中国上市公司分红的动因研究——基于机器学习的证据  

Motivation for Dividends of Listed Companies in China:Evidence Based on Machine Learning

作　　者：陈运森[1] 周金泳 彭嘉续 CHEN Yun-sen;ZHOU Jin-yong;PENG Jia-xu(School of Accountancy,Central University of Finance and Economics)

机构地区：[1]中央财经大学会计学院

出　　处：《中国工业经济》2024年第5期155-173,共19页China Industrial Economics

基　　金：国家自然科学基金面上项目“随机监管的资本市场经济后果及监管联动性研究:基于证监会双随机检查的准自然实验”(批准号72272168);国家自然科学基金青年项目“上市公司研发投入的数据缺失:基于披露机制建模和机器学习的方法与应用”(批准号72201288);国家自然科学基金面上项目“企业数字化转型的赋能机制与赋能效果研究:基于人力资本结构调整视角”(批准号72272167)。

摘　　要：企业分红不仅是股东投资回报的直接体现,更是“建设以投资者为本的资本市场”的重要手段,但“股利之谜”一直是理论与实践中的基础性难题。本文采用机器学习方法,评估了不同类别公司特征对分红行为的预测效果,旨在更准确地解读中国企业的分红动因。结果发现:(1)以渐进梯度回归树和随机森林为代表的集成学习方法在预测公司分红行为方面表现突出,显著优于多元线性回归等传统统计模型。(2)生命周期特征和公司税率特征对中国资本市场分红实践的预测效果较好,第一类和第二类代理问题相关特征次之,而融资需求特征和投资者情绪特征在企业分红行为的预测效果上相对较弱。(3)在各类分红动因变量中,留存收益资产比、公司税率特征、其他应收款资产比和上一期股利支付水平对分红水平的预测贡献较大。异质性分析显示,影响企业分红的主要因素在半强制分红政策、股息税改革和《现金分红指引》等股利政策颁布前后存在差异,在不同产权性质以及自由现金流状况的企业之间也有所不同,是否采取股利迎合策略和投资者现金股利情绪也会影响公司的现金分红行为。本文为“加强现金分红监管、增强投资者回报”的监管实践提供了理论支持,也为深入理解中国现代资本市场的分红实践和活跃资本市场提供了启示。Corporate dividends are not only a direct reflection of shareholders’return on investment but also serve to bolster investor confidence.It is an important means to“build an investororiented capital market”.However,the“dividend puzzle”remains a fundamental challenge in both theory and practice,as existing theories fail to fully explain the true motivations behind dividend payouts.This paper employs a variety of machine learning methods to analyze the mechanisms of corporate dividends under the influence of multiple factors,and evaluates the predictive effect of different types of company characteristics on dividend behavior,aiming to more accurately interpret the dividend behavior of China’s enterprises.The results are as follows.Firstly,the ensemble learning methods represented by the gradient boosting regression tree and the random forest are outstanding in predicting and explaining corporate dividend behavior,which significantly exceed conventional statistical models such as multiple linear regression.Secondly,among the characteristics of multiple companies,the life cycle characteristics and corporate tax rate characteristics have a better prediction effect on the dividend practice of China’s capital market,followed by the first and second types of agency problems,while the financing demand and investor sentiment characteristics have a weaker prediction effect on the dividend behavior of enterprises.Thirdly,among the characteristic variables of various dividend drivers,the ratio of retained earnings to assets,the characteristics of corporate tax rate,the ratio of other receivables to assets,and the level of dividend payment in the previous period contribute the most to the prediction of dividend level.Heterogeneity analysis shows that the main factors affecting corporate dividends are significantly different before and after the promulgation of dividend policies such as semi-mandatory dividend policy,dividend tax reform,and the Cash Dividend Guidelines.There are significant differences between ent

关 键 词：股利政策 企业分红 机机器学习 股利之谜 

分 类 号：F275[经济管理—企业管理]