November 24 Finance Seminar, No. 344
Time: 14:00-15:30, Tuesday, November 24, 2026
Location: Room 603, Starr Building(史带楼603室)
Host: Prof. Kemin Wang (王克敏 教授)
Department of Finance, FDSM
Topic: More Than a Number: The Value of Human Reasoning as AI Input
Speaker: Dr. Zhenghui Ni (倪政辉 助理教授)
Renmin University of China
Abstract:
Do sell-side analysts’ explanations improve large language model (LLM) earnings forecasts beyond their numerical estimates? We progressively augment GPT-5-Chat with analyst EPS forecasts and structured reasoning extracted from analyst reports. The standalone model considers more information-cue categories, whereas analysts offer more diverse interpretations within categories. Adding report reasoning reduces mean price-scaled absolute forecast error by 10.5% beyond analyst estimates alone. The fully augmented model outperforms analyst consensus and reduces error by 21.2% relative to the standalone model. These findings show that analysts’ articulated reasoning contains predictive information beyond their point forecasts.
Bio:
倪政辉是中国人民大学商学院助理教授,2024 年在新加坡国立大学获得博士学位。他的研究领域聚焦人工智能、信息与资本市场,尤其关注生成式人工智能如何影响信息处理、企业信息披露和投资者行为。他目前的研究课题包括:人工智能信息处理与自愿性信息披露、人工智能应用中断的经济后果,以及人类推理在大语言模型应用中的作用。他的研究成果发表于Journal of Accounting and Economics和Journal of Financial and Quantitative Analysis等期刊。他主持国家自然科学基金青年科学基金项目。
复旦大学管理学院金融与财务学系