【题  目】Common AI Information and Capital Allocation

【时  间】2026年9月30日（周三） 14:00-15:30

【地  点】后主楼1610会议室

【主讲人】孙浩宁 博士研究生（清华大学经济管理学院）

【主持人】许敏波 教授（北京师范大学经济与工商管理学院）

摘要：This paper studies how common reliance on AI-generated information affects institutional portfolios and real capital allocation. We construct a revealed measure of managers' alignment with AI-extractable information and find that manager pairs with greater common reliance subsequently hold more similar portfolios, particularly after ChatGPT's public release and when language models have a clearer information advantage. In a portfolio model, investors choose between private research and an AI signal with a shared model error. Common AI exposure generates portfolio comovement and survives aggregation across investors. We then study a production economy in which financial prices aggregate investors' information and firms learn from those prices before investing. Managers may underestimate the correlation between their AI signal and the information in prices. Under complete correlation neglect, lower AI costs raise aggregate TFP locally when adoption is limited, while improvements in AI accuracy can initially reduce TFP by inducing greater exposure to the common error. The two local results extend beyond the power-cost specification under stated conditions on information acquisition costs.

报告人简介：

孙浩宁，清华大学经济管理学院经济系博士研究生，导师为董丰教授，本科毕业于清华大学经济管理学院，主要研究方向为宏观金融、数字经济与人工智能的宏观影响等，主持国家自然科学基金青年学生基础研究项目（博士研究生），研究成果发表在International Economic Review、Review of Economic Dynamics、《管理世界》、《管理科学学报》、《经济学季刊》等国内外期刊。