Can Machines Understand Human Skills? Insights from Analyst Selection

Abstract

This paper presents a machine learning method to analyze analysts’ forecasting decisions and detect their firm-specific skills. Machine vs. human assessment of human skills differ in important dimensions: Machines rely on nonlinear functions of analyst characteristics, such as past accuracy and efforts, to identify analyst skill, while human experts lean on relation-based information such as brokerage size. Our model allows the formation of a “smart” consensus of the forecasts by machine-identified skilled analysts, which better proxies earning news before earnings announcements than the traditional analyst consensus. Investment strategies based on revisions of machine-identified skilled analysts generate significant abnormal returns and explain the market anomaly of post-analyst revision drifts. Our machine learning framework has the potential to be applied to other settings that involve human skills, such as the evaluation of job candidates and the compilation of political and macroeconomic forecasts.

Keywords: FinTech, Machine Learning, Artificial Intelligence, Analyst Forecast, Analyst Skill

JEL Classification: C45,D80,G11,G14,G23,M41