高管个人特征与公司业绩——基于机器学习的经验证据

Managerial individual characteristics and corporate performance:Evidence from a machine learning approach

在目前的公司治理文献中,大部分的高管特征研究一方面仅关注单一的高管特征与公司业绩之间的关联,缺乏全面的高管特征分析;另一方面主要围绕因果推断进行研究,缺乏从预测能力出发的系统定量的结论.本文首次采用机器学习算法中的Boosting回归树,全面考察了多维度高管特征对公司业绩的预测性.以我国2008年~2016年的上市公司为样本,研究了高管的多维个人特征是否能预测公司业绩,并进一步分析了对公司业绩预测能力较强的高管个人特征及其预测模式.研究发现:1)整体而言,在我国公司CEO和董事长的特征对公司业绩的预测能力较弱;2)在众多高管个人特征之中,高管持股比例和年龄对公司业绩的预测能力较强;3)高管持股比例和年龄与公司业绩之间的关联都呈现出非线性的特点,与以往的理论较为吻合.本研究不仅利用机器学习方法从一个更为全面的视角对中国的高管特征进行了研究,也为公司高管聘任和激励机制设计等方面提供了有益的启发.

In the existing literature on corporate governance,most of the research related to managerial characteristics has two main limitations.First,most of the papers focus on the relationship between one managerial individual characteristic and corporate performance but lack a comprehensive understanding of the potential non-linear relationship and interactions among some of the important independent variables.Second,existing research tests casual inference but ignores the predictive performance of the model.In this paper,we first examine if managerial individual characteristics can predict corporate performance by using a machine learning approach:Boosting regression trees.Using a sample of listed firms in the Chinese A-share market from 2008 to 2016,we study whether these individual characteristics could predict corporate performance.The evidence shows that:1)The individual characteristics of Chinese executives including CEOs and chairmen could predict corporate performance only to a limited degree.2)Among multiple individual characteristics,managerial ownership and executive age are the two most important predictors of corporate performance.3)The relations between predictors and corporate performance are non-linear,consistent with the prior literature.This paper initiates a new,more thorough perspective in Chinese executive research using machine learning methods and has important implications for selecting executives and designing incentive mechanisms.