主讲人：
刘行（清华大学五道口金融学院助理教授）
主持人：
（北大经院）贾若
（人大财金）陈泽
（清华经管）冯润桓
参与老师：
（北大经院）郑伟
（人大财金）魏丽
时间：
2024年5月14日（周二）
10：00-11：30
线上形式：
腾讯会议
会议号：311 919 936
线下地点：
北京大学经济学院302会议室
主讲人简介：
刘行，清华大学五道口金融学院助理教授。他的研究方向包括公司金融、金融科技和保险市场。他的研究关注无形资产的经济价值，如公司文化、工作的非物质激励维度、知识产权、以及新技术对信息生产和代理冲突的影响。他的研究成果目前发表于
Journal of Financial and Quantitative Analysis
和
Management Science
。刘行博士于2023年获得加拿大英属哥伦比亚大学金融学博士学位。此前他在清华大学获得金融专业硕士学位，在中国人民大学财政金融学院保险系获得经济学学士学位。
摘要：
Selection markets create a multitasking environment where intermediary agents often need to increase consumer take-up as well as resolve information asymmetries about consumer expected cost during the sales process. I study how artificial intelligence (AI) affects attention allocation and information production in human-intermediated markets by analyzing a large-scale randomized experiment conducted by a top insurance agency in China. In the experiment, the firm provided treated agents with an AI-generated estimation of consumer demand for insurance, based on consumer digital footprints on the advertisements on social media; these footprints were available to all agents prior to the experiment. I show that AI demand prediction shifts agents' attention to converting high-intent consumers, improving agents' sales by 14%. As an unintended consequence, AI-generated demand information reduces agents' own information acquisition and increases adverse selection, consistent with attention models and a crowding out of risk information. Moreover, treated agents bring in riskier consumers but do not match them to more expensive products to achieve stronger incentive compatibility. The findings suggest that a common application of AI to predict consumer demand can have side effects on human information production, market efficiency, and can exacerbate agency conflicts when intermediary agents capitalize on AI.
