficulty in long run. A small risk would increase the threshold computation power to attack by 36%. A protocol that includes the orphan blocks in difficulty recalculation defenses such attacks.
Presentations: ABFER 8th Annual Conference (Special Session on Blockchain & Cryptocurrency)
Abstract: This paper highlights the role of mutual fund families to explain the power-law tail distribution and the organizational structure of the asset management industry with a double-sided matching model. Specifically, there is a labor market where mutual fund families choose both the quality and the quantity of managers to hire, and an asset market where cash flows into fund portfolios. Fund families choose the investment intensity for new funds which cost managerial units, and design the riskiness of the alpha generating technology. We find the conditions under which better fund families hire both better managers and more of them, and launch more funds. We also find that better advisors introduce riskier portfolio products.
with David Xiaoyu Xu
Abstract: This paper highlights the role of mutual fund families to explain the power-law tail distribution and the organizational structure of the asset management industry with a double-sided matching model. Specifically, there is a labor market where mutual fund families choose both the quality and the quantity of managers to hire, and an asset market where cash flows into fund portfolios. Fund families choose the investment intensity for new funds which cost managerial units, and design the riskiness of the alpha generating technology. We find the conditions under which better fund families hire both better managers and more of them, and launch more funds. We also find that better advisors introduce riskier portfolio products.