嘉宾简介：修大成，芝加哥大学布斯商学院计量与统计教授。中国科学技术大学数学,理工学学士，美国普林斯顿大学应用数学硕士、博士。主要研究领域为Financial Econometrics, Empirical Asset Pricing, Machine Learning in Finance, High-Dimensional Statistics, Quantitative Finance等。研究成果发表于Journal of Econometrics，Review of Financial Studies，Journal of Finance，Annals of Statistics，Journal of Business & Economic Statistics等国际知名期刊。
内容摘要：Because machine learning can handle a large number of predictive variables and has a variety of functional forms, the application of machine learning methods in the financial field is always a concerned issue in the cademia and industry.
This paper applies a variety of representative machine learning methods to solve the most studied and classic problem in the field of empirical asset pricing: measuring the risk premium of assets. This paper focuses on comparing the different methods. It is found that using machine learning to predict can bring huge economic benefits to investors, which is better than the long-term regression analysis strategy in the literature. Among them, classification tree and neural network are the two learning methods with the best performance. Compared with other methods, they take into account the nonlinear relationship and interaction between variables and effectively improve the prediction accuracy.