德勤-复旦会计论坛系列讲座之一百七十九

时   间:2021年1月5日(周二)下午14:00

地   点:史带楼303室

主持人:张新 副教授

主讲人Sean S. Cao (Georgia State University)

主    题:Analyst Skill and Crowd Wisdom: Information Aggregation from Machine Learning

摘    要:In capital markets, it is important to effectively aggregate information from heterogeneous agents to form crowd wisdom for investment decisions. Such aggregation is challenging and complex because it involves each agent’s information and expertise, firm and macro-level factors, and their high-dimensional interactions. In this study, we use machine learning (ML) to provide a novel methodology to identify analyst skill, and effectively aggregate forecasting opinions of analysts to form a crowd wisdom-based earnings forecast for each stock and at each given quarter. We find that the ML-based forecast consensus contains more valuable information than the mean consensus used in the literature when predicting subsequent earnings. Further, the ML consensus can predict 3-day abnormal returns around the earnings announcement date, suggesting that our machine learning aggregation model is not yet incorporated by the market immediately. The predictive power of ML is stronger when stocks are covered by more analysts, when analyst forecast errors are greater, and when the market is more volatile or in a crisis. Overall, our ML-based consensus provides two implications for scholars and investors: 1) a new earnings expectation measure that better measures earnings surprise. The difference between ML consensus and mean consensus captures what AI knows about earning news while the market does not know until right before earning announcement, and 2) it delivers profitable returns for investors. The method of information aggregation we develop can also be applied to other settings such as online forums, political opinions, and macroeconomic outlooks.

 

会计学系

2020-12-24