Research

Working Papers

Memory and Generative AI (solo)

Large language models rely on associative memory to make decisions, even if the memories are entirely unrelated to the decision domain.
--- Presentations at Harvard, SAIF, ABFER 2025 Poster session, CICF2025, CFRC 2025, scheduled presentations at AEA2026, Fudan SOM, AFA 2026 Poster session

AI as Decision-Maker: Ethics and Risk Preferences of LLMs (with Shumiao Ouyang and Hayong Yun)

Large language models do exhibit consistent risk preferences and ethical fine-tuning has unintended consequences on their risk preferences.
--- Presentations at ABFER-JFDS 2024, OxNLP, GPI, Oxford Finance, SBS Board, SAIF*（×2）, PKU NSD, ZJU Econ, SFS Cavalcade 2024, CREDIT 2024, Adam Smith Junior 2024, and CBF 2024, 4th Hongkong AI Conference 2025*, ES World Congress 2025, scheduled presentations at NAWM Econometric Society 2026, Paris December Meeting 2025, GSU AI conference 2025 (* denotes presentation by Xingjian)

Carbon emission and asset prices: new evidence from machine learning (with Feng Li)

We predict the carbon emissions of US-listed firms with XGBoost and find a reversed carbon premium after 2016. (This is my second-year summer paper.)
--- Presentations at CICF2023, CFRI-CIRF 2023 joint conference, CMCSR2023, SBSICF2023

Clustered by images: Convolutional neural networks, investor heterogeneity, and Chinese stock market predictability (《图以类聚：卷积神经网络，投资者异质性与中国股市可预测性》 in Chinese, with Yuqiao Fang and Feng Li)

We propose a simple but effective method to improve the predictability power of CNN in the Chinese Stock market.
--- Presentations at CFAC2024. Accpected at Journal of Management Sciences in China （《管理科学学报》）

Work in Progress

Improving Investor Trading Habit via Digital Companion (with Xiaomeng Lu, Jun Qian and Shang-Jin Wei)

We conducted experiments and found that digital nudging is effective at changing investors' behavior.
--- Partnered with the Ant Group.